CN102116659B - Interval convergence based stock bin level detection method - Google Patents

Abstract

The invention is a method for continuously detecting the material level of a silo based on a digital camera and auxiliary lighting equipment and adopting an image processing method. The present invention shoots images by means of two classifications, performs image preprocessing according to the image characteristics, and calculates the image entropy of the processed image, and obtains an approximation interval after the primary classification (coarse classification) image is processed and analyzed. The subdivision (subdivision) takes images, calculates the image entropy of the preprocessed image, selects a suitable image for image material level line detection, obtains the material level line, and obtains the bin material level value after calculation.

Description

A method for detection of silo material level based on interval convergence

technical field

The invention relates to the detection of the material level of a silo. The invention specifically relates to the non-contact dynamic detection of the material level of the silo by using a digital camera to take images and perform image processing. the

Background technique

Silo material level detection is an important measure for safe production. Over the years people have adopted various methods to detect the material level of the silo. Commonly used detection methods are: hammer type, electrode type, capacitive type, machine rod type, weighing type, rotary wing wheel type, radar type, ultrasonic type, laser type, nuclear type, etc. Among them, the hammer type, electrode type, capacitive type, machine rod type, weighing type and rotary wing wheel type are contact measurement methods, and the rest are non-contact measurement methods. The methods that can measure the limit position are: hammer type, electrode type, nuclear type and laser type. The methods for continuous measurement of material level include: radar type, ultrasonic type, machine rod type, weighing type, nuclear type, laser type, etc. the

Depth measurement by means of image processing is widely used. Most of the current methods are to use digital cameras to automatically focus, directly shoot images, and then perform depth and height measurements through image processing. For details, see the invention patent of CN1378086. However, for solid material silos, due to the extremely harsh image shooting environment, the captured images have the following characteristics:

(1) The silo has high dust concentration, high humidity, low illumination, poor image quality, and it is difficult for the camera to automatically focus. the

(2) The illuminance fluctuates frequently in the silo. For example, there are many large-scale equipment in the coal mine, and the power grid is greatly disturbed, resulting in illuminance fluctuations. the

(3) For some special occasions, such as underground coal bunkers in coal mines, due to explosion-proof requirements, the power of equipment and lighting should be as small as possible. the

Therefore, due to the special image environment, it is difficult to meet the real-time and reliability requirements of silo level measurement by using ordinary image processing methods for depth measurement, and it is also difficult to achieve long-term and stable detection. the

The patent publication number is CN101270981, which is a material level measurement method and device based on machine vision. It proposes a material level detection method for underground coal mine silos. However, the reliability of this method is difficult to guarantee, and the actual application effect is not good. the

technical problem:

Due to the special image environment of the silo, the current existing methods have the following problems:

(1) It is difficult for digital cameras to automatically focus and take reliable images;

(2) The method of measuring material level based on such images has low reliability and precision;

(3) Since the silo is a closed and semi-closed environment, it is difficult to monitor and verify the operation of the equipment. There is no self-test mechanism based on the reliability of the detection system, and the reliability of the system cannot be guaranteed, which is difficult for practical application. the

The invention uses a digital camera to take images and perform image processing for non-contact dynamic detection of the material level of the silo, and proposes a method for calculating the characteristic texture image entropy and PCNN for material level calculation for the silo environment. The medium image shooting adopts the method of two stages, which has the following advantages:

(1) The method of calculating feature texture image entropy and PCNN for material level calculation reduces the requirements for image quality and enhances the adaptability to the harsh image environment of the silo;

(2) The image shooting in the detection adopts the method of two classifications, which increases the real-time performance and reliability of the material level detection. the

Contents of the invention

The invention consists of three parts: digital camera, auxiliary lighting and image processing. the

Digital camera:

The digital camera should be installed on the top of the silo, avoiding the silo discharge port and related discharge port facilities, so as not to be blocked from the shooting angle (the layout of the digital camera is shown in Figure 1). The digital camera system shall include a transparent sealing cover, a dust removal device for the sealing cover and respective fixing devices. the

The relevant parameters of the digital camera are selected according to the size of the silo and the intensity of the auxiliary lighting. the

Auxiliary lighting:

The auxiliary lighting adopts a group of spotlights, one of which is set adjacent to the digital camera, and the rest are set on the circumference of the coal bunker wall at an equal angle. The light source adopts a monochromatic light source, and a red light or an infrared light source with a longer wavelength is selected. According to the actual size of the coal bunker and the environment, select the power of the light source and the number of spotlights (the auxiliary lighting layout is shown in Figure 2). the

Auxiliary lighting is mainly designed for the image characteristics of silos with particularly harsh image environments, such as underground coal bunkers in coal mines. There are two main difficulties in silo image imaging. One is restricted by production conditions, and the silo environment has high dust concentration and humidity. The large feature causes the illumination to attenuate quickly and the floodlight is insufficient; secondly, for safety reasons, the lighting power should be as low as possible. The multi-angle lighting design can effectively overcome the above difficulties. the

Image processing includes:

(1) The method of calculating the image entropy of each image separately after preprocessing the image:

The flow chart of image entropy processing is shown in Figure 3. the

Image entropy calculation method:

(a) Grayscale stretching of the image

Since the purpose of image processing here is to detect material level, grayscale images can be used. The grayscale adopts 8-bit grayscale. Since the gray scale of the silo image is often unevenly distributed, it is generally necessary to pre-stretch the gray scale. the

Methods as below:

When the grayscale is a discrete value, the frequency approximates the probability value, that is:

p _r (r _k )=n _k /n 0≤r _k ≤1 k=0,1,...,l-1;

In the formula: l is the total number of gray levels, p _r (r _k ) is the probability of taking the kth gray level value, n _k is the number of times the kth level gray level appears in the image, and n is the total number of pixels in the image.

${\mathrm{the\; s}}_k=T\left(r_k\right)={\mathrm{\Σ}}_{j=0}^k\frac{{\mathrm{no}}_j}{\mathrm{no}}={\mathrm{\Σ}}_{j=0}^k{p}_r\left(r_j\right)$ 0 ≤ r _k ≤ 1 k = 0, 1, ..., l-1;

(b) Differentiate the image

Record an image as X(l, j), and the differentiated image as Y(l, j). but:

Y(1,j)=X(1,j);

Y(l,j)=X(l,j)-X(i-1,j), (i>1);

The image mainly contains two parts, the upper part with the warehouse wall as the main body and the lower part with the material surface as the main body. The lower part with the material surface as the main body presents a mixture of bright and dark small areas due to the irregular light reflection. In the image Y obtained after calculation, the gray value of the upper part with the warehouse wall as the main body approaches 0, and the lower part with the material surface as the main body obtains the boundaries of bright and dark small areas, and this part will be used as the main body of information entropy calculation . the

(c) Perform binary segmentation on the image

In order to further clarify the differentiated image, a threshold is selected to perform binary segmentation between the image background and the obtained small area boundary. Get the image Z(i, j);

(d) Calculate the information entropy of the binary image

Image entropy H(P): H(P)=-P ₁ lnP ₁ -P ₀ lnP ₀ ;

Among them, P ₁ and P ₀ represent the probability when Z is 1 and 0 respectively.

(2) The calculation method of the material level value obtained by calculating the material level edge:

For the selected P images, first, compare the edge of the material level with the respective corresponding material level scale images to obtain a set of P material level values L ₁ (i=1,...,P), and then use the following formula The actual material level value L is obtained: L=(L1+L2+...+Lp)/p.

The method of obtaining the scale image of the material level: Divide the silo with a depth of h into w parts at the interval of the absolute error value Δh, w=h/Δh, corresponding to w gears, and the focal length of the camera lens corresponds to the object distance from the first gear At the beginning, go to the wth gear in sequence, and shoot according to the following method: when the focal length of the camera lens corresponds to the object distance at the Sth subdivision gear (S=1, 2, ..., w), the inner empty silo Set the ruler on the warehouse wall, and set the ruler successively at (S-t*Δh), (S-(t-1)*Δh), ..., (S-Δh), S, (S+Δh), ... , (S+(t-1)*Δh), and (S+t*Δh), take images respectively to obtain a group of (2t+1) images. After image processing of this group of images, the corresponding The material level scale images of S subdivided gears; according to the above method, the material level scale images of w subdivided gears in the whole silo can be obtained to form a material level scale image group of the whole silo. the

The silo material level detection method based on interval convergence includes the following steps:

(1) Divide into rough grades:

Roughly divide the bin with a depth of h into N equal parts (coarse classification), and use program control to adjust the focal length of the digital camera lens to capture images according to the corresponding division of the object distance, and obtain a set of N frames corresponding to the division image;

(2) Determine the convergence interval:

After the image is preprocessed, the image entropy of each image is calculated separately, and K images with larger entropy values (1<K<N) are selected, and the smallest gear among the gears corresponding to these K images is Ni, and the largest is Ni. If the stall is Nj, a convergence interval [Ni, Nj] is determined. the

(3) Divide into subdivision files:

Subdivide the interval [Ni, Nj] into M equal parts (subdivision files), and adjust the focal length of the digital camera lens to shoot images according to the corresponding division of the object distance, and obtain a set of M frames corresponding to the division of gears image;

(4) Edge detection and calculation of material level value:

After the image is preprocessed, the image entropy of each image is calculated separately, P images with larger entropy value are selected, and the material level edge is calculated by using the material level edge detection algorithm, and the material level value is calculated. the

The selection method of rough classification N and subdivision classification M:

The subdivision gear is set according to the absolute error value of the silo material level detection, that is, the gear interval of the subdivision gear is equal to the absolute error value; the rough subdivision is based on the full-scale actual distance of the silo material level detection and the detection speed It is required to be set, and the distance between the rough classification gears is generally an integer multiple of the distance between the fine classification gears. the

The value of the boundary point Ni, Nj of the convergence interval [Ni, Nj]:

For the convergence interval [Ni, Nj] determined by the images taken in rough bins, the ratio of the image entropy values of the images corresponding to Ni and Nj (comparing the smaller entropy value with the larger entropy value) should be greater than the specified ratio. the

Reliability Analysis:

Reliability Analysis 1:

According to the characteristics of the image entropy distribution, a convergence interval [Ni, Nj] is determined in the images taken in coarse classification; when the value is taken, the ratio of the image entropy value of the image corresponding to Ni and Nj (with a smaller entropy The entropy value with relatively large value) should be greater than the specified ratio k1. The value of K1 is taken according to the image environment of the silo, which is related to the environmental parameters such as the material type of the silo, the illuminance of the silo, humidity, and dust concentration. K1 should generally be greater than 0.8. If this condition cannot be met, it means that during the shooting process, the system is working abnormally and the data is unavailable. The image needs to be taken again. If the problem is still not resolved, it means that the system may have a hardware failure and needs to be repaired. the

The program flow of reliability analysis 1 is shown in Figure 4. the

Reliability Analysis 2:

The images taken by the meter subdivision file are B1, B2...B10, and the P images with larger entropy values in B1, B2...B10 are selected for material level detection, and the material level value (denoted as The error between M(i), i=1, 2, ..., P) should be smaller than the minimum error Y* required by the project, and it should be within the corresponding gear range. the

The program flow of reliability analysis 2 is shown in Figure 5. the

System workflow:

然后继续进行下一轮检测。  First, the height of the silo is recorded as h, the absolute error value of the required material level detection is recorded as d, and the classification is divided into subdivision and coarse classification. Roughly divide the full-scale range of the material level of the silo into N equal parts (coarse classification), and use program control to adjust the focal length of the digital camera lens to capture images according to the corresponding division of the object distance, and obtain a group of N corresponding division divisions. After the image is preprocessed, the image entropy of each image is calculated separately. For the image taken by this method, the closer the camera focal length is to the material level, the clearer the captured image and the greater the image entropy. Choose a higher entropy value. Large K images (1<K<N), remember that the smallest gear among the gears corresponding to these K images is Ni, and the largest gear is Nj, and a convergence interval [Ni, Nj] is determined, then this The interval should contain the actual material level value; the interval [Ni, Nj] is subdivided into M equal parts (subdivision files), according to the gears corresponding to the object distance, the digital camera lens focal length is used to adjust the program control method to shoot images, and the corresponding Divide a group of M images of stalls, preprocess the images and calculate the image entropy of each image respectively, then the entropy value of this group of images should be relatively close, because the material level of the silo is often not flat, each time The characteristics presented after blanking vary greatly, so the entropy value of the image cannot accurately reflect the material level, so the position of the image with the largest entropy value cannot be simply taken as the gear closest to the material level. At this time, select P images with a large entropy value, use the PCNN method to calculate the edge of the material level, and then obtain a set of P material level values L ₁ (i=1,... , P), and then get the actual material level value L by the following formula:

Then proceed to the next round of testing.

The working flow diagram of the silo material level measurement system is shown in Figure 6. the

Description of drawings

Figure 1 layout of digital camera

Figure 2 Auxiliary lighting layout diagram

Figure 3 Flow chart of image entropy processing

Figure 4 Reliability Analysis 1 Program Flow

Figure 5 Reliability analysis 2 program flow

Figure 6 Workflow diagram of the silo material level measurement system

Detailed ways

The present invention is further explained as follows with reference to accompanying drawing in conjunction with embodiment:

Taking the underground coal bunker of a coal mine as an example, the height of the coal bunker is 40m, the diameter is 8m, and the absolute error value of detection is 0.5m. The minimum silo limit is 4m. the

Set the subdivision level to 1m, and the coarse level to 4m. The digital camera uses a program to adjust the focal length of the lens, and the auxiliary lighting uses 8 spotlights. the

Before detection, the background image of the material level scale is prefabricated. the

Divide the silo with a depth of 40m into w parts at the interval of absolute error value Δh=0.5m, w=h/Δh=80, corresponding to w gears, and the focal length of the camera lens corresponding to the object distance starts from the first gear, Sequence to the wth gear, and shoot according to the following method: when the focal length of the camera lens corresponds to the object distance at the Sth subdivision gear (S=1, 2, ..., w), the inner empty silo wall Set the scale on the top, and set the scale successively at (S-t*Δh), (S-(t-1)*Δh), ..., (S-Δh), S, (S+Δh), ..., ( At S+(t-1)*Δh), (S+t*Δh) (here t=4), take images respectively to obtain a group of (2t+1=9) images, and perform image processing on this group of images After that, the material level scale image corresponding to the S subdivision level can be obtained; according to the above method, the material level scale images of w subdivision levels in the whole silo can be obtained, forming a group of material level scale images of the whole silo. the

Select k1=0.8, absolute error value=0.25m. the

Choose k=3, P=4. the

See attached drawing 1, which describes the installation position of the digital camera system. The digital camera system should be installed on the top of the silo, as close to the wall of the silo as possible, avoiding the silo discharge port and related facilities, so as not to be blocked from shooting angle. The digital camera system should include a transparent sealing cover, a dust removal device of the sealing cover and respective fixing devices, and the device should also meet the safety requirements of the application. It should meet the electrical explosion-proof requirements of underground coal mines. the

Referring to Figure 2, the installation of auxiliary lighting is described. The auxiliary lighting system adopts a group of spotlights, one of which is adjacent to the digital camera, and the rest are arranged on the circumference of the coal bunker wall at an equal angle. The light source adopts a monochromatic light source, and a red light or an infrared light source with a longer wavelength is selected. Select the power of the light source and the number of spotlights according to the actual size and environment of the coal bunker. Accompanying drawing 2 has described the situation of 8 spotlights. When the present invention is used in coal bunkers in coal mines, the device used in the present invention should also meet the electrical explosion-proof requirements of coal mines. the

The auxiliary lighting system is mainly designed for the image characteristics of silos with particularly harsh image environments, such as underground coal bunkers in coal mines. There are two main difficulties in image imaging of silos. One is that due to production conditions, the silo environment has high dust The characteristic of high humidity causes the illumination to attenuate quickly and the floodlight is insufficient; secondly, for safety reasons, the lighting power should be as low as possible. The multi-angle lighting design can effectively overcome the above difficulties. the

Referring to accompanying drawing 3, have described the feature image image entropy calculation method of detection method of the present invention, at first carry out preprocessing operation to image, comprise the following steps: gray stretching, differential calculation, binary segmentation, then calculate its image entropy. the

The flow chart of material level detection is shown in Figure 6. the

When starting the detection, firstly, 101, a group of 10 images can be obtained, proceed 102, 103 in sequence, and then judge 104, remember that the smallest of the three entropy values is Z1, and the largest is Z2, k'=Z1/Z2, then when When k'>k1, the verification is passed, indicating that the captured image is reliable, and the system is working normally. At this time, the verification flag is cleared, and the next step is performed. When k'<=k1, the captured image is unreliable, and the system is working Abnormal, at this time the verification flag should be zero, first set the verification flag, and then judge 105, when the flag X1 is 1, re-execute 101, 102, 103, and then enter 104 again, if the system The abnormality is only caused by accidental interference, then it should return to normal at this time, then k'>k1, the verification can pass, if there is a functional failure of the system, the verification cannot pass, when judging the verification flag, the flag is 1 , then a fault alarm will be issued. the

After the verification in 104 is passed, proceed to 106, and take pictures successively according to the subdivided files within the material level detection interval determined by one classification, and a group of 9 images can be obtained, and proceed to 107 and 108 in sequence, and use the method of PCNN to detect the material level The edge is detected, and then compared with the material level scale image at the corresponding gear to obtain the material level value. Finally judge 109, analyze the obtained P (P=4) material level values, calculate the difference, the error should be less than the required precision value, if the verification is passed, then continue to verify whether the material level value is at the corresponding gear, if passed, then It shows that the image is reliable and the system is working normally. At this time, the verification flag is cleared and the next step is performed. If any of the above two judgments fails, set the flag bit X1, and then judge 105, when the flag bit X1 is 1, re-execute 101, 102, 103, 104, 106, 107, 108, 109, if the system If the abnormality is only caused by occasional interference, it should return to normal at this time, and the verification can pass. If there is a functional failure of the system, the verification cannot pass. When judging the verification flag, if the flag is 1, a fault alarm will be issued. After passing 109, execute 110, calculate the average value and output the material level value of the silo, and then proceed to the next round of detection. the

Claims (4)

1. A silo material level detection method based on interval convergence, characterized in that: the depth of the silo of h is roughly divided into N equal parts, which are recorded as rough classification, and the stalls correspondingly divided according to the object distance are adopted by the program The control method adjusts the focal length of the digital camera lens to shoot images, and obtains a group of N images corresponding to the division of gears. After the images are preprocessed, the image entropy of each image is calculated separately, and K images with larger entropy values are selected, 1<K <N, K takes an integer value, remember that the smallest gear among the gears corresponding to the K images is Ni, and the largest gear is Nj, then a convergence interval [Ni, Nj] is determined; the interval [Ni, Nj ] is subdivided into M equal parts, which are recorded as subdivision files. According to the stalls corresponding to the object distance, the focal length of the digital camera lens is adjusted to take images by program control, and a group of M images corresponding to the divided stalls are obtained. After preprocessing, calculate the image entropy of each image separately, select P images with larger entropy values, use the PCNN material level edge detection algorithm to calculate the material level edge, and compare the material level edge with the corresponding material level scale image, Calculate the material level value. 2. The detection method according to claim 1, characterized in that: the method of comparing the edge of the material level with the respective corresponding material level scale images to calculate the material level value is specifically, for the selected P images, First, compare the material level edge with the corresponding material level scale images to obtain a set of P material level values L _i , where i=1,...,P, i take integer values, and then the actual Material level value L: L=(L ₁ +L ₂ +...+L _p )/P. 3. The detection method according to claim 1, characterized in that: the division method of the coarse classification and the subdivision is that the subdivision is set according to the absolute error value Δh of the detection of the material level in the silo, that is, the fine The gear spacing of the binning is equal to the absolute error value Δh; the coarse binning is set according to the depth h of the silo and the speed requirements for detection, and the gear spacing of the coarse binning is an integer multiple of the spacing of the finer bins. 4. detection method as claimed in claim 1, it is characterized in that: the convergence interval [Ni, Nj] determined by the image taken by coarse sub-grading, the ratio of the image entropy value of the corresponding image of Ni and Nj, during calculation, Use the smaller entropy value to compare the larger entropy value, which should be greater than the specified ratio.

Images