CN117197682B - Method for blind pixel detection and removal by long-wave infrared remote sensing image - Google Patents
Method for blind pixel detection and removal by long-wave infrared remote sensing image Download PDFInfo
- Publication number
- CN117197682B CN117197682B CN202311119819.5A CN202311119819A CN117197682B CN 117197682 B CN117197682 B CN 117197682B CN 202311119819 A CN202311119819 A CN 202311119819A CN 117197682 B CN117197682 B CN 117197682B
- Authority
- CN
- China
- Prior art keywords
- pixel
- blind
- pixels
- value
- remote sensing
- 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.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 24
- 238000000034 method Methods 0.000 title claims abstract description 14
- 238000013145 classification model Methods 0.000 claims abstract description 20
- 238000006467 substitution reaction Methods 0.000 claims abstract description 8
- 238000004519 manufacturing process Methods 0.000 claims abstract description 5
- 238000007477 logistic regression Methods 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 230000008901 benefit Effects 0.000 abstract description 6
- 230000000694 effects Effects 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
Landscapes
- Image Analysis (AREA)
- Transforming Light Signals Into Electric Signals (AREA)
Abstract
The invention discloses a method for blind pixel detection and removal of a long-wave infrared remote sensing image, which relates to the technical field of satellite remote sensing image processing and comprises the following steps: acquiring multi-scene continuous frame long wave infrared remote sensing images; constructing a 3D-CNN pixel classification model; inputting an image to be subjected to blind pixel detection into a model; classifying by using a 3D-CNN network model; acquiring pixel position coordinates of blind pixels and manufacturing an infrared image blind pixel position table; traversing the classification category to which the non-blind pixel adjacent to the blind pixel position belongs; calculating the average value of all pixels in each classification category; acquiring the median value of all the mean values; using a median value to replace the gray value of the blind pixel point pixel; and by analogy, completing pixel value substitution of all the blind pixels, and obtaining the infrared image after blind pixel compensation. The invention has the advantages that: modeling is carried out based on multi-scene continuous frame images, and blind pixel categories are classified based on pixel level classification, so that automatic detection and removal of blind pixels are realized.
Description
Technical Field
The invention relates to the technical field of satellite remote sensing image processing, in particular to a method for blind pixel detection and removal by a long-wave infrared remote sensing image.
Background
Due to various limitations of factors such as technology, manufacturing materials and the like, the infrared focal plane array detector has a blind pixel problem in the imaging process, and the blind pixels comprise fixed blind pixels and random blind pixels. The existence of the blind pixels can interfere with the subsequent detection and identification of the target and seriously affect the imaging quality, so that the blind pixels in the infrared image are detected and removed.
There are also some methods for researching detection and removal of infrared image blind pixels at present, and the blind pixel detection method is that an image frame is divided according to the size of a window by an infrared blind pixel detection algorithm based on an adjustable threshold window, and the size relation between the average value and standard deviation in the window is calculated by adjusting the threshold value in the window, so that the purpose of judging the blind pixels is achieved. In the prior art, a 3 sigma detection method based on image Gaussian normal distribution is provided, windowing is performed by taking a pixel as a center, the average value and standard deviation after windowing are calculated, and then whether the pixel is a blind pixel or not is judged by comparing whether the deviation between the gray level of the pixel in the center of the window and the average value is larger than 3 times of the standard deviation. Blind pixel removing methods such as median substitution, adjacent point substitution, first-order linear interpolation substitution, and intra-neighborhood multipoint average substitution.
The existing blind pixel detection method has the problems that the blind pixels are missed to detect, the infrared small target is easily misjudged as the blind pixels, newly added blind pixels in the image are difficult to detect, the effect on complex scenes is poor, and the like. The existing blind pixel removing method is easy to generate compensation failure problem for continuous blind pixels in practical engineering application.
Disclosure of Invention
In order to solve the defects of the prior art, the invention aims to provide a method for carrying out blind pixel detection and removal on long-wave infrared remote sensing images based on modeling of multi-scene continuous frame images, classifying blind pixel categories based on pixel level classification, and realizing automatic detection and removal of blind pixels.
In order to solve the technical problems, the technical scheme provided by the invention is as follows: a method for blind pixel detection and removal by long-wave infrared remote sensing images comprises the following steps:
S1, acquiring a long-wave infrared remote sensing image of a multi-scene continuous frame; the long-wave infrared remote sensing image simultaneously comprises space information and time information;
S2, integrating the space information and the time information in the long-wave infrared remote sensing image together to construct a space-time combined 3D-CNN pixel classification model, wherein pixels belonging to blind pixels are classified into 5 types, and other pixels are classified into 5 types according to the DN of the gray value of the pixel; if the gray value of other pixels is Max and Min, i= (Max-min+1)/5, the classification rule is as follows:
s3, extracting a W X H space-time cube in the long-wave infrared remote sensing image as sample data, wherein W X W is the size of a space neighborhood, H is the number of continuous frames in the long-wave infrared remote sensing image, and dividing the sample data into a training set and a verification set according to the proportion of 1:1:8;
S4, inputting the sample data into the 3D-CNN pixel classification model, and training the 3D-CNN pixel classification model to obtain a trained 3D-CNN pixel classification model; the training set enters four convolution layers, the convolution kernel size is 3 multiplied by 7, the step length is 2, dropout with the ratio of 0.3 is added after the last convolution layer, then the training set enters a full-connection layer, the full-connection layer changes a three-dimensional feature cube into a one-dimensional feature vector, and finally the class of the sample data is output by using a Logistic regression classifier aiming at a multitasking softmax; the loss function of the 3D-CNN pixel classification model is as follows: Wherein m represents the size of mini-batch, and x i and z i represent the predicted value and the actual value of the ith sample data in each batch respectively; loss is the loss function;
S5, inputting the image to be subjected to blind pixel detection into a trained 3D-CNN pixel classification model for classification, and outputting the class of the classified pixel, wherein the class comprises the class of the blind pixel;
S6, acquiring pixel position coordinates of blind pixels, and manufacturing an infrared image blind pixel position Excel table; the Excel table comprises three columns, wherein the first column is a serial number, the values are 1 to n, n is the total number of blind pixels, the second column is an abscissa value of a blind pixel coordinate, and the third column is an ordinate value corresponding to the abscissa value of the blind pixel coordinate;
S7, traversing classification categories to which non-blind pixel adjacent to the blind pixel position belongs, specifically: judging a neighborhood region where the blind pixel is located according to the position where the blind pixel is located in the image; determining non-blind pixel according to the neighborhood region, and determining a specific category belonging to the classification category according to the determined non-blind pixel;
s8, calculating the average value of all pixels in each classification category;
S8, obtaining the median value of all the average values;
S9, using a median value to replace the gray value of the blind pixel point pixel;
s10, and so on, completing pixel value substitution of all the blind pixels so as to obtain the infrared image after blind pixel compensation.
Compared with the prior art, the invention has the advantages that:
1. the invention combines the advantages of the 3D-CNN model using the 3D convolution cube, which can extract the time dimension characteristics, and uses the continuous frame infrared image to construct the model, which can more accurately distinguish the blind pixel category from other pixel categories, and avoid the omission of the blind pixel. The model is classified based on pixel level, so that the problem of misjudging a small target as a blind pixel can be solved;
2. The invention can detect the fixed blind pixels in the image and accurately identify the newly increased blind pixels which are difficult to solve in the past;
3. according to the blind pixel removing method provided by the invention, the pixel values of the classified non-blind pixel categories are only used for calculation, so that the problem of compensation failure aiming at continuous blind pixels can be avoided;
4. The invention also has better detection effect on the blind pixels in the complex scene by establishing the blind pixel detection model.
Drawings
FIG. 1 is a flow chart of the present invention.
Detailed Description
The following description of the technical solutions in the embodiments of the present invention will be clear and complete, and it is obvious that the described embodiments are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The present invention will be described in further detail with reference to the following examples in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
Examples
A method for blind pixel detection and removal by long-wave infrared remote sensing images comprises the following steps:
S1, acquiring a long-wave infrared remote sensing image of a multi-scene continuous frame; the long-wave infrared remote sensing image simultaneously comprises space information and time information;
S2, integrating the space information and the time information in the long-wave infrared remote sensing image together to construct a space-time combined 3D-CNN pixel classification model, wherein pixels belonging to blind pixels are classified into 5 types, and other pixels are classified into 5 types according to the DN of the gray value of the pixel; if the gray value of other pixels is Max and Min, i= (Max-min+1)/5, the classification rule is as follows:
s3, extracting a W X H space-time cube in the long-wave infrared remote sensing image as sample data, wherein W X W is the size of a space neighborhood, H is the number of continuous frames in the long-wave infrared remote sensing image, and dividing the sample data into a training set and a verification set according to the proportion of 1:1:8;
S4, inputting the sample data into the 3D-CNN pixel classification model, and training the 3D-CNN pixel classification model to obtain a trained 3D-CNN pixel classification model; the training set enters four convolution layers, the convolution kernel size is 3 multiplied by 7, the step length is 2, dropout with the ratio of 0.3 is added after the last convolution layer, then the training set enters a full-connection layer, the full-connection layer changes a three-dimensional feature cube into a one-dimensional feature vector, and finally the class of the sample data is output by using a Logistic regression classifier aiming at a multitasking softmax; the loss function of the 3D-CNN pixel classification model is as follows: Wherein m represents the size of mini-batch, and x i and z i represent the predicted value and the actual value of the ith sample data in each batch respectively; loss is the loss function;
S5, inputting the image to be subjected to blind pixel detection into a trained 3D-CNN pixel classification model for classification, and outputting the class of the classified pixel, wherein the class comprises the class of the blind pixel;
S6, acquiring pixel position coordinates of blind pixels, and manufacturing an infrared image blind pixel position Excel table; the Excel table comprises three columns, wherein the first column is a serial number, the values are 1 to n, n is the total number of blind pixels, the second column is an abscissa value of a blind pixel coordinate, and the third column is an ordinate value corresponding to the abscissa value of the blind pixel coordinate;
S7, traversing classification categories to which non-blind pixel adjacent to the blind pixel position belongs, specifically: judging a neighborhood region where the blind pixel is located according to the position where the blind pixel is located in the image; determining non-blind pixel according to the neighborhood region, and determining a specific category belonging to the classification category according to the determined non-blind pixel; assuming that a certain blind pixel is located at a non-boundary position of an image and the pixel coordinates are (i, j), eight adjacent pixels are in total in a 3×3 neighborhood, the pixel coordinates are (i-1, j-1), (i-1, j), (i, j+1), (i, j-1), (i, j+1), (i+1, j), (i+1, j+1), and the blind pixel is located at the boundary of the image. It is further assumed that these eight pixels belong to five of the classification categories, namely category 1, category 2, category 3, category 4, category 5;
| (i-1，j-1) | (i-1，j) | (i，j+1) |
| (i，j-1) | (i，j) | (i，j+1) |
| (i+1，j-1) | (i+1，j) | (i+1，j+1) |
The subsequent step calculates the average value of all pixels in each classification category, for example, calculates the average value of the gray values of the pixels included in each of the five categories, which are respectively 350, 760, 1190, 780, 800; acquiring the median value of all the mean values, wherein the median value is 780; the median value is used to replace the gray value of the blind pixel, such as assigning 780 the pixel value of the single blind pixel (i, j) described above.
S8, calculating the average value of all pixels in each classification category;
S8, obtaining the median value of all the average values;
S9, using a median value to replace the gray value of the blind pixel point pixel;
s10, and so on, completing pixel value substitution of all the blind pixels so as to obtain the infrared image after blind pixel compensation.
The advantage of time dimension characteristics can be extracted by combining a 3D-CNN model with a 3D convolution cube, a classification model based on pixel levels is constructed by using continuous frame infrared images, so that blind pixel categories are classified, blind pixel categories are detected, the blind pixel categories are more accurately distinguished from other pixel categories, missing detection and false detection of the blind pixels are avoided, and newly added blind pixels can be detected. The method for replacing the blind pixel gray value by using the gray average value of a certain non-blind pixel class in the blind pixel neighborhood is provided, so that the problem of compensation failure aiming at continuous blind pixels is avoided.
The invention and its embodiments have been described without limitation, and the examples shown are only one of the embodiments of the invention, without the actual embodiment being limited thereto. In summary, those skilled in the art, having benefit of this disclosure, will appreciate that many changes can be made without departing from the spirit and scope of the invention as disclosed herein.
Claims (1)
1. The method for blind pixel detection and removal by using the long-wave infrared remote sensing image is characterized by comprising the following steps of:
S1, acquiring a long-wave infrared remote sensing image of a multi-scene continuous frame; the long-wave infrared remote sensing image simultaneously comprises space information and time information;
S2, integrating the space information and the time information in the long-wave infrared remote sensing image together to construct a space-time combined 3D-CNN pixel classification model, wherein pixels belonging to blind pixels are classified into 5 types, and other pixels are classified into 5 types according to the DN of the gray value of the pixel; if the gray value of other pixels is Max and Min, i= (Max-min+1)/5, the classification rule is as follows:
s3, extracting a W X H space-time cube in the long-wave infrared remote sensing image as sample data, wherein W X W is the size of a space neighborhood, H is the number of continuous frames in the long-wave infrared remote sensing image, and dividing the sample data into a verification set, a test set and a training set according to the proportion of 1:1:8;
S4, inputting the sample data into the 3D-CNN pixel classification model, and training the 3D-CNN pixel classification model to obtain a trained 3D-CNN pixel classification model; the training set enters four convolution layers, the convolution kernel size is 3 multiplied by 7, the step length is 2, dropout with the ratio of 0.3 is added after the last convolution layer, then the training set enters a full-connection layer, the full-connection layer changes a three-dimensional feature cube into a one-dimensional feature vector, and finally the class of the sample data is output by using a Logistic regression classifier aiming at a multitasking softmax; the loss function of the 3D-CNN pixel classification model is as follows: Wherein m represents the size of mini-batch, and x j and z j represent the predicted value and the actual value of the jth sample data in each batch, respectively; loss is the loss function;
S5, inputting the image to be subjected to blind pixel detection into a trained 3D-CNN pixel classification model for classification, and outputting the class of the classified pixel, wherein the class comprises the class of the blind pixel;
S6, acquiring pixel position coordinates of blind pixels, and manufacturing an infrared image blind pixel position Excel table; the Excel table comprises three columns, wherein the first column is a serial number, the values are 1 to n, n is the total number of blind pixels, the second column is an abscissa value of a blind pixel coordinate, and the third column is an ordinate value corresponding to the abscissa value of the blind pixel coordinate;
S7, traversing classification categories to which non-blind pixel adjacent to the blind pixel position belongs, specifically: judging a neighborhood region where the blind pixel is located according to the position where the blind pixel is located in the image; determining non-blind pixel according to the neighborhood region, and determining a specific category belonging to the classification category according to the determined non-blind pixel;
s8, calculating the average value of all pixels in each classification category;
S8, obtaining the median value of all the average values;
S9, using a median value to replace the gray value of the blind pixel point pixel;
s10, and so on, completing pixel value substitution of all the blind pixels so as to obtain the infrared image after blind pixel compensation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311119819.5A CN117197682B (en) | 2023-09-01 | 2023-09-01 | Method for blind pixel detection and removal by long-wave infrared remote sensing image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311119819.5A CN117197682B (en) | 2023-09-01 | 2023-09-01 | Method for blind pixel detection and removal by long-wave infrared remote sensing image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117197682A CN117197682A (en) | 2023-12-08 |
| CN117197682B true CN117197682B (en) | 2024-06-25 |
Family
ID=88984350
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311119819.5A Active CN117197682B (en) | 2023-09-01 | 2023-09-01 | Method for blind pixel detection and removal by long-wave infrared remote sensing image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117197682B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117475242B (en) * | 2023-12-27 | 2024-03-08 | 南京信息工程大学 | Infrared detector blind pixel dynamic detection method based on progressive classification model |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113192059A (en) * | 2021-05-24 | 2021-07-30 | 北京市遥感信息研究所 | On-orbit ship detection method based on infrared remote sensing image |
| CN115019156A (en) * | 2022-06-06 | 2022-09-06 | 中国科学院光电技术研究所 | Blind flash element detection and compensation method based on space-time characteristics |
Family Cites Families (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101908209B (en) * | 2010-07-29 | 2012-03-28 | 中山大学 | A Blind Element Compensation Algorithm for Infrared Thermal Image Based on Cubic Spline |
| US9876968B2 (en) * | 2014-12-08 | 2018-01-23 | Ci Systems (Israel) Ltd. | Drift correction method for infrared imaging device |
| CN105928622A (en) * | 2016-04-26 | 2016-09-07 | 成都市晶林科技有限公司 | Infrared focal plane detector blind pixel correction method |
| CN108805853A (en) * | 2017-04-28 | 2018-11-13 | 武汉多谱多勒科技有限公司 | A kind of infrared image blind pixel detection method |
| CN107346533A (en) * | 2017-07-12 | 2017-11-14 | 中国科学院上海技术物理研究所 | A kind of blind-pixel elimination method of view-based access control model characteristic |
| CN108491852B (en) * | 2018-01-31 | 2021-08-06 | 中国科学院上海技术物理研究所 | A Blind Element Detection Method for Area Array Infrared Focal Plane |
| CN109584205A (en) * | 2018-10-19 | 2019-04-05 | 中国人民解放军海军工程大学 | Blind element detection method and device |
| CN111353495B (en) * | 2018-12-20 | 2023-07-18 | 中国科学院沈阳自动化研究所 | A Statistical Analysis and Evaluation Method of Infrared Imaging Blind Elements |
| CN111353968B (en) * | 2018-12-20 | 2023-03-31 | 中国科学院沈阳自动化研究所 | Infrared image quality evaluation method based on blind pixel detection and analysis |
| CN110598736B (en) * | 2019-08-06 | 2022-12-20 | 西安理工大学 | Power equipment infrared image fault positioning, identifying and predicting method |
| CN110440930B (en) * | 2019-08-09 | 2021-11-19 | 上海集成电路研发中心有限公司 | Low-noise infrared detector structure, chip and testing method thereof |
| CN112419359A (en) * | 2020-11-10 | 2021-02-26 | 中国电子科技集团公司第十一研究所 | Infrared dim target detection method and device based on convolutional neural network |
| CN114897712A (en) * | 2022-04-07 | 2022-08-12 | 昆明物理研究所 | Infrared image blind pixel compensation method based on two wave bands |
| CN115100068A (en) * | 2022-07-08 | 2022-09-23 | 郑州信大先进技术研究院 | Infrared image correction method |
| CN115880241A (en) * | 2022-11-29 | 2023-03-31 | 凌云光技术股份有限公司 | A method and system for detecting and processing infrared image blind elements |
| CN116630227A (en) * | 2023-03-07 | 2023-08-22 | 中国科学院微电子研究所 | A blind element detection method, device, computer equipment and medium |
-
2023
- 2023-09-01 CN CN202311119819.5A patent/CN117197682B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113192059A (en) * | 2021-05-24 | 2021-07-30 | 北京市遥感信息研究所 | On-orbit ship detection method based on infrared remote sensing image |
| CN115019156A (en) * | 2022-06-06 | 2022-09-06 | 中国科学院光电技术研究所 | Blind flash element detection and compensation method based on space-time characteristics |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117197682A (en) | 2023-12-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111383209B (en) | Unsupervised flaw detection method based on full convolution self-encoder network | |
| CN108961235B (en) | Defective insulator identification method based on YOLOv3 network and particle filter algorithm | |
| CN112819772B (en) | High-precision rapid pattern detection and recognition method | |
| CN111784633B (en) | Insulator defect automatic detection algorithm for electric power inspection video | |
| CN119006469B (en) | Automatic detection method and system for surface defects of substrate glass based on machine vision | |
| CN115147418B (en) | Compression training method and device for defect detection model | |
| CN111080582B (en) | Method for detecting defects of inner and outer surfaces of workpiece | |
| CN112233067A (en) | Hot rolled steel coil end face quality detection method and system | |
| CN113393426B (en) | Steel rolling plate surface defect detection method | |
| US12217411B2 (en) | Inspection apparatus, unit selection apparatus, inspection method, and computer-readable storage medium storing an inspection program | |
| CN115082449B (en) | Electronic component defect detection method | |
| CN114187267B (en) | Stamping part defect detection method based on machine vision | |
| CN117522864B (en) | European pine plate surface flaw detection method based on machine vision | |
| CN115063430A (en) | Electric pipeline crack detection method based on image processing | |
| CN117197682B (en) | Method for blind pixel detection and removal by long-wave infrared remote sensing image | |
| WO2025148572A1 (en) | Target detection method and apparatus, electronic device, and storage medium | |
| CN111814852A (en) | Image detection method, image detection device, electronic equipment and computer-readable storage medium | |
| CN111724376A (en) | A paper defect detection method based on texture feature analysis | |
| CN116563262A (en) | Building crack detection algorithm based on multiple modes | |
| CN114863464B (en) | Second-order identification method for PID drawing picture information | |
| CN113496159A (en) | Multi-scale convolution and dynamic weight cost function smoke target segmentation method | |
| CN114943738A (en) | Sensor packaging curing adhesive defect identification method based on visual identification | |
| CN119380123A (en) | A power grid anomaly detection method and system based on multidimensional residual attention mechanism | |
| CN117455917B (en) | Establishment of false alarm library of etched lead frame and false alarm on-line judging and screening method | |
| CN117314901B (en) | A scale-adaptive neural network system for chip detection |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |