CN117952935A - Photovoltaic panel shadow heating spot identification method based on visible light image threshold segmentation - Google Patents
Photovoltaic panel shadow heating spot identification method based on visible light image threshold segmentation Download PDFInfo
- Publication number
- CN117952935A CN117952935A CN202410121171.3A CN202410121171A CN117952935A CN 117952935 A CN117952935 A CN 117952935A CN 202410121171 A CN202410121171 A CN 202410121171A CN 117952935 A CN117952935 A CN 117952935A
- Authority
- CN
- China
- Prior art keywords
- image
- hot spot
- photovoltaic panel
- visible light
- threshold
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 61
- 230000011218 segmentation Effects 0.000 title claims abstract description 44
- 238000010438 heat treatment Methods 0.000 title claims abstract description 18
- 238000001514 detection method Methods 0.000 claims abstract description 44
- 238000012549 training Methods 0.000 claims abstract description 20
- 230000008030 elimination Effects 0.000 claims abstract description 9
- 238000003379 elimination reaction Methods 0.000 claims abstract description 9
- 238000005286 illumination Methods 0.000 claims abstract description 7
- 238000004458 analytical method Methods 0.000 claims description 15
- 238000003706 image smoothing Methods 0.000 claims description 10
- 230000008859 change Effects 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 8
- 238000010845 search algorithm Methods 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 4
- 230000007797 corrosion Effects 0.000 claims description 4
- 238000005260 corrosion Methods 0.000 claims description 4
- 230000000877 morphologic effect Effects 0.000 claims description 4
- 238000011176 pooling Methods 0.000 claims description 4
- 238000013459 approach Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 description 11
- 238000010586 diagram Methods 0.000 description 4
- 238000002372 labelling Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 235000002566 Capsicum Nutrition 0.000 description 2
- 239000006002 Pepper Substances 0.000 description 2
- 241000722363 Piper Species 0.000 description 2
- 235000016761 Piper aduncum Nutrition 0.000 description 2
- 235000017804 Piper guineense Nutrition 0.000 description 2
- 235000008184 Piper nigrum Nutrition 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 150000003839 salts Chemical class 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000003703 image analysis method Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000011423 initialization method Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration using local operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/17—Terrestrial scenes taken from planes or by drones
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Remote Sensing (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
A photovoltaic panel shadow heating spot identification method based on visible light image threshold segmentation comprises the following steps: establishing a photovoltaic panel infrared image data set and a photovoltaic panel visible light image data set; removing local highlight and uneven illumination influence in the original image by adopting a background elimination method based on an image Gaussian scale space; then searching an optimal global segmentation threshold value by using a robust maximum entropy method of one-dimensional histogram reconstruction; constructing a hot spot detection model formed by shadow shielding, and training the hot spot detection model formed by shadow shielding by using a photovoltaic panel infrared image data set; using photovoltaic panel visible light image data; and obtaining a visible light image with shadow features of the photovoltaic panel to be identified and an infrared image at the same position, sending the infrared image into a hot spot detection model formed by shadow shielding, obtaining a hot spot position and drawing the hot spot position in the visible light image. Compared with the prior art, the method and the device have the advantages that the hot spot position formed by shading in the visible light image is effectively obtained, and the recognition accuracy and the recognition efficiency are improved.
Description
Technical Field
The invention relates to the technical field of solar photovoltaic panel hot spot recognition, in particular to a photovoltaic panel shadow hot spot recognition method based on visible light image threshold segmentation.
Background
Solar photovoltaic power generation is used as clean energy source for sustainable development, and has great effect on sustainable development of environment and economy. The hot spot is a main type of solar photovoltaic panel faults, and is caused by the fact that the photovoltaic panel is shielded by foreign matters, so that current under the photovoltaic panel is uneven, and further the local abnormal heating phenomenon of the photovoltaic panel, which is formed by the faults of electronic devices, is caused. Typically, infrared hot spots on photovoltaic panels can be found using infrared imagers. Unlike hot spots caused by direct contact of paper, leaves and the like with the surface of the photovoltaic panel, shadow formed on the surface of the photovoltaic panel by trunks, equipment posts and the like which are close to the photovoltaic panel but not in direct contact with the surface of the photovoltaic panel also can form hot spots under sunlight irradiation. The hot spots have the properties of non-permanence, non-statics and the like along with the change of the irradiation angle of sunlight, and have influence on the acquisition and identification of whether the hot spot faults exist in the photovoltaic panel truly by using an infrared imager.
The existing image analysis method for the hot spots of the photovoltaic panel generally adopts an infrared image shot by an infrared imager to analyze, an image processing technology is used for finding the positions of the hot spots in the infrared image, and a method for identifying the hot spots of the photovoltaic module by fusing infrared and visible light images is also adopted. However, in the prior art, whether the hot spot of the photovoltaic panel is caused by shadow cannot be accurately judged, so that the follow-up judgment of the fault type and the damage degree of the hot spot of the photovoltaic panel and the treatment measures to be taken can be influenced.
Disclosure of Invention
The invention aims to solve the technical problem of providing a photovoltaic panel shadow heating spot identification method based on visible light image threshold segmentation, which accurately identifies the photovoltaic panel heating spot and avoids false identification caused by shadow.
In order to solve the technical problems, the invention adopts the following technical scheme:
a photovoltaic panel shadow heating spot identification method based on visible light image threshold segmentation comprises the following steps: step1, establishing a photovoltaic panel infrared image data set and a photovoltaic panel visible light image data set;
step2, removing local highlight and uneven illumination influence in the original image by adopting a background elimination method based on an image Gaussian scale space; then searching an optimal global segmentation threshold value by using a robust maximum entropy method of one-dimensional histogram reconstruction, and extracting a visible light shadow feature image;
Step3, constructing a hot spot detection model, and training the hot spot detection model by using a photovoltaic panel infrared image dataset;
step4, constructing a visible light image data set training hot spot cause analysis model of the photovoltaic panel;
step5, obtaining a visible light image with shadow features of the photovoltaic panel to be identified and an infrared image at the same position, sending the infrared image into a hot spot detection model, obtaining a hot spot position and drawing the hot spot position in the visible light image, and identifying whether the hot spot formed by shadow shielding is formed or not.
The specific steps of Step2 include:
Step2.1, performing image smoothing by adopting a two-dimensional Gaussian filter kernel, performing background estimation by creating an image Gaussian scale space, reconstructing an image one-dimensional histogram by using the gray scale of each point of the image and the gray scale of a neighborhood median, completing global threshold segmentation of the target image by combining a maximum entropy method focusing on a small target, and finally removing a few isolated noise around hot spots by using two mathematical basic morphological operations of expansion and corrosion to obtain a final segmentation result;
Step2.2, image complex background elimination: checking image smoothing by using a two-dimensional Gaussian filter, and calculating a difference value between the image smoothing and an original image to eliminate complex background influence on the basis of background estimation by creating an image Gaussian scale space; the two-dimensional Gaussian function is widely applied to the fields of image denoising, feature extraction and the like;
let the original image be I (x, y), and the scale space L (x, y, σi) obtained by the convolution operation of the original image and the two-dimensional gaussian filter kernel of the scale change is:
L(x，y，σi)＝G(x,y,σi)×I(x，y)；
Thus, there are:
Wherein: sigma i is a scale factor, and a Gaussian scale space is constructed by adjusting the size of sigma i; the relationship between the scale factors is:
σi＝kiσ0，k＞1；i＝1，2，3，…；
Wherein sigma 0 is an initial scale factor, and is obtained through hot spot size information in the picture; a gaussian differential scale space D (x, y, σi) is obtained in the gaussian scale space, and D (x, y, σi) can be obtained by subtracting two adjacent image scale spaces L (x, y, σi), that is:
D(x，y，σi)＝Li+1(x,y,σi)-Li(x,y,σi)；
As can be seen from the gaussian filter characteristics, as the scale factor σ i increases, L i(x,y,σi) the larger the smooth scale is, the gradually converging to the background, so that the sum of the adjacent two-layer image difference D i(x,y,σi) pixels in the image scale space becomes smaller; let the mean value of the pixels in image D i(x,y,σi) be E I:
Wherein m×n is an image size; according to E i change, judging the background approach degree, stopping creating the scale space when E i is less than or equal to delta, and determining the last layer of image L i(x,y,σi) of the scale space as the optimal background, wherein delta is a given error threshold;
Step2.3, graying treatment: assuming that the original gray-scale image I (x, y) is composed of a bright background and a dark target, taking each layer of the scale space L (x, y, sigma i) as a background, and making a difference with the original gray-scale image I (x, y) to obtain a target image D i (x, y) with the background eliminated by the current layer, namely:
Di(x,y)＝|I(x,y)-Li(x,y,σi)|；
And (3) obtaining a final target image by adopting a weighted average mode for each layer of the target image, wherein the operation formula is as follows:
Wherein: n is the total number of Gaussian scale space layers; w i is the weight of each layer of the target image D i (x, y), and each expression is:
When the original grayscale image I (x, y) is composed of a dark background and a bright target:
when the original grayscale image I (x, y) is composed of a bright background and a dark target:
Step2.4, realizing threshold segmentation by a robust maximum entropy method; reconstructing an image one-dimensional histogram by using the gray level of each point of the image and the neighborhood space information, and realizing global threshold segmentation of the target image by combining a maximum entropy method.
The concrete process of the step2.4 is as follows:
The gray level number of the pixels of the original image I (x, y) is M, the gray value of each pixel element point is I, and the gray median value in the h multiplied by h neighborhood of each pixel element point is j; a two-dimensional histogram of the original image Guan Yuandian, which corresponds to the neighborhood median, can be constructed, a few noise and edge pixel points are removed, other pixel points are distributed along the diagonal of the rectangle of the two-dimensional histogram, and the effective searching range of the threshold value is in a rectangular area taking the diagonal as a central line; taking the diagonal OB as a new coordinate horizontal axis, taking a straight line perpendicular to the OB as a new coordinate vertical axis, and counting the number of points projected to the diagonal OB in a rectangular area to construct a new one-dimensional histogram;
let F (a, b) be any point in the rectangular area, the point of the point projected ON the diagonal OB is denoted as a, and the straight line l crosses a and F (a, b) and the i-axis to intersect at a point N, so as to obtain |on|=a+b;
in Rt Δoan, the projection value |oa| of the point F (a, b) on the diagonal OB is:
the OA is the pixel value of the point F (a, b) in the new one-dimensional histogram, and the distribution range is Representing a downward rounding; the process of finding the optimal threshold in the one-dimensional histogram by using the maximum entropy method is as follows:
Let p i be the probability of the gray value i (i is more than or equal to 0 and less than M) of the image element point, w o(t),wb (t) respectively represents the accumulated probability of the background and foreground pixels of t threshold segmentation, and the entropy values of the threshold t foreground class and the background class are respectively as follows:
Wherein: The selection criteria for the optimal global threshold t * may be expressed as:
the specific steps of Step5 are as follows:
step5.1, obtaining a visible light image with shadow characteristics of a photovoltaic panel to be identified and an infrared image at the same position;
step5.2, sending the infrared image of the photovoltaic panel to be identified into a trained hot spot detection model to output the central coordinate and length and width data of a prediction frame of the hot spots in the infrared image, namely obtaining the hot spot positions of the photovoltaic panel;
And step5.3, drawing a hot spot prediction frame in the visible light image, analyzing the cause of hot spots in the hot spot prediction frame in the visible light image by using a hot spot cause analysis model, acquiring the hot spot cause corresponding to the hot spot position, finally acquiring the hot spot position on the photovoltaic panel and the hot spot cause of the hot spots in each hot spot position, and identifying whether the hot spots are formed by shadow shielding.
The hot spot detection model is an improved Faster R-CNN detection model, a feature map module is arranged in front of ROIAlign parts of the improved Faster R-CNN detection model, and a search algorithm is arranged after a prediction frame is selected ROIAlign.
The feature map module performs gaussian modeling on the fed feature map, judges whether a set threshold is reached in the feature map, adjusts the super parameter along with the training wheel number, judges the output convolution layer by adopting an N-dimensional column vector in the N-dimensional convolution layer, and uses L T to represent threshold loss, and adjusts a proper threshold through iteration, wherein:
where L T is the threshold map penalty, R d is the box prediction of the threshold map, And outputting the position prediction of the layer for the feature.
In ROIAlign, a search algorithm is set after a prediction frame is selected, whether a region is continuous or not is analyzed, the prediction frame is divided into a plurality of parts, a plurality of parts of prediction frames are screened, a value with characteristics exceeding a preset value is found in the prediction frame to be a larger value, the maximum value X +sx、Y+sx and the minimum value X +i,、Y+i, of the position coordinates are selected in the larger characteristic value, and the box is readjusted to be a new box only comprising a plurality of values with the maximum characteristics.
In ROIAlign above, the range of the characteristic value in the prediction frame is 0-1, and the value defined as greater than or equal to 0.7 is defined as larger value.
The above-mentioned hot spot cause analysis model is ResNet model.
The Res Net model basic framework adopts a Res Net 101 structure, input data of the Res Net 101 structure is a photovoltaic panel image obtained through pretreatment, output data is codes corresponding to hot spot causes, after convolution pooling in the Res Net 101 structure, features are continuously extracted through a redundancy layer, and finally, the Res Net 101 structure passes through a full-connection layer with 1000 dimensions, and a softmax is used as an activation function to output a result.
The cross entropy loss function L between the ResNet model output result and the actual classification y is as follows:
Where N is the dimension of the output layer.
The photovoltaic panel shadow heating spot identification method based on visible light image threshold segmentation provided by the invention has the following beneficial effects:
1) According to the invention, a Gaussian scale space is constructed through convolution operation by utilizing the characteristic of a two-dimensional Gaussian function, an optimal background layer of the denoised image is judged according to gray information of the differential scale space in the space, then a shadow image with shadow shielding characteristics is extracted by utilizing an improved background difference method, and the difference value is adaptively adjusted to be positive and negative according to the brightness and darkness relationship between a target and a background, so that the interference of uneven illumination and local highlight on the subsequent global threshold segmentation is eliminated; the probability of the gray value of the image pixel point is used for respectively solving the accumulated probability and entropy value of the target background and the foreground pixel, and the global segmentation threshold is determined according to the principle of the maximum entropy value, so that adverse effects caused by low contrast of a shadow area in a target image are solved, and meanwhile, the robustness of a processing process is enhanced.
2) According to the invention, based on the hot spot detection model and the hot spot cause analysis model, the infrared image and the visible light image of the photovoltaic panel with the foreign object shadow can be respectively detected, the hot spot position in the infrared image is obtained, the same position of the visible light image is analyzed based on the hot spot position to analyze the hot spot cause, the hot spot position is combined with the cause obtained by analysis, the problem that whether the hot spot is caused by the shadow can not be rapidly distinguished only by using the infrared image in the actual environment is solved, the recognition detection effect is improved, and the recognition efficiency is improved.
3) The hot spot detection model is based on an improved Mask R-CNN model, and the recognition accuracy and recognition efficiency of the Mask R-CNN model on the hot spot position in an infrared image are further improved through the feature map module, the hot spot contour is accurately positioned, and the analysis accuracy of judging whether the hot spot is caused by shadow or not is improved.
4) The invention adopts a double-data set and a double-model mechanism, can ensure that the model output feature vector has better training performance, overcomes the problem of insufficient hot spot target identification capability in the traditional detection, and ensures that the target position of hot spot detection is more accurate.
Drawings
The invention is further illustrated by the following examples in conjunction with the accompanying drawings:
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a schematic diagram of the structure of an improved Mask R-CNN model according to the present invention;
FIG. 3 is a schematic flow chart of the hot spot detection model and the hot spot cause analysis model in the invention respectively being Mask R-CNN model and Res Net model;
Fig. 4 is a pixel point coordinate diagram of a two-dimensional histogram of the gray value of the original image Guan Yuandian and the neighborhood median.
Detailed Description
The technical scheme of the invention is described in detail below with reference to the accompanying drawings and examples.
As shown in fig. 1, a method for identifying a photovoltaic panel shadow heating spot based on threshold segmentation of a visible light image comprises the following steps:
step1, establishing a photovoltaic panel infrared image data set and a photovoltaic panel visible light image data set;
step2, removing local highlight and uneven illumination influence in the original image by adopting a background elimination method based on an image Gaussian scale space; then searching an optimal global segmentation threshold value by using a robust maximum entropy method of one-dimensional histogram reconstruction, and extracting a visible light shadow feature image;
Step3, constructing a hot spot detection model, and training the hot spot detection model by using a photovoltaic panel infrared image dataset;
step4, constructing a visible light image data set training hot spot cause analysis model of the photovoltaic panel;
step5, obtaining a visible light image with shadow features of the photovoltaic panel to be identified and an infrared image at the same position, sending the infrared image into a hot spot detection model, obtaining a hot spot position and drawing the hot spot position in the visible light image, and identifying whether the hot spot formed by shadow shielding is formed or not.
The specific steps of Step2 include:
Step2.1, performing image smoothing by adopting a two-dimensional Gaussian filter kernel, performing background estimation by creating an image Gaussian scale space, reconstructing an image one-dimensional histogram by using the gray scale of each point of the image and the gray scale of a neighborhood median, completing global threshold segmentation of the target image by combining a maximum entropy method focusing on a small target, and finally removing a few isolated noise around hot spots by using two mathematical basic morphological operations of expansion and corrosion to obtain a final segmentation result;
Step2.2, image complex background elimination: checking image smoothing by using a two-dimensional Gaussian filter, and calculating a difference value between the image smoothing and an original image to eliminate complex background influence on the basis of background estimation by creating an image Gaussian scale space; the two-dimensional Gaussian function is widely applied to the fields of image denoising, feature extraction and the like;
let the original image be I (x, y), and the scale space L (x, y, σi) obtained by the convolution operation of the original image and the two-dimensional gaussian filter kernel of the scale change is:
L(x，y，σi)＝G(x,y,σi)×I(x，y)；
Thus, there are:
Wherein: sigma i is a scale factor, and a Gaussian scale space is constructed by adjusting the size of sigma i; the current value of the scale factor sigma i mainly comprises an equidistant method and a proportional method, and the proportional method which can better show the continuity between scales is used in the invention; the relationship between the scale factors is:
σi＝kiσ0，k＞1；i＝1，2，3，…；
Wherein sigma 0 is an initial scale factor, and is obtained through hot spot size information in the picture; a gaussian differential scale space D (x, y, σi) is obtained in the gaussian scale space, and D (x, y, σi) can be obtained by subtracting two adjacent image scale spaces L (x, y, σi), that is:
D(x，y，σi)＝Li+1(x,y,σi)-Li(x,y,σi)；
As can be seen from the gaussian filter characteristics, as the scale factor σ i increases, L i(x,y,σi) the larger the smooth scale is, the gradually converging to the background, so that the sum of the adjacent two-layer image difference D i(x,y,σi) pixels in the image scale space becomes smaller; let the mean value of the pixels in image D i(x,y,σi) be E I:
Wherein m×n is an image size; according to E i change, judging the background approach degree, stopping creating the scale space when E i is less than or equal to delta, and determining the last layer of image L i(x,y,σi) of the scale space as the optimal background, wherein delta is a given error threshold;
Step2.3, graying treatment: assuming that the original gray-scale image I (x, y) is composed of a bright background and a dark target, taking each layer of the scale space L (x, y, sigma i) as a background, and making a difference with the original gray-scale image I (x, y) to obtain a target image D i (x, y) with the background eliminated by the current layer, namely:
Di(x,y)＝|I(x,y)-Li(x,y,σi)|；
And (3) obtaining a final target image by adopting a weighted average mode for each layer of the target image, wherein the operation formula is as follows:
Wherein: n is the total number of Gaussian scale space layers; w i is the weight of each layer of the target image D i (x, y), and each expression is:
The method solves the problem that the influence of uneven illumination in the image to be processed is not suitable for the characteristics of uneven local reflection of light, abrupt change of image background gray scale and the like of the image. In order to solve the characteristic problems of the image to be processed, and can be adjusted automatically according to the actual situation, the background elimination method is improved as follows:
When the original grayscale image I (x, y) is composed of a dark background and a bright target:
when the original grayscale image I (x, y) is composed of a bright background and a dark target:
The improved background elimination method takes an estimated background as a boundary, gray abrupt noise in the image background is automatically shielded, and a target foreground part is extracted; meanwhile, the robustness of background estimation in the complex background is enhanced, and when the proportion of the local foreground and the background of the image is not great, the situation of extracting the background by mistake is effectively prevented;
Step2.4, realizing threshold segmentation by a robust maximum entropy method; the contrast between the shadow image of the target image D (x, y) after the background is eliminated and the background is lower, and the segmentation of the shadow image is still interfered by the noise of the surrounding similar gray level; and reconstructing a one-dimensional histogram of the image by considering the gray level of each point of the image and the neighborhood space information, and realizing global threshold segmentation of the target image by combining a maximum entropy method.
The concrete process of the step2.4 is as follows:
The gray level number of the pixels of the original image I (x, y) is M, the gray value of each pixel element point is I, and the gray median value in the h multiplied by h neighborhood of each pixel element point is j; a two-dimensional histogram of the original image Guan Yuandian, in which the gray value corresponds to the neighborhood median, can be constructed, and the coordinates of the pixel points are shown in fig. 4; removing few noise and edge pixel points, wherein other pixel points are distributed along the diagonal line of the rectangular of the two-dimensional histogram, and the effective searching range of the threshold value is in a rectangular area taking the diagonal line as a central line; taking the diagonal OB as a new coordinate horizontal axis, taking a straight line perpendicular to the OB as a new coordinate vertical axis, and counting the number of points projected to the diagonal OB in a rectangular area to construct a new one-dimensional histogram;
Let F (a, B) be any point in the rectangular area, the point projected ON the diagonal line 0B is denoted as a, the straight line l crosses a and F (a, B) and the i-axis to intersect at a point N, and |on|=a+b can be obtained;
in Rt Δoan, the projection value |oa| of the point F (a, b) on the diagonal OB is:
the OA is the pixel value of the point F (a, b) in the new one-dimensional histogram, and the distribution range is Representing a downward rounding; the process of finding the optimal threshold in the one-dimensional histogram by using the maximum entropy method is as follows:
Let p i be the probability of the gray value i (i is more than or equal to 0 and less than M) of the image element point, w o(t),wb (t) respectively represents the accumulated probability of the background and foreground pixels of t threshold segmentation, and the entropy values of the threshold t foreground class and the background class are respectively as follows:
Wherein: The selection criteria for the optimal global threshold t * may be expressed as:
The global threshold segmentation method for the image by using the maximum entropy method can segment the image area more completely, and the processing process is robust.
The specific steps of Step5 are as follows:
step5.1, obtaining a visible light image with shadow characteristics of a photovoltaic panel to be identified and an infrared image at the same position;
step5.2, sending the infrared image of the photovoltaic panel to be identified into a trained hot spot detection model to output the central coordinate and length and width data of a prediction frame of the hot spots in the infrared image, namely obtaining the hot spot positions of the photovoltaic panel;
And step5.3, drawing a hot spot prediction frame in the visible light image, analyzing the cause of hot spots in the hot spot prediction frame in the visible light image by using a hot spot cause analysis model, acquiring the hot spot cause corresponding to the hot spot position, finally acquiring the hot spot position on the photovoltaic panel and the hot spot cause of the hot spots in each hot spot position, and identifying whether the hot spots are formed by shadow shielding.
The hot spot detection model is an improved Faster R-CNN detection model, a feature map module is arranged in front of ROIAlign parts of the improved Faster R-CNN detection model, and a search algorithm is arranged after a prediction frame is selected ROIAlign.
The feature map module carries out Gaussian modeling on the fed feature map, judges whether a set threshold value is reached in the feature map, wherein the threshold value is that the super parameter is adjusted along with the number of training wheels, and for the visible photovoltaic image, as the boundary of the photovoltaic group string is usually white and is obviously distinguished from pixels in a background area, the selection of the image segmentation threshold value can be carried out through gray level; the gray original image noise type is mainly composed of salt and pepper noise, and denoising can be realized by selecting a median filtering method; smoothing the image after removing salt and pepper noise, and eliminating the influence of uneven illumination and partial gray abrupt change; searching the difference between the background and the original image by estimating the background, and eliminating the influence of the complex background; and judging the output convolution layer by adopting an N-dimensional column vector in the N-dimensional convolution layer, wherein L T is used for representing the threshold loss, and an appropriate threshold is adjusted through iteration, wherein:
where L T is the threshold map penalty, R d is the box prediction of the threshold map, And outputting the position prediction of the layer for the feature.
In ROIAlign, a search algorithm is set after a prediction frame is selected, whether a region is continuous or not is analyzed, the prediction frame is divided into a plurality of parts, a plurality of parts of prediction frames are screened, a value with characteristics exceeding a preset value is found in the prediction frame to be a larger value, the maximum value X max、Ymax and the minimum value X min、Ymin of the position coordinates are selected in the larger characteristic value, and the box is readjusted to be a new box only comprising a plurality of values with the maximum characteristics.
In ROIAlign above, the range of the characteristic value in the prediction frame is 0-1, and the value defined as greater than or equal to 0.7 is defined as larger value.
The above-mentioned hot spot cause analysis model is ResNet model.
The Res Net model basic framework adopts a Res Net 101 structure, input data of the Res Net 101 structure is a photovoltaic panel image obtained through pretreatment, output data is codes corresponding to hot spot causes, after convolution pooling in the Res Net 101 structure, features are continuously extracted through a redundancy layer, and finally, the Res Net 101 structure passes through a full-connection layer with 1000 dimensions, and a softmax is used as an activation function to output a result.
The cross entropy loss function L between the ResNet model output result and the actual classification y is as follows:
Where N is the dimension of the output layer.
Preferably, step1 further includes labeling four vertex coordinates of each hot spot in each image in the photovoltaic panel infrared image data set, obtaining coordinates, frames and detection information type data of the hot spots, and constructing a hot spot position label corresponding to the image in the photovoltaic panel infrared image database.
Preferably, step1 further includes marking a hot spot cause of each image in the visible light image dataset of the photovoltaic panel, constructing a hot spot cause label corresponding to the picture in the visible light image dataset, and re-marking a hot spot formed by shading.
Preferably, the unmanned aerial vehicle provided with the double cameras is used for collecting the infrared image and the visible light image of the same position of the photovoltaic panel to be identified, and the double cameras are respectively an infrared thermal imager and a visible light camera.
Examples:
A photovoltaic panel shadow heating spot identification method based on visible light image threshold segmentation comprises the following steps:
s1: and establishing a photovoltaic panel infrared image data set and a photovoltaic panel visible light image data set.
In this embodiment, the unmanned aerial vehicle equipped with the double cameras collects the infrared image and the visible light image of the same position of the photovoltaic panel to be identified, so as to obtain an infrared image data set of the photovoltaic panel and an image in the visible light image data set of the photovoltaic panel, and one infrared image in the infrared image data set of the photovoltaic panel corresponds to one visible light image of the photovoltaic panel in the visible light image data set of the photovoltaic panel.
S2: and carrying out image smoothing by adopting a two-dimensional Gaussian filter kernel, and carrying out background estimation by creating an image Gaussian scale space. Reconstructing a one-dimensional histogram of the image by using the gray scale of each point of the image and the gray scale of the neighborhood median, and completing global threshold segmentation of the target image by combining a maximum entropy method focusing on a small target; finally, using two mathematical basic morphological operations of expansion and corrosion to remove surrounding few isolated noise, and finally using a robust maximum entropy method of one-dimensional histogram reconstruction to find an optimal global segmentation threshold value and extracting visible light shadow feature images; labeling four vertex coordinates of each hot spot in each image in the photovoltaic panel infrared image data set, obtaining coordinates, frames and detection information type data of the hot spots, constructing a hot spot position label corresponding to a picture in a photovoltaic panel infrared image database, labeling a hot spot cause of each image in the photovoltaic panel visible light image data set, and constructing a hot spot cause label corresponding to the picture in the visible light image data set.
In this embodiment, labeling labels of the visible light image dataset of the photovoltaic panel are labeled and classified by using different shielding objects such as shadows, dust, leaves and the like as labels.
In this embodiment, the photovoltaic panel infrared images in the constructed photovoltaic panel infrared image data set are divided into a training set and a testing set according to a preset ratio, and in this embodiment, a ratio of 4:1 is adopted.
S3: and constructing a hot spot detection model, and training the hot spot detection model by using the photovoltaic panel infrared image data set.
As shown in FIG. 2, the hot spot detection model is an improved Faster R-CNN detection model, a feature map module is arranged in front of ROIAlign parts of the improved Faster R-CNN detection model, and a search algorithm is arranged in ROIAlign after a prediction frame is selected.
In the actual detection, the detection target is the photovoltaic panel hot spots in the infrared image, and the photovoltaic panel hot spots have the characteristics of uneven distribution and different sizes, but have pixel consistency, namely the pixel values of the hot spot area images are converged, and the pixel values are higher; by utilizing the characteristic, ROIAlign in the Mask R-CNN model is adjusted in a targeted manner.
Specifically, as shown in fig. 2, before ROIAlign, performing hot spot pre-estimation on the feature map by using a feature map module, performing gaussian modeling on the fed feature map by using the feature map module, judging whether a set threshold is reached in the feature map, wherein the threshold is adjusted by using a super parameter along with the number of training rounds, judging the output convolution layer by using an N-dimensional column vector in the N-dimensional convolution layer, wherein L T is used for representing threshold loss, and performing iterative adjustment to obtain a proper threshold, wherein:
Where L T is the threshold map penalty, R q is the box prediction of the threshold map, And outputting the position prediction of the layer for the feature.
Continuing to ROIAlign the feature diagram meeting the conditions; secondly, in ROIAlign, a search algorithm is set after a prediction frame is selected, whether the areas are continuous or not is analyzed, and the prediction frame is required to be divided into a plurality of small prediction frames because a plurality of hot spot characteristic areas possibly exist in the prediction frame; and screening a plurality of smaller prediction frames, finding a characteristic larger value in the prediction frames, selecting a position coordinate maximum value X max、Ymax and a position coordinate minimum value X min、Ymin from the larger characteristic values, readjusting the box to be a new box only comprising a plurality of values with the largest characteristics, and carrying out subsequent flow.
In the embodiment, a common image recognition model ResNet classifies causes of the hot spot position phenomenon of the common photovoltaic panel image; the ResNet model can alleviate the degradation problem caused by the depth of the network, and a direct connection channel is added in the network. In order to improve the recognition accuracy, a 101-layer model design is arranged; the basic framework of the hot spot cause recognition model adopts a ResNet 101 structure, and input data is a photovoltaic panel common image obtained through pretreatment; the output data is codes corresponding to the hot spot causes; after the input data is subjected to convolution pooling, the features are continuously extracted through the redundancy layer, and finally, the input data passes through a full-connection layer with 1000 dimensions, and a result is output by taking softmax as an activation function.
The cross entropy loss function between the output result and the actual classification using the deep learning model is as follows:
In the embodiment, a batch gradient descent method is adopted for the loss function, and the weight parameters of the deep learning model are optimized so that the classification error rate is minimum; each super parameter configuration in the training process: the weight initialization adopts a normal random initialization method, and the batch size is selected to be the most proper according to the performance and the memory capacity of the GPU used for training; the learning rate adopts a dynamic learning rate; after training is completed, error assessment is performed using an independent test set loss function.
After the network structure is set and the training set is completed, respectively pre-training the double models by using a data set such as COCO2017, and learning network parameters in multiple iterations by using the models, so that the models have good detection performance; and then performing transfer learning training by using the pre-trained and learned infrared hot spot detection model Mask R-CNN and the common image recognition model Res Net.
After preprocessing an image, inputting a feature map into a Mask R-CNN, sampling on an output feature pyramid and an intermediate layer feature pyramid to obtain a predicted target area, firstly, judging the output convolution layer by adopting an N-dimensional column vector in an N-dimensional convolution layer through a ROIAlign in a predictor, representing threshold loss, and iteratively adjusting to obtain a proper threshold; continuing to ROIAlign the feature diagram meeting the conditions; and finally, outputting the central coordinates and the length and width data of the prediction frame and the type of the prediction frame by the network.
S4: and acquiring a visible light image with shadow features of the photovoltaic panel to be identified and an infrared image at the same position, sending the infrared image into a hot spot detection model, acquiring hot spot positions, drawing the hot spot positions in the visible light image, analyzing images in the hot spot positions of the visible light image by using a hot spot cause analysis model, and acquiring hot spot causes corresponding to the hot spot positions.
The step S4 specifically includes:
s41: the infrared image and the visible light image of the same position of the photovoltaic panel to be identified are acquired, and in the embodiment, the infrared image and the visible light image of the same position of the photovoltaic panel to be identified are acquired by adopting an unmanned aerial vehicle with double cameras.
S42: sending the infrared image of the photovoltaic panel to be identified into a trained hot spot detection model, and outputting the central coordinate and length and width data of a hot spot prediction frame in the infrared image, namely obtaining the hot spot position of the photovoltaic panel;
S43: and drawing a hot spot prediction frame in the visible light image, analyzing the cause of the hot spot in the hot spot prediction frame in the visible light image by utilizing a hot spot cause analysis model, acquiring the hot spot cause corresponding to the hot spot position, and finally acquiring the hot spot position on the photovoltaic panel and the hot spot cause of the hot spot in each hot spot position.
S5: and acquiring a visible light image with shadow features of the photovoltaic panel to be identified and an infrared image at the same position, sending the infrared image into a hot spot detection model, acquiring a hot spot position and drawing the hot spot position in the visible light image, thereby identifying whether the hot spot formed by shadow shielding is generated.
The above embodiments are merely examples and are not intended to limit the scope of the present invention; these embodiments may be implemented in various other ways, and various omissions, substitutions, and changes may be made without departing from the scope of the technical idea of the present invention.
Claims (10)
1. A photovoltaic panel shadow heating spot identification method based on visible light image threshold segmentation is characterized by comprising the following steps:
step1, establishing a photovoltaic panel infrared image data set and a photovoltaic panel visible light image data set;
step2, removing local highlight and uneven illumination influence in the original image by adopting a background elimination method based on an image Gaussian scale space; then searching an optimal global segmentation threshold value by using a robust maximum entropy method of one-dimensional histogram reconstruction, and extracting a visible light shadow feature image;
Step3, constructing a hot spot detection model, and training the hot spot detection model by using a photovoltaic panel infrared image dataset;
step4, constructing a visible light image data set training hot spot cause analysis model of the photovoltaic panel;
step5, obtaining a visible light image with shadow features of the photovoltaic panel to be identified and an infrared image at the same position, sending the infrared image into a hot spot detection model, obtaining a hot spot position and drawing the hot spot position in the visible light image, and identifying whether the hot spot formed by shadow shielding is formed or not.
2. The method for identifying the shadow heating spot of the photovoltaic panel based on the threshold segmentation of the visible light image according to claim 1, wherein the specific Step of Step2 comprises the following steps:
Step2.1, performing image smoothing by adopting a two-dimensional Gaussian filter kernel, performing background estimation by creating an image Gaussian scale space, reconstructing an image one-dimensional histogram by using the gray scale of each point of the image and the gray scale of a neighborhood median, completing global threshold segmentation of the target image by combining a maximum entropy method focusing on a small target, and finally removing a few isolated noise around hot spots by using two mathematical basic morphological operations of expansion and corrosion to obtain a final segmentation result;
Step2.2, image complex background elimination: checking image smoothing by using a two-dimensional Gaussian filter, and calculating a difference value between the image smoothing and an original image to eliminate complex background influence on the basis of background estimation by creating an image Gaussian scale space; the two-dimensional Gaussian function is widely applied to the fields of image denoising, feature extraction and the like;
let the original image be I (x, y), and the scale space L (x, y, σi) obtained by the convolution operation of the original image and the two-dimensional gaussian filter kernel of the scale change is:
L(x，y，σi)＝G(x,y,σi)×I(x，y)；
Thus, there are:
Wherein: sigma i is a scale factor, and a Gaussian scale space is constructed by adjusting the size of sigma i; the relationship between the scale factors is:
σi＝kiσ0，k＞1；i＝1，2，3，…；
Wherein sigma 0 is an initial scale factor, and is obtained through hot spot size information in the picture; a gaussian differential scale space D (x, y, σi) is obtained in the gaussian scale space, and D (x, y, σi) can be obtained by subtracting two adjacent image scale spaces L (x, y, σi), that is:
D(x，y，σi)＝Li+1(x,y,σi)-Li(x,y,σi)；
As can be seen from the gaussian filter characteristics, as the scale factor σ i increases, L i(x,y,σi) the larger the smooth scale is, the gradually converging to the background, so that the sum of the adjacent two-layer image difference D i(x,y,σi) pixels in the image scale space becomes smaller; let the mean value of the pixels in image D i(x,y,σi) be E I:
Wherein m×n is an image size; according to E i change, judging the background approach degree, stopping creating the scale space when E i is less than or equal to delta, and determining the last layer of image L i(x,y,σi) of the scale space as the optimal background, wherein delta is a given error threshold;
Step2.3, graying treatment: assuming that the original gray-scale image I (x, y) is composed of a bright background and a dark target, taking each layer of the scale space L (x, y, sigma i) as a background, and making a difference with the original gray-scale image I (x, y) to obtain a target image D i (x, y) with the background eliminated by the current layer, namely:
Di(x,y)＝|I(x,y)-Li(x,y,σi)|；
And (3) obtaining a final target image by adopting a weighted average mode for each layer of the target image, wherein the operation formula is as follows:
Wherein: n is the total number of Gaussian scale space layers; w i is the weight of each layer of the target image D i (x, y), and each expression is:
When the original grayscale image I (x, y) is composed of a dark background and a bright target:
when the original grayscale image I (x, y) is composed of a bright background and a dark target:
Step2.4, realizing threshold segmentation by a robust maximum entropy method; reconstructing an image one-dimensional histogram by using the gray level of each point of the image and the neighborhood space information, and realizing global threshold segmentation of the target image by combining a maximum entropy method.
3. The method for identifying the shadow heating spot of the photovoltaic panel based on the threshold segmentation of the visible light image according to claim 2, wherein the step2.4 specifically comprises the following steps:
The gray level number of the pixels of the original image I (x, y) is M, the gray value of each pixel element point is I, and the gray median value in the h multiplied by h neighborhood of each pixel element point is j; a two-dimensional histogram of the original image Guan Yuandian, which corresponds to the neighborhood median, can be constructed, a few noise and edge pixel points are removed, other pixel points are distributed along the diagonal of the rectangle of the two-dimensional histogram, and the effective searching range of the threshold value is in a rectangular area taking the diagonal as a central line; taking the diagonal OB as a new coordinate horizontal axis, taking a straight line perpendicular to the OB as a new coordinate vertical axis, and counting the number of points projected to the diagonal OB in a rectangular area to construct a new one-dimensional histogram;
let F (a, b) be any point in the rectangular area, the point of the point projected ON the diagonal OB is denoted as a, and the straight line l crosses a and F (a, b) and the i-axis to intersect at a point N, so as to obtain |on|=a+b;
in Rt Δoan, the projection value |oa| of the point F (a, b) on the diagonal OB is:
the OA is the pixel value of the point F (a, b) in the new one-dimensional histogram, and the distribution range is Representing a downward rounding; the process of finding the optimal threshold in the one-dimensional histogram by using the maximum entropy method is as follows:
Let p i be the probability of the gray value i (i is more than or equal to 0 and less than M) of the image element point, w o(t),wb (t) respectively represents the accumulated probability of the background and foreground pixels of t threshold segmentation, and the entropy values of the threshold t foreground class and the background class are respectively as follows:
Wherein: The selection criteria for the optimal global threshold t * may be expressed as:
4. the method for identifying the shadow heating spot of the photovoltaic panel based on the threshold segmentation of the visible light image according to claim 3, wherein the specific steps of Step5 are as follows:
step5.1, obtaining a visible light image with shadow characteristics of a photovoltaic panel to be identified and an infrared image at the same position;
step5.2, sending the infrared image of the photovoltaic panel to be identified into a trained hot spot detection model to output the central coordinate and length and width data of a prediction frame of the hot spots in the infrared image, namely obtaining the hot spot positions of the photovoltaic panel;
And step5.3, drawing a hot spot prediction frame in the visible light image, analyzing the cause of hot spots in the hot spot prediction frame in the visible light image by using a hot spot cause analysis model, acquiring the hot spot cause corresponding to the hot spot position, finally acquiring the hot spot position on the photovoltaic panel and the hot spot cause of the hot spots in each hot spot position, and identifying whether the hot spots are formed by shadow shielding.
5. The method for identifying the hot spots of the photovoltaic panel shadow based on the threshold segmentation of the visible light image according to claim 4, wherein the hot spot detection model is an improved Faster R-CNN detection model, a feature map module is arranged in front of ROIAlign parts of the improved Faster R-CNN detection model, and a search algorithm is arranged after a prediction frame is selected in ROIAlign.
6. The method for identifying photovoltaic panel shadow heating spot based on visible light image threshold segmentation according to claim 5, wherein the feature map module performs gaussian modeling on the fed feature map, determines whether a set threshold is reached in the feature map, the threshold is adjusted according to the number of training rounds, determines the output convolution layer by using an N-dimensional column vector in the N-dimensional convolution layer, and uses L l to represent a threshold loss, and iteratively adjusts a suitable threshold, wherein:
Where L l is the threshold map penalty, R q is the box prediction of the threshold map, And outputting the position prediction of the layer for the feature.
7. The method for identifying photovoltaic panel shadow heating spot based on visible light image threshold segmentation according to claim 6, wherein in ROIAlign, a search algorithm is set after a prediction frame is selected, whether a region is continuous or not is analyzed, the prediction frame is divided into a plurality of parts, a plurality of parts of prediction frames are screened, a value with characteristics exceeding a preset value is found in the prediction frame to be a larger value, a maximum value X +sx、Y+sx and a minimum value X +i,、Y+i, of position coordinates are selected in the larger characteristic value, and a box is readjusted to be a new box only comprising a plurality of values with the largest characteristics.
8. The method for identifying photovoltaic panel shadow heating spots based on visible light image threshold segmentation according to claim 7, wherein in ROIAlign, the range of characteristic values in a prediction frame is 0-1, and a value more than or equal to 0.7 is defined as a larger value.
9. The method for identifying the hot spots of the photovoltaic panel shadow based on the threshold segmentation of the visible light image according to claim 8, wherein the hot spot cause analysis model is ResNet models, a base frame of the Res Net model adopts a Res Net 101 structure, input data of the Res Net 101 structure is a photovoltaic panel image obtained through pretreatment, output data is codes corresponding to the hot spot cause, after convolution pooling in the Res Net 101 structure, the characteristic is continuously extracted through a redundancy layer, and finally, the result is output through a full-connection layer with 1000 dimensions by taking softmax as an activation function.
10. The method for identifying photovoltaic panel shadow heating spots based on visible light image threshold segmentation according to claim 9, wherein the cross entropy loss function L between the Res Net model output result and the actual classification y is:
Where N is the dimension of the output layer.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2023118224538 | 2023-12-27 | ||
| CN202311822453 | 2023-12-27 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117952935A true CN117952935A (en) | 2024-04-30 |
Family
ID=90803278
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410121171.3A Pending CN117952935A (en) | 2023-12-27 | 2024-01-29 | Photovoltaic panel shadow heating spot identification method based on visible light image threshold segmentation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117952935A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118864391A (en) * | 2024-07-04 | 2024-10-29 | 无锡云程电力科技有限公司 | Photovoltaic module hot spot detection method and system based on infrared thermal imager |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104123734A (en) * | 2014-07-22 | 2014-10-29 | 西北工业大学 | Visible light and infrared detection result integration based moving target detection method |
| CN105261037A (en) * | 2015-10-08 | 2016-01-20 | 重庆理工大学 | Moving object detection method capable of automatically adapting to complex scenes |
| AU2020103901A4 (en) * | 2020-12-04 | 2021-02-11 | Chongqing Normal University | Image Semantic Segmentation Method Based on Deep Full Convolutional Network and Conditional Random Field |
| CN113076816A (en) * | 2021-03-17 | 2021-07-06 | 上海电力大学 | Solar photovoltaic module hot spot identification method based on infrared and visible light images |
| CN113989260A (en) * | 2021-11-09 | 2022-01-28 | 浙江浙能嘉华发电有限公司 | An edge segmentation method of photovoltaic panels based on UAV infrared images |
| CN116503354A (en) * | 2023-04-27 | 2023-07-28 | 清华大学 | Photovoltaic battery hot spot detection and evaluation method and device based on multi-modal fusion |
-
2024
- 2024-01-29 CN CN202410121171.3A patent/CN117952935A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104123734A (en) * | 2014-07-22 | 2014-10-29 | 西北工业大学 | Visible light and infrared detection result integration based moving target detection method |
| CN105261037A (en) * | 2015-10-08 | 2016-01-20 | 重庆理工大学 | Moving object detection method capable of automatically adapting to complex scenes |
| AU2020103901A4 (en) * | 2020-12-04 | 2021-02-11 | Chongqing Normal University | Image Semantic Segmentation Method Based on Deep Full Convolutional Network and Conditional Random Field |
| CN113076816A (en) * | 2021-03-17 | 2021-07-06 | 上海电力大学 | Solar photovoltaic module hot spot identification method based on infrared and visible light images |
| CN113989260A (en) * | 2021-11-09 | 2022-01-28 | 浙江浙能嘉华发电有限公司 | An edge segmentation method of photovoltaic panels based on UAV infrared images |
| CN116503354A (en) * | 2023-04-27 | 2023-07-28 | 清华大学 | Photovoltaic battery hot spot detection and evaluation method and device based on multi-modal fusion |
Non-Patent Citations (4)
| Title |
|---|
| 余松煜等: "数字图像处理", 28 February 2007, 上海：上海交通大学出版社, pages: 302 - 306 * |
| 李宏等: "舰载飞机飞行试验测试工程", 31 March 2019, 北京：航空工业出版社, pages: 115 * |
| 王培元等: "航空侦察图像三维重建理论与工程实现", 30 September 2022, 北京：国防工业出版社, pages: 27 - 36 * |
| 高凯亮;覃团发;陈跃波;常侃;: "一种混合高斯背景模型下的像素分类运动目标检测方法", 南京大学学报(自然科学版), no. 02, 30 March 2011 (2011-03-30), pages 100 - 105 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118864391A (en) * | 2024-07-04 | 2024-10-29 | 无锡云程电力科技有限公司 | Photovoltaic module hot spot detection method and system based on infrared thermal imager |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107680054B (en) | Multi-source image fusion method in haze environment | |
| Xiao et al. | Making of night vision: Object detection under low-illumination | |
| CN111046880B (en) | Infrared target image segmentation method, system, electronic equipment and storage medium | |
| CN110264448B (en) | A method of insulator fault detection based on machine vision | |
| CN105069807B (en) | A kind of stamped workpieces defect inspection method based on image procossing | |
| CN112288008B (en) | Mosaic multispectral image disguised target detection method based on deep learning | |
| CN107784669A (en) | A kind of method that hot spot extraction and its barycenter determine | |
| AU2020272936B2 (en) | Methods and systems for crack detection using a fully convolutional network | |
| CN107633226A (en) | A kind of human action Tracking Recognition method and system | |
| CN116664565A (en) | Hidden crack detection method and system for photovoltaic solar cell | |
| CN107944354B (en) | Vehicle detection method based on deep learning | |
| CN119091234B (en) | Intelligent decision-making and response method and system based on data analysis | |
| Cruz-Rojas et al. | A novel comparison of image semantic segmentation techniques for detecting dust in photovoltaic panels using machine learning and deep learning | |
| CN116486273B (en) | A Method for Extracting Water Body Information from Small Sample Remote Sensing Images | |
| CN111985381A (en) | A Guided Region Dense Crowd Counting Method Based on Scaled Convolutional Neural Networks | |
| CN110660065A (en) | An Infrared Fault Detection and Recognition Algorithm | |
| CN113033558A (en) | Text detection method and device for natural scene and storage medium | |
| CN115830004A (en) | Surface defect detection method, device, computer equipment and storage medium | |
| CN107992863B (en) | Multi-resolution grain insect variety visual identification method | |
| CN117952935A (en) | Photovoltaic panel shadow heating spot identification method based on visible light image threshold segmentation | |
| CN110047041A (en) | A kind of empty-frequency-domain combined Traffic Surveillance Video rain removing method | |
| CN118823481A (en) | A bridge detection image recognition method and system based on artificial intelligence | |
| CN116129280B (en) | Method for detecting snow in remote sensing image | |
| Khoomboon et al. | A land cover mapping algorithm for thin to medium cloud-covered remote sensing images using a level set method | |
| CN119478956B (en) | Automatic identification card graph cutting and OCR (optical character recognition) method and system based on AI (advanced technology attachment) |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |