A belt detection method for a belt conveyor

Abstract

The present invention relates to the technical field of image processing. More specifically, the present invention relates to a method for detecting the belt of a belt conveyor, including collecting an image of the conveyor belt and obtaining a pothole image of the image by using a pothole recognition network; taking any pothole in the current frame pothole image as a target pothole, calculating a matching index between the target pothole and any pothole in the next frame pothole image to determine the target pothole in the consecutive frame pothole images, wherein the matching index between two potholes is positively correlated with the similarity between the surrounding potholes of the two potholes and is also positively correlated with the similarity between the two potholes; calculating the clarity of each target pothole in the consecutive frame pothole images. Through the calculation of the wear coefficient, the present invention can quantitatively analyze the wear of the conveyor belt and make a safety judgment based on a set threshold. This enables the wear detection not to rely on subjective evaluation, but to judge whether an alarm is needed through clear criteria, thereby enhancing the reliability of decision-making.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing. More specifically, the present invention relates to a method for detecting the belt of a belt conveyor. Background Art

[0002] Belt conveyors are widely used in multiple industries such as mining, manufacturing, and logistics for efficient material transportation. As its core component, the condition of the conveyor belt is directly related to the reliability and safety of the entire system. In coal mining, the conveyor belt is an essential and important equipment in coal mines, thermal power plants, and coal transportation systems. It is mainly applied in the processes of coal extraction, transportation, and loading and unloading, and is one of the key links in modern coal production. With the continuous development of the coal industry, the role of the conveyor belt as a transportation tool has become increasingly prominent. Its main function is to transport coal from the extraction site to the processing facility or to other transportation tools, saving a large amount of labor and time costs. However, long-term operation and harsh working environments make coal conveyor belts prone to problems such as wear, aging, and damage, thus affecting production efficiency and safety.

[0003] In related technologies, for example, the Chinese patent application document with the publication number CN112598720A discloses a method and system for detecting the wear state of a conveyor belt based on binocular photography. According to the surface contour feature information, the uniformity of the distribution spacing of the contour lines of the surface contour is determined, and thus the distribution state of the contour lines of the surface contour is determined according to the uniformity of the distribution spacing of the contour lines at each pixel point; finally, the surface wear position area of the conveyor belt is determined according to the uniformity of the distribution spacing of the contour lines of the surface contour, so as to accurately locate the surface wear position area of the conveyor belt.

[0004] During the process of coal material falling onto the conveyor belt, the impact of the coal material on the conveyor belt will cause different degrees of pits on the conveyor belt. However, since the size of the pits is uncertain, some smaller pits will have problems of inaccurate identification, and thus it is impossible to accurately detect the wear of the conveyor belt. Summary of the Invention

[0005] The present invention provides a method for detecting the belt of a belt conveyor, aiming to solve the problem in related technologies that the impact of coal material on the conveyor belt will cause different degrees of pits on the conveyor belt. However, since the size of the pits is uncertain, some smaller pits will have problems of inaccurate identification, and thus it is impossible to accurately detect the wear of the conveyor belt.

[0006] The present invention provides a method for detecting the belt of a belt conveyor, including: collecting an image of the conveyor belt and obtaining a pothole image of the image by using a pothole recognition network; taking any pothole in the current frame pothole image as a target pothole, calculating a matching index between the target pothole and any pothole in the next frame pothole image to determine the target pothole in the consecutive frame pothole images, wherein the matching index between two potholes is positively correlated with the similarity between the surrounding potholes of the two potholes and is also positively correlated with the similarity between the two potholes; calculating the clarity of each target pothole in the consecutive frame pothole images, and selecting the pothole image corresponding to the highest clarity of the pothole as the target image, wherein the clarity of the pothole reflects the degree of chaos of the pixel values in the pothole and the size of the difference in pixel values in the pothole; calculating the wear coefficient of the target pothole in the target image, and performing wear detection on the conveyor belt according to the size of the wear coefficient, wherein the wear coefficient of the target pothole is positively correlated with the area of the target pothole and is also positively correlated with the ratio of the minimum pixel value of the target pothole to the average pixel value of the normal area. By combining image processing technology and a deep learning network, the pothole recognition network is used to accurately extract the pothole features on the conveyor belt. By analyzing the pothole images and calculating parameters such as the matching index and clarity of the potholes, the detection accuracy of the wear condition of the conveyor belt can be effectively improved. Generally speaking, this method can efficiently, accurately and real-time detect the wear condition of the coal conveyor belt, improve safety and reduce manual intervention.

[0007] Further, performing wear detection on the conveyor belt according to the size of the wear coefficient includes: in response to the wear coefficient being greater than the wear evaluation threshold, giving an alarm prompt; in response to the wear coefficient being less than or equal to the wear evaluation threshold, not giving an alarm. By setting the wear evaluation threshold, when the wear coefficient is greater than the wear evaluation threshold, an alarm prompt can be triggered. This can timely discover potential safety hazards and reduce the risk of accidents.

[0008] Further, the calculation formula for the wear coefficient of the target pothole in the target image is: ; where is the wear coefficient of the target pothole in the target image, is the number of pixel points of the target pothole in the target image, is the pixel value sequence of the target pothole in the target image, is the mean value of the pixel values of the non-pothole area in the target image, is the hyperbolic tangent function. Considering multiple factors such as the area and pixel value difference of the target pothole when calculating the wear coefficient can comprehensively reflect the wear condition of the conveyor belt.

[0009] Further, calculating the wear coefficient of the target pothole further includes: obtaining the difference between the pixel values of the target pothole and the non-pothole area in the target image to obtain a pothole difference map, performing k-means clustering on the pothole difference map, determining the number of clustering clusters using the silhouette coefficient method, and calculating the wear coefficient of the target pothole based on the number of clustering clusters. The calculation formula is: ; where is the wear coefficient of the target pothole, is the pixel value sequence composed of the pixel points of the clustering centers of all clustering clusters, is the average value of the pixel values of the non-pothole area in the target image, is the number of clustering centers of all clustering clusters, is the hyperbolic tangent function.

[0010] Further, calculating the clarity of each target pothole in the consecutive frame pothole images, the calculation formula is: ; where is the clarity of the target pothole in the frame pothole image, is the proportion of the pixel value in the target pothole in the frame pothole image, is the pixel value sequence of the target pothole in the frame pothole image.

[0011] Further, calculating the matching index between the target pothole and any pothole in the next frame pothole image, the calculation formula is: ; where is the matching index between the target pothole and the th pothole in the next frame pothole image, is the pixel value matrix of the target pothole, is the pixel value matrix of the th pothole in the next frame, is the pixel value matrix of the th pothole in the next frame, is the position vector formed by the geometric center of the th pothole around the target pothole pointing to the geometric center of the target pothole, is the position vector formed by the geometric center of the th pothole around the th pothole in the next frame pothole image pointing to the geometric center of the th pothole, and are both similarity calculations.

[0012] Further, the training process of the pothole recognition network includes: marking the pothole pixel points in the image of the conveyor belt as 1 and the non-pothole pixel points as 0 to obtain a 0-1 labeled image; inputting the 0-1 labeled image into a semantic segmentation network for training; and completing the training when the loss function is less than a preset value or reaches a preset number of training times to obtain the pothole recognition network. By training the pothole recognition network, the system can automatically extract effective features from the image. And with the continuous optimization of network training, the recognition accuracy and adaptability of the system will gradually improve, enhancing the feasibility of its long-term use.

[0013] Further, obtaining the pothole image of the image by using the pothole recognition network includes: inputting the image of the conveyor belt into the pothole recognition network to obtain a 0-1 labeled image; and multiplying the 0-1 labeled image matrix by the image matrix of the conveyor belt to obtain the pothole image.

[0014] Further, the image of the conveyor belt is collected by a depth camera, where the pixel points in the image contain gray value information and depth information. By using the combination of depth information and gray value, not only the shape of the pothole can be detected, but also the depth and size of the pothole can be more accurately identified, improving the adaptability in complex working environments.

[0015] Further, obtaining each pothole in the pothole image includes: using the connected component analysis method to obtain each connected component in the pothole image, where one connected component is one pothole.

[0016] Beneficial Effects

[0017] (1) By automatically identifying the pothole image on the conveyor belt through the deep learning algorithm, the pothole features can be efficiently extracted from complex images, reducing manual intervention and improving the automation degree of detection.

[0018] (2) By calculating the clarity of each frame of the target pothole image, the quality of the pothole image can be judged to ensure that the clearest image is selected as the final target image. The clarity is not only related to the difference and chaos degree of the pixels in the pothole, but also can effectively distinguish relatively blurred or low-quality images, avoiding misinterpretation of the images. By selecting the image with the highest clarity as the target image, it can ensure that the subsequent wear coefficient calculation is based on high-quality data, improving the accuracy of the detection results.

[0019] (3) Through the calculation of the wear coefficient, the wear of the conveyor belt can be quantitatively analyzed and a safety judgment can be made according to the set threshold. This makes the wear detection not rely on subjective evaluation, but judge whether to alarm through clear criteria, thus enhancing the reliability of decision-making. Description of the Drawings

[0020] Figure 1It is a flowchart schematically showing the calculation of the wear coefficient of a conveyor belt according to an embodiment of the present invention. Detailed implementation manners

[0021] The following will describe in detail the specific implementation manners of the present invention in conjunction with the accompanying drawings.

[0022] As Figure 1 shown, S101: Obtain a pitted image of the conveyor belt.

[0023] Specifically, the surface of the conveyor belt is photographed by an image acquisition device to obtain an image of the conveyor belt, and a semantic segmentation network (CNN) is used to identify pits in the image and extract the pitted image. Among them, the image acquisition device is a depth camera, so that each pixel point in the image contains depth information and grayscale value information, which is convenient for subsequent wear detection of the conveyor belt.

[0024] In one embodiment, a semantic segmentation network (CNN) is used to identify pits in the image and extract the pitted image, including: training a pit recognition network, inputting the image of the conveyor belt into the trained pit recognition network to obtain the pitted image. Among them, the process of training the pit recognition network is: marking the pit pixel points in the image of the conveyor belt as 1 and the non-pit pixel points as 0 to obtain a 0-1 labeled image; inputting the 0-1 labeled image into the semantic segmentation network for training; completing the training when the loss function is less than a preset value or reaches a preset number of training times to obtain the pit recognition network. Then, the image of the conveyor belt is input into the pit recognition network to obtain a 0-1 labeled image, and the 0-1 labeled image matrix is multiplied by the image matrix of the conveyor belt to obtain the pitted image. Thus, the pitted image corresponding to each frame of the conveyor belt image can be obtained.

[0025] Training the pitted image through a deep learning network can continuously optimize the algorithm and improve the adaptability of the system in different environments. In different working environments and on different types of conveyor belts, the system can automatically adjust the recognition and detection strategies according to the training data and real-time images, maintain a high detection accuracy, reduce manual operations, and reduce human omissions and errors, thereby improving the reliability of the overall system.

[0026] In one embodiment, after obtaining the pitted images corresponding to each frame of the conveyor belt image, a connected component analysis method is used to obtain each connected component in the pitted image, where one connected component represents one pit. Thus, each pit in the pitted image can be obtained.

[0027] It should be noted that since the conveyor belt runs at a constant speed during actual operation, that is, the surface images of the conveyor belt at fixed intervals are collected at equal time intervals, and then continuous-frame pit images can be obtained. However, during the process of the conveyor belt running at a constant speed, some of the collected images are not clear. Therefore, it is necessary to screen the continuous-frame pit images to select the pit images with the highest clarity, ensure that the subsequent wear coefficient calculation is based on high-quality data, and thus improve the accuracy of the analysis of the wear degree of the conveyor belt.

[0028] S102: Obtain the position of any pit in the continuous-frame pit images.

[0029] Specifically, taking any pit in the current-frame pit image as the target pit, calculate the matching index between the target pit and any pit in the next-frame pit image, and determine the target pit in the continuous-frame pit images according to the magnitude of the matching index. Among them, the matching index between two pits is positively correlated with the similarity between the surrounding pits of the two pits and is also positively correlated with the similarity between the two pits. Specifically, for all pits at the edge in the next-frame pit image, obtain the matching index of all pits. When the matching index reaches the maximum value, it indicates that this pit and the target pit are the same pit. Thus, the position of the target pit in the current-frame pit image in the next-frame pit image can be obtained, and then the position of the target pit in the continuous-frame pit images can be determined.

[0030] In one embodiment, calculate the matching index between the target pit and any pit in the next-frame pit image, and the calculation formula is: ; In the formula, is the matching index between the target pit and the th pit in the next-frame pit image, is the pixel value matrix of the target pit, is the pixel value matrix of the th pit in the next frame, is the pixel value matrix of the th pit in the next frame, is the position vector composed of the geometric centers of the th pit around the target pit pointing to the geometric center of the target pit, is the position vector composed of the geometric centers of the th pit around the th pit in the next-frame pit image pointing to the geometric center of the th pit, and are both similarity calculations.

[0031] Among them, is the similarity degree between the k-th pothole in the next-frame pothole image and the internal pixel value matrix of the target pothole. The closer this value is to 1, the more similar the two potholes are. is the similarity degree between the four potholes closest to the k-th pothole in the target pothole and the next-frame pothole image respectively and the four potholes closest to the target pothole in the current-frame pothole image, including the similarity comparison of positions and the internal pixel value matrices in the potholes. The larger this value is, the higher the similarity of the pothole distribution around the target pothole and the k-th pothole in the next-frame pothole image, and the more likely the target pothole and the k-th pothole in the next-frame pothole image are the same pothole.

[0032] By matching the potholes through the above method, the position of the target pothole in the adjacent-frame pothole images can be accurately found, and the mis-matching of the target pothole caused by similar potholes can be effectively avoided through the positions and internal pixel value distributions of the adjacent potholes, ensuring the accuracy of calculating the clarity degree of the subsequent potholes.

[0033] It should be noted that the user can appropriately increase or decrease the number of calculated adjacent potholes according to the actual application scenario. In principle, the smaller the coal material, the more adjacent potholes should be involved in the calculation. The reason is that the coal feeding device is located above the conveyor belt, and the coal can fall onto the conveyor belt after passing through the coal feeding device and impact the conveyor belt. Also, the smaller the coal material, the relatively more impact times the conveyor belt will receive, and the impact forces are also more similar, resulting in a generally consistent pothole size. Therefore, more pothole distribution features are needed to improve the matching accuracy.

[0034] S103: Calculate the clarity degree of each target pothole in the consecutive-frame pothole images and select the target image.

[0035] Specifically, calculate the clarity degree of each target pothole in the consecutive-frame pothole images, and the consecutive-frame pothole images all contain complete target potholes, and select the pothole image corresponding to the highest clarity degree of the pothole as the target image, where the clarity degree of the pothole reflects the degree of chaos of the pixel values in the pothole and the size of the difference in the pixel values in the pothole.

[0036] In one embodiment, calculate the clarity degree of each target pothole in the consecutive-frame pothole images, and the calculation formula is: ; in the formula, is the clarity degree of the target pothole in the -th frame pothole image, is the proportion of the pixel value in the target pothole in the -th frame pothole image, is the pixel value sequence of the target pothole in the -th frame pothole image.

[0037] Among them, is the degree of chaos of the target pothole pixel values in any frame of pothole images. The larger this value is, the more pixel value information is contained in the target pothole in any frame of pothole images; The larger it is, the greater the difference in pixel value levels contained in the target pothole in the said any frame of pothole images, and the clearer the pothole is. Thus, the clarity of the target pothole in each consecutive frame of pothole images can be obtained, and the pothole image corresponding to the highest clarity of the pothole is selected as the target image. By obtaining the clearest pothole image corresponding to the target pothole during a continuous shooting process, the accuracy of calculating the wear coefficient of the pothole in the subsequent process is improved.

[0038] S104: Calculate the wear evaluation of the target pothole.

[0039] Specifically, calculate the wear coefficient of the target pothole in the target image, and perform wear detection on the conveyor belt according to the magnitude of the wear coefficient. Among them, the wear coefficient of the target pothole is positively correlated with the area of the target pothole, and is also positively correlated with the ratio of the minimum pixel value of the target pothole to the average pixel value of the normal area.

[0040] In one embodiment, the calculation formula for the wear coefficient of the target pothole in the target image is: ; In the formula, is the wear coefficient of the target pothole in the target image, is the number of pixel points of the target pothole in the target image, is the pixel value sequence of the target pothole in the target image, represents the minimum value of the pixel points in the pixel value sequence of the target pothole in the target image, is the average value of the pixel values of the non-pothole area in the target image, is the hyperbolic tangent function. Among them, is the comparison between the minimum value of the pixel value sequence of the clearest target pothole and the average value of the pixel values of the non-pothole area. The larger this value is, the more serious the wear condition of the clearest target pothole is; is the expression of the pothole size. The larger this value is, the larger the pothole area. It should be noted that when calculating the wear coefficient, multiple factors such as the area of the target pothole and pixel value differences are considered, which can comprehensively reflect the wear condition of the conveyor belt.

[0041] In another embodiment, another method for calculating the wear coefficient of the target pothole in the target image is also provided. The specific steps are as follows: Subtract the mean value of the pixel values of the non-pothole area from the pixel values of the target pothole in the target image to obtain a pothole difference map, and perform k-means clustering on the pothole difference map. Among them, set the initial number of clustering clusters k to 1, obtain the silhouette coefficient value using the silhouette coefficient method, and get a silhouette coefficient value each time one more clustering cluster is added. Set the silhouette coefficient threshold to 0.5. When the silhouette coefficient value is greater than the silhouette coefficient threshold, the optimal number of clustering clusters is obtained at this time. It should be noted that the more the number of clustering clusters, the worse the quality of the pothole. Because one clustering cluster means that the clustering center has received greater wear. Calculate the wear coefficient of the target pothole according to the above method, and the calculation formula is: . Wherein, is the wear coefficient of the target pothole, is the pixel value sequence composed of the pixel points of the clustering centers of all clustering clusters, is the mean value of the pixel values of the non-pothole area in the target image, is the number of clustering centers of all clustering clusters, is the hyperbolic tangent function. The wear degree of the worst position point in the target pothole can be found as the overall wear degree of the target pothole, and at the same time, the number of wear points in the target pothole is taken into account as the overall quality of the pothole.

[0042] S105: Perform wear detection on the conveyor belt according to the magnitude of the wear coefficient.

[0043] In one embodiment, if the wear coefficient is greater than the wear evaluation threshold, it indicates that the wear at the target pothole position does not meet the usage requirements, then an alarm prompt is given to remind the relevant staff that repair or replacement is needed; if the wear coefficient is less than or equal to the wear evaluation threshold, no alarm is made and the conveyor belt can be continued to be used. In this embodiment, the empirical value of the wear evaluation threshold is 0.8. In other embodiments, the empirical value of the wear evaluation threshold can be 0.85 or 0.78, etc., which can be adjusted according to the specific implementation situation. By setting the wear evaluation threshold, when the wear coefficient is greater than the wear evaluation threshold, an alarm prompt can be triggered, potential safety hazards can be discovered in time, and the risk of accidents can be reduced.

[0044] Through the above steps, combining image processing technology and deep learning network, the pothole features on the conveyor belt are accurately extracted using the pothole recognition network. By analyzing the pothole image and calculating parameters such as the matching index and clarity of the pothole, the detection accuracy of the wear condition of the conveyor belt can be effectively improved. Generally speaking, this method can efficiently, accurately and real-time detect the wear condition of the coal conveyor belt, improve safety, and reduce manual intervention.

[0045] The above-described embodiments merely represent several implementation manners of the present invention. The description thereof is relatively specific and detailed, but it should not be construed as a limitation to the scope of the patent application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all fall within the protection scope of the present invention.

Claims

1. A belt conveyor belt detection method, characterized in that: include: Collecting an image of the conveyor belt, and obtaining a pothole image of the image using a pothole recognition network; According to any pothole in the pothole image of the current frame being the target pothole, a matching index between the target pothole and any pothole in the pothole image of the next frame is calculated to determine the target pothole in the pothole images of the consecutive frames, wherein the matching index of the two potholes is positively correlated with the similarity of the pothole distribution between the surrounding potholes of the two potholes and is also positively correlated with the similarity between the two potholes; Calculating the clarity of each target pit in the continuous frame pit images, wherein the continuous frame pit images all contain complete target pits, and selecting the pit image corresponding to the highest pit clarity as the target image, wherein the pit clarity reflects the degree of confusion of pixel values ​​in the pit and the size of the difference of pixel values ​​in the pit; Calculate the wear coefficient of the target pothole in the target image: The pixel value of the target pothole in the target image is obtained by subtracting the pixel value of the non-pothole area to obtain the pothole difference map, and the pothole difference map is clustered by k-means. The number of clusters is determined by the silhouette coefficient method, and the wear coefficient of the target pothole is calculated based on the number of clusters. The calculation formula is: ； In the formula, is the wear coefficient of the target pothole, is the pixel value sequence composed of the center pixels of all clusters. is the mean pixel value of the non-hole area in the target image, is the number of cluster centers of all clusters, is the hyperbolic tangent function; The wear of the conveyor belt is detected according to the size of the wear coefficient, wherein the wear coefficient of the target pothole is positively correlated with the area of ​​the target pothole, and is also positively correlated with the ratio of the minimum pixel value of the target pothole to the average pixel value of the normal area.

2. The belt conveyor belt detection method according to claim 1, characterized in that: According to the size of the wear coefficient, the conveyor belt is subjected to wear detection, including: In response to the wear coefficient being greater than a wear evaluation threshold, an alarm is issued; In response to the wear coefficient being less than or equal to the wear evaluation threshold, no alarm is issued.

3. The belt conveyor belt detection method according to claim 1, characterized in that: The calculation formula of the wear coefficient of the target pothole in the target image also includes: ； In the formula, is the wear coefficient of the target pothole in the target image, is the number of pixels of the target holes in the target image, is the pixel value sequence of the target hole in the target image, is the mean pixel value of the non-hole area in the target image, is the hyperbolic tangent function.

4. The belt conveyor belt detection method according to claim 1, characterized in that: Calculate the clarity of each target pit in the continuous frame pit image, the calculation formula is: ； In the formula, For the The clarity of the target pit in the frame pit image, For the The pixel value of the target hole in the frame hole image is The proportion of For the The pixel value sequence of the target pothole in the frame pothole image.

5. The belt conveyor belt detection method according to claim 1, characterized in that: Calculate the matching index between the target pothole and any pothole in the next pothole image. The calculation formula is: ； In the formula, The target pothole and the next frame of the pothole image The matching index of the holes, is the pixel value matrix of the target pothole, The target pothole is surrounded by The position vector is composed of the geometric center of each pit pointing to the geometric center of the target pit. In the next frame of the pothole image, Around the pothole The geometric center of the pit points to The position vector composed of the geometric centers of the pits, For the next frame The pixel value matrix of the pits, For the next frame The pixel value matrix of the pits, Both are similarity calculations.

6. The belt conveyor belt detection method according to claim 1, characterized in that: The training process of the pothole recognition network includes: The pothole pixels in the conveyor belt image are marked as 1, and the non-pothole pixels are marked as 0, to obtain a 0-1 marked image; Input the 0-1 labeled image into a semantic segmentation network for training; In response to the loss function being less than a preset value or reaching a preset number of training times, the training is completed to obtain a pothole recognition network.

7. The belt conveyor belt detection method according to claim 1, characterized in that: Obtaining a pothole image of the image using a pothole recognition network includes: Inputting the image of the conveyor belt into a pothole recognition network to obtain a 0-1 labeled image; The pothole image is obtained by multiplying the 0-1 labeled image matrix with the image matrix of the conveyor belt.

8. The belt conveyor belt detection method according to claim 1, characterized in that: The image of the conveyor belt is captured by a depth camera, wherein the pixels in the image contain grayscale value information and depth information.

9. The belt conveyor belt detection method according to claim 1, characterized in that: Get each pothole in the pothole image, including: The connected domain analysis method is used to obtain the connected domains in the pothole image, where one connected domain is a pothole.

Images