Belt conveyor longitudinal tearing detection system and method based on computer vision

Abstract

According to the belt conveyor longitudinal tearing detection method based on computer vision, reflective strips are arranged on the non-bearing surface of a conveying belt of a belt conveyor at equal intervals, and a light source, a ccd camera and a photodiode are arranged on a rack. The photodiodes are used for detecting light reflection of the reflective strips to trigger the camera to shoot the non-bearing surface of the conveying belt, numbers and other marks arranged on the conveying belt are used for distinguishing unavailable areas, photos of all the shot areas are compared with initial photos, whether changes exist or not is judged, and the detection accuracy is improved. And if yes, performing artificial intelligence identification on the state of the change area by using a depth vision technology. And a traditional picture comparison technology is combined with a computer vision technology, so that the detection efficiency is improved.

Description

Technical Field

[0001] The present invention relates to a technology for detecting longitudinal tearing of a belt conveyor, and in particular to a method for detecting longitudinal tearing of a belt conveyor based on computer vision and a detection device thereof. Background Art

[0002] Belt conveyors are essential equipment for material transportation in industries such as mining, ports, and metallurgy. Over long periods of operation, conveyor belts can tear longitudinally due to material impact, equipment wear, and other factors. Traditional inspection methods rely primarily on manual inspections, which are inefficient and prone to missed inspections. Therefore, developing an efficient and automated inspection method is of great practical significance. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for detecting longitudinal tears in belt conveyors based on computer vision, thereby improving the efficiency and accuracy of detecting longitudinal tears in belt conveyors through computer vision technology. To achieve the above purpose, the present invention adopts the following technical solutions:

[0004] A method for detecting longitudinal tearing of a belt conveyor based on computer vision comprises the following steps:

[0005] Step 1: Reflective strips 1 are arranged at equal intervals on the non-load-bearing surface of the conveyor belt to divide the non-load-bearing surface into multiple areas;

[0006] Step 2: A light source 2, a CCD camera 3, a photodiode 4, and a roller brush 7 are arranged on the frame of the belt conveyor. The field of view of the CCD camera 3 can completely cover the non-load-bearing surface between two adjacent reflective strips 1 to capture a complete image of an area of ​​the non-load-bearing surface. The light emitted by the light source 2 is reflected by the reflective strips 1 and then received by the photodiode 4. During the movement, the non-load-bearing surface is first cleaned by the roller brush 7 and then illuminated by the planar light spot projected by the light source 2.

[0007] Step 3: Use the photodiode 4 to detect the reflection of the reflective strip 1 to trigger the CCD camera to photograph the non-load-bearing surface of the conveyor belt. The image of the non-load-bearing surface of the conveyor belt is photographed by the CCD camera 3 and transmitted to the on-site computer 5 for processing;

[0008] Step 4: Initial image acquisition and storage: When the system is first run, the initial image of each area on the non-load-bearing surface of the conveyor belt is collected as a reference for subsequent comparisons;

[0009] Step 5: During the subsequent operation of the belt conveyor, image acquisition and real-time image comparison are performed. The on-site computer 5 uses an image comparison algorithm to compare the real-time acquired regional image with the initial image to detect whether there are obvious changes. If there are obvious changes, the process proceeds to step 6; otherwise, the process continues to step 5.

[0010] Step 6: Anomaly identification is performed on the image of the area with obvious changes. The on-site computer 5 transmits the image of the area with obvious changes to the server 6. The server 6 identifies the image of the area with obvious changes and determines whether it is a longitudinal tear fault. If a longitudinal tear fault is found, the process proceeds to step 7; otherwise, the process proceeds to step 5.

[0011] Step 7: Longitudinal tear alarm. When the longitudinal tear fault of the belt conveyor is identified, the server 6 sends an alarm signal and records the fault area and time so that maintenance personnel can handle it in time.

[0012] Furthermore, in the method for detecting longitudinal tearing of a belt conveyor based on computer vision according to the present invention, step 5 preferably includes the following sub-steps:

[0013] Sub-step 5.1: Before performing image comparison, pre-process the collected non-load-bearing surface image of the conveyor belt to improve the accuracy and efficiency of the comparison;

[0014] Sub-step 5.2: Compare the real-time acquired image with the stored initial image to detect whether there is any change.

[0015] Furthermore, in the method for detecting longitudinal tearing of a conveyor belt based on computer vision according to the present invention, the sub-step 5.1 preferably includes the following operations:

[0016] Grayscale conversion: Convert the color image output by the CCD camera into a grayscale image to simplify subsequent processing steps.

[0017] Furthermore, in the method for detecting longitudinal tearing of a belt conveyor based on computer vision according to the present invention, the sub-step 5.2 preferably includes the following operations:

[0018] Image alignment: Calculate the transformation parameters based on the reflection strip 1 in the image, and transform the image according to the calculated transformation parameters so that the two images being compared are aligned in space.

[0019] Calculate image difference: Calculate the difference between the two aligned images. Each pixel value in the difference image reflects the degree of change of the two image pixels at the corresponding position, which can directly show the difference between the two images.

[0020] Thresholding: Thresholding is performed on the difference image to extract the areas with significant changes. By setting the change threshold, the pixel values ​​in the difference image that are less than the change threshold are set to 0, and the pixel values ​​that are greater than the change threshold are set to 1.

[0021] Count different pixels: Count the number of pixels in the thresholded image with a median value of 1, that is, the number of different pixels. If this number exceeds the pre-set different pixel threshold, it is determined that the conveyor belt is in an abnormal state.

[0022] Furthermore, in the computer vision-based belt conveyor longitudinal tear detection method described in the present invention, step 6 preferably includes the following operations: sub-step 6.1 of preprocessing the image of the area with obvious changes, and sub-step 6.2 of detecting longitudinal cracks in the preprocessed image.

[0023] Furthermore, in the method for detecting longitudinal tearing of a conveyor belt based on computer vision according to the present invention, the sub-step 6.1 preferably includes the following operations:

[0024] Grayscale: convert the image output by CCD camera 3 into a grayscale image;

[0025] Noise removal: Use Gaussian filtering and median filtering techniques to remove noise from the image, enhance image clarity, and provide a clearer image for subsequent edge detection and recognition operations;

[0026] Edge detection: Use edge detection algorithms to extract edge information from images, including noise suppression, gradient calculation, non-maximum suppression, and dual threshold detection;

[0027] Binarization: Convert the image into a binary image through threshold segmentation.

[0028] Furthermore, in the method for detecting longitudinal tearing of a belt conveyor based on computer vision according to the present invention, the sub-step 6.2 preferably includes the following operations:

[0029] Grid division: Divide the binary image into an N*N grid, where the size of each grid unit is k*k pixels;

[0030] Node initialization: Calculate the grayscale mean of each grid cell. If the mean is greater than the preset threshold T, it is marked as a crack node with a value of 1, otherwise it is marked as a background node with a value of 0;

[0031] Node connectivity calculation: Calculate the connectivity of the nodes using the 8-connectivity rule. For each crack node, check its connectivity with adjacent nodes. If connected, mark it as a crack region.

[0032] Crack morphology analysis: calculate the length and width of the crack in the crack area;

[0033] Crack detection: Calculate the ratio of the crack length to the crack width. When the ratio of the crack length to the width is greater than a threshold value U and the crack length is greater than a threshold value V, it is determined that the belt conveyor has longitudinal tearing.

[0034] Furthermore, in the method for detecting longitudinal tearing of a conveyor belt based on computer vision according to the present invention, the node connectivity calculation preferably includes the following operations:

[0035] Initialize a label matrix to record the crack region to which each node belongs.

[0036] Traverse all nodes and assign a new label to the unlabeled crack nodes.

[0037] Recursively or iteratively check its neighboring nodes, and if the neighboring node is a crack node and is unlabeled, assign the same label.

[0038] Furthermore, in the computer vision-based belt conveyor longitudinal tear detection method of the present invention, the crack morphology analysis preferably includes the following detailed steps:

[0039] Scan the connected domain from top to bottom and record the position of the first white pixel as (x1, y1).

[0040] Scan the connected domain from bottom to top and record the position of the first white pixel as (x2, y2).

[0041] Scan the connected domain from left to right and record the position of the first white pixel as (x3, y3).

[0042] Scan the connected domain from right to left and record the position of the first white pixel as (x4, y4);

[0043] Xleft =min{x1, x2, x3, x4},

[0044] Xright =max{x1, x2, x, x},

[0045] Yup = min{y1, y 2, y 3, y 4},

[0046] Ydown =max{y1, y2, y3, y4};

[0047] The length of the crack = Ydown-Yup, and the width of the crack = Xright-Xleft.

[0048] Furthermore, in the method for detecting longitudinal tearing of a conveyor belt based on computer vision of the present invention, the sub-step 6.1 of pre-processing the image of the area with obvious changes by the server 6 preferably includes:

[0049] Resize the image to a fixed size to facilitate input into the deep learning model for processing;

[0050] Normalize the image and map the pixel value range to the [0, 1] interval to improve the training effect and convergence speed of the model;

[0051] Enhance the image by performing operations such as random rotation, flipping, cropping, and adding noise to increase data diversity and prevent model overfitting.

[0052] Furthermore, in the method for detecting longitudinal tearing of a conveyor belt based on computer vision according to the present invention, the sub-step 6.2 of performing longitudinal crack detection on the pre-processed image by the server 6 preferably includes:

[0053] A convolutional neural network (CNN) was selected as the basic architecture to build a deep learning model. Its output layer was configured with two output nodes to predict whether there was a longitudinal crack and the severity of the crack.

[0054] The pre-collected conveyor belt images are divided into training sets and test sets according to a certain ratio for model training and evaluation;

[0055] The training set is input into the deep learning model, and the output of the model is calculated through forward propagation. The loss function value is then calculated based on the difference between the output and the true label. The gradient of the loss function with respect to the model parameters is then calculated through the backpropagation algorithm, and the model parameters are updated using the optimization algorithm. This process is repeated until the accuracy and loss value of the model on the training set meet the preset requirements or the number of training rounds is completed.

[0056] The images that need to be detected for longitudinal tearing are processed in the same preprocessing method as the training set, and then input into the trained deep learning model to output the prediction results of whether there is a longitudinal tear and the prediction results of the tear severity.

[0057] Furthermore, the computer vision-based belt conveyor longitudinal tear detection method described in the present invention preferably sets an identifier on the non-load-bearing surface of the belt near each reflective strip, and divides the conveyor belt into several independent detection areas according to the identified identifier.

[0058] At the same time, the present invention also provides a detection device for the above-mentioned belt conveyor longitudinal tear detection method, comprising: a reflective strip 1, a light source 2, a CCD camera 3, a photodiode 4, an on-site computer 5, a server 6, and a roller brush 7;

[0059] On the non-load-bearing surface of the conveyor belt of the belt conveyor, a reflective strip 1 is provided at intervals of L, wherein the distance L is the same as the width of the conveyor belt;

[0060] The light source 2 is a high-brightness LED lamp that can project a planar light spot and illuminate the non-load-bearing surface of the conveyor belt of the belt conveyor. The planar light spot can evenly illuminate the non-load-bearing surface between two adjacent reflective strips 1.

[0061] The CCD camera 3 is an industrial camera with high resolution. The installation position of the CCD camera 3 is adjusted so that the field of view of the CCD camera 3 can completely cover the non-bearing surface between two adjacent reflective strips 1.

[0062] The light source 2 and the photodiode 4 are respectively installed on the left and right sides of the CCD camera 3, so that the light emitted by the light source 2 can be reflected by the reflective strip 1 and then received by the photodiode 4;

[0063] The roller brush 7 is arranged on one side of the light source 2, so that the non-load-bearing surface of the conveyor belt is first cleaned by the roller brush 7 during movement and then illuminated by the planar light spot projected by the light source 2, so as to prevent dust and the like from adhering to the non-load-bearing surface of the conveyor belt and interfering with the subsequent longitudinal tear detection;

[0064] The on-site computer 5 is electrically connected to the light source 2, the CCD camera 3, and the photodiode 4. The on-site computer 5 is capable of controlling the on and off of the light source 2 through software program settings, triggering the CCD camera 3 to capture and receive images output by the CCD camera 3;

[0065] The on-site computer 5 is connected to the server 6 via a network.

[0066] Furthermore, in the detection device described in the present invention, the light emitted by the light source 2 is preferably diffusely reflected on the non-load-bearing surface of the conveyor belt and mirror-reflected on the reflective strip 1, so that when the reflective strip 1 passes near the photodiode 4, the electrical signal output by the photodiode 4 increases. The on-site computer 5 performs rising edge detection on the received electrical signal output by the photodiode 4 to obtain a trigger signal, and controls the CCD camera 3 to take pictures according to the trigger signal.

[0067] The beneficial effects of the present invention are:

[0068] 1. In the present invention, the light emitted by the light source 2 is diffusely reflected on the non-load-bearing surface of the conveyor belt and then specularly reflected on the reflective strip 1. As a result, when the reflective strip 1 passes near the photodiode 4, the amount of light received by the photodiode 4 increases, thereby increasing the electrical signal output by the photodiode 4. The on-site computer 5 detects the rising edge of the received electrical signal output by the photodiode 4 to obtain a trigger signal, and controls the CCD camera 3 to capture images based on the trigger signal.

[0069] 2. The present invention can quickly identify whether there are changes in the surface of the conveyor belt by performing real-time image comparison on the on-site computer 5, thereby detecting possible tears or other abnormal conditions; by using the server 6 to further identify the changed area, the rich computing power resources on the server 6 can be used to use complex computer vision algorithms for offline analysis to determine whether it is a longitudinal tear fault; through the above method, efficient and automated detection of longitudinal tears in the conveyor belt of the belt conveyor can be achieved, which not only improves the accuracy and efficiency of the detection, but also allows one server 6 to host multiple on-site computers 5 and simultaneously detect multiple belt conveyors.

[0070] 3. The present invention sets identifiers on the conveyor belt to distinguish different areas. Each identifier corresponds to a specific area. The identifier in the automatic recognition image can determine the specific area corresponding to the current image. By setting identifiers in each area, the system can perform image comparison and anomaly detection more efficiently, which not only improves the accuracy and speed of detection, but also facilitates fault location and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Attachment Figure 1 This is a schematic diagram of the structure of a belt conveyor longitudinal tear detection device based on computer vision;

[0072] Attachment Figure 2 This is a flow chart of the belt conveyor longitudinal tear detection method based on computer vision. DETAILED DESCRIPTION

[0073] The present invention will now be described in further detail with reference to the accompanying drawings and preferred embodiments. These drawings are simplified schematic diagrams that illustrate the basic structure of the present invention only in a schematic manner, and therefore only show components relevant to the present invention. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present invention, and are not to be construed as limiting the present invention.

[0074] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "setting" and "connection" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or mutual cooperation. It can be a mechanical connection or an electrical connection. It can be directly connected or indirectly connected through an intermediate medium, or it can be internal communication between two components. In addition, in this article, "coupling" of two devices means that the laser beam emitted by one of the devices is incident on the other device. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0075] Example 1:

[0076] As attached Figure 1 As shown, a computer vision-based device for detecting longitudinal tears in a belt conveyor is provided. The belt conveyor system is equipped with reflective strips 1, a light source 2, a CCD camera 3, a photodiode 4, an on-site computer 5, a server 6, and a roller brush 7. Installation and commissioning are performed according to the method of the present invention. After installation, the system is started for testing. When the photodiode detects reflections from the reflective strips, it triggers the CCD camera to capture the image. The system compares the captured image with the initial image and, if significant changes are detected in the image, further analysis is performed on the changed areas.

[0077] Select a reflective strip 1 of appropriate width and length and place a reflective strip 1 on the non-load-bearing surface of the conveyor belt of the conveyor at intervals of L, where the distance L is the same as the width of the conveyor belt. The aspect ratio of the length to width of the reflective strip 1 is greater than 10, preferably greater than 20.

[0078] The light source 2 is a high-brightness LED lamp that can project a planar light spot and illuminate the non-load-bearing surface of the conveyor belt of the belt conveyor. The planar light spot can evenly illuminate the non-load-bearing surface between two adjacent reflective strips 1.

[0079] The CCD camera 3 is an industrial camera with high resolution. The installation position of the CCD camera 3 is adjusted so that the field of view of the CCD camera 3 can completely cover the non-bearing surface between two adjacent reflective strips 1 .

[0080] The light source 2 and the photodiode 4 are respectively installed on the left and right sides of the CCD camera 3 , so that the light emitted by the light source 2 can be reflected by the reflective strip 1 and then received by the photodiode 4 .

[0081] The roller brush 7 is arranged on one side of the light source 2, so that the non-load-bearing surface of the conveyor belt is first cleaned by the roller brush 7 during movement and then illuminated by the planar light spot projected by the light source 2, so as to prevent dust and the like from adhering to the non-load-bearing surface of the conveyor belt and interfering with subsequent longitudinal tear detection.

[0082] On-site computer 5 is electrically connected to light source 2, CCD camera 3, and photodiode 4. Software is used to program on-site computer 5 to control the on-off state of light source 2, trigger CCD camera 3 to capture images, and receive images output by CCD camera 3. On-site computer 5 is connected to server 6 via a network.

[0083] Since the light emitted by the light source 2 is diffusely reflected on the non-load-bearing surface of the conveyor belt and is mirror-reflected on the reflective strip 1, the light received by the photodiode 4 when the reflective strip 1 passes near the photodiode 4 increases, thereby increasing the electrical signal output by the photodiode 4. The on-site computer 5 detects the rising edge of the received electrical signal output by the photodiode 4 to obtain a trigger signal, and controls the CCD camera 3 to take pictures according to the trigger signal.

[0084] As attached Figure 2 As shown in FIG, the belt conveyor longitudinal tear detection method based on computer vision includes the following steps:

[0085] Step 1: Reflective strips 1 are arranged at equal intervals on the non-bearing surface of the conveyor belt to divide the non-bearing surface into multiple areas.

[0086] The reflective strips 1 are preferably made of a highly reflective material, such as reflective tape or paint. The material should be wear-resistant and weather-resistant to suit the conveyor belt's operating environment. The reflective strips 1 are installed at even intervals on the non-load-bearing surface of the conveyor belt. Reflective strips 1 are placed at regular intervals to ensure that every area is accurately detected. The reflective strips 1 are securely affixed to the non-load-bearing surface of the conveyor belt using strong glue or other fixing devices to prevent them from falling off during operation.

[0087] Step 2: Arrange the light source 2, CCD camera 3, photodiode 4, and roller brush 7 on the frame.

[0088] Light source 2 preferably utilizes a high-brightness LED lamp due to its high brightness, low energy consumption, and long lifespan. Mounted above the conveyor frame, the light source projects a planar light spot onto the non-load-bearing surface of the conveyor belt. This planar light spot evenly illuminates the non-load-bearing surface between two adjacent reflective strips 1. To ensure uniform illumination of the reflective strips 1, the light source angle should be adjusted based on site conditions for optimal reflective effect. The light source's brightness and on / off control should be configured to accommodate varying ambient lighting conditions, ensuring the reflective strips are clearly visible in all lighting conditions.

[0089] CCD camera 3 is preferably a high-resolution industrial camera to ensure clear image detail. Mount CCD camera 3 on the frame, maintaining a certain distance from reflective strips 1. Adjust the camera's mounting position, determining its height and angle based on the conveyor belt's width and the required inspection accuracy, so that the camera's field of view fully covers the non-load-bearing surface between two adjacent reflective strips 1, ensuring that a complete image of a region of the non-load-bearing surface is captured.

[0090] The photodiode 4 preferably has a fast response speed and high sensitivity to ensure accurate detection of the light reflected by the reflective strip. The light source 2 and the photodiode 4 are respectively mounted on the left and right sides of the CCD camera 3 so that the light emitted by the light source 2 is reflected by the reflective strip 1 and can be received by the photodiode 4.

[0091] The roller brush 7 is arranged on one side of the light source 2, so that the non-load-bearing surface of the conveyor belt is first cleaned by the roller brush 7 during movement and then illuminated by the planar light spot projected by the light source 2, so as to prevent dust and the like from adhering to the non-load-bearing surface of the conveyor belt and interfering with subsequent longitudinal tear detection.

[0092] Step 3: Use the photodiode 4 to detect the reflection of the reflective strip 1 to trigger the CCD camera to photograph the non-load-bearing surface of the conveyor belt.

[0093] When reflective strips 1 pass near photodiode 4, the amount of light received by photodiode 4 increases, causing the electrical signal output by photodiode 4 to increase. On-site computer 5 detects the rising edge of the received electrical signal output by photodiode 4 to generate a trigger signal. Based on this trigger signal, CCD camera 3 is controlled to capture an image. CCD camera 3 captures an image of the non-load-bearing surface of the conveyor belt and transmits it to on-site computer 5 for processing. Triggered by the output signal from the photodiode, CCD camera 3 captures an image that covers the entire area between two adjacent reflective strips 1.

[0094] Step 4: Initial image acquisition and storage:

[0095] When the system is first run, an initial image of each area on the non-load-bearing surface of the conveyor belt is collected as a reference for subsequent comparisons.

[0096] Step 5: During the operation of the belt conveyor, perform image acquisition and real-time image comparison.

[0097] When the reflective strip 1 passes near the photodiode 4, the electrical signal output by the photodiode 4 increases. The on-site computer 5 detects the rising edge of the received electrical signal output by the photodiode 4 to obtain a trigger signal, and controls the CCD camera 3 to take pictures according to the trigger signal.

[0098] The on-site computer 5 uses an image comparison algorithm to compare the real-time collected image with the initial image to detect whether there are obvious changes, such as tears, damage, etc. If there are obvious changes, then go to step 6, otherwise continue to step 5.

[0099] Step 6: Identify anomalies in images of areas with significant changes.

[0100] If an abnormal change is detected, the on-site computer 5 transmits the image of the area with obvious changes to the server 6. The server 6 uses the changed area to further identify whether it is a longitudinal tear fault. If a longitudinal tear fault is found, the process proceeds to step 7, otherwise it proceeds to step 5.

[0101] Step 7: Longitudinal tear alarm.

[0102] When a longitudinal tear fault is identified in the belt conveyor, the server 6 sends an alarm signal and records the fault area and time so that maintenance personnel can handle it in time.

[0103] By performing real-time image comparison on the on-site computer 5 , it is possible to quickly identify whether changes have occurred on the conveyor belt surface, thereby detecting possible tears or other abnormalities.

[0104] By using server 6 to further identify the changed area, the abundant computing resources on server 6 can be used to perform offline analysis using complex computer vision algorithms to determine whether it is a longitudinal tear fault.

[0105] The above method can realize efficient and automated detection of longitudinal tearing of conveyor belts of conveyor belts, which not only improves the accuracy and efficiency of detection, but also allows one server 6 to host multiple on-site computers 5 and detect multiple conveyor belts at the same time.

[0106] Example 2:

[0107] Based on the solution of Example 1, step 5 further includes the following detailed sub-steps:

[0108] Sub-step 5.1: Before performing image comparison, pre-process the collected image of the non-load-bearing surface of the conveyor belt to improve the accuracy and efficiency of the comparison. The pre-processing preferably includes the following operations:

[0109] Grayscale conversion: Converting the color image output by the CCD camera to a grayscale image simplifies subsequent processing steps. This reduces computational effort and highlights structural features within the image. Common weighted averaging methods can be used. For example, based on the human eye's sensitivity to different colors, the red, green, and blue components are multiplied by different weights (e.g., 0.299, 0.587, and 0.114), and then the grayscale value is added together.

[0110] Geometric correction: In actual conveyor belt usage scenarios, vibrations during operation and other factors may cause geometric distortion in the captured images. Geometric correction is achieved through image registration technology, ensuring that images captured at different times maintain consistent geometry.

[0111] Sub-step 5.2: Comparing the real-time acquired image with the stored initial image to detect whether there is a change, preferably including the following operations:

[0112] Image alignment: In real-time, the captured real-time image and the initial image may differ in position, angle, and other aspects. Image alignment allows them to be aligned to the same coordinate system. This process requires first determining a geometric transformation model, such as an affine transformation or perspective transformation. The transformation parameters are then calculated based on the reflection strips 1 in the image. Finally, the real-time image is transformed according to the calculated transformation parameters, ensuring that the two images being compared are spatially aligned.

[0113] Calculating image differences: After image alignment, the difference between the two aligned images is calculated. Each pixel value in the difference image reflects the degree of change between the two image pixels at the corresponding location. For example, if a pixel value in the difference image is 0, it means that the pixel value at that location in the live image and the original image is the same; if the pixel value is not 0, it indicates that there is a change. By calculating image differences, the difference between the two images can be directly displayed.

[0114] Thresholding: To ignore minor changes, the differential image is thresholded to extract areas of significant change. In actual conveyor belt inspections, slight environmental fluctuations or errors in the image acquisition process may cause minor changes in the differential image. These minor changes may not represent an actual abnormal condition of the conveyor belt. By setting an appropriate threshold, pixel values ​​in the differential image below the threshold are set to 0, and pixel values ​​above the threshold are set to 1, resulting in a binary image. Regions with a value of 1 in this binary image represent areas of significant change. The choice of threshold depends on the specific application and noise level and needs to be adjusted based on actual conditions.

[0115] Counting Distinct Pixels: This counts the number of pixels in the thresholded binary image that have a value of 1, representing the number of distinct pixels. By counting the number of distinct pixels, the degree of difference between the live image and the initial image can be quantified. If this number exceeds a pre-set threshold, it indicates an abnormality in the conveyor belt, such as a tear, wear, or foreign matter.

[0116] As a preferred embodiment, the threshold is selected to be 5% of the total number of pixels, that is, if the number of different pixels exceeds 10% of the total number of pixels, it is determined that the conveyor belt is in an abnormal state.

[0117] Furthermore, the preferred threshold is 10% or 15% of the total number of pixels.

[0118] Example 3:

[0119] Based on the solution of the aforementioned embodiment, step 6 further includes the following detailed sub-steps: the server 6 pre-processes the images of the area with obvious changes and performs longitudinal crack detection to improve the accuracy and efficiency of longitudinal tear detection.

[0120] The server 6 performs a sub-step 6.1 of pre-processing the image of the region with obvious changes, preferably including the following operations:

[0121] Grayscale: Convert the color image output by the CCD camera into a grayscale image.

[0122] Noise Removal: During image acquisition, noise often occurs due to environmental factors such as fine dust. This noise can interfere with accurate image analysis. Filtering techniques, such as Gaussian filtering and median filtering, are used to remove noise from images. Gaussian filtering smoothes the image by taking a weighted average of the values ​​of each pixel and its neighboring pixels, reducing the impact of noise. Median filtering sorts the pixel values ​​within a pixel's neighborhood and takes the median value as the new value for that pixel. Noise removal enhances image clarity, providing a clearer image foundation for subsequent edge detection and recognition.

[0123] Edge detection: An edge detection algorithm is used to extract edge information from an image. The Sobel algorithm detects edges by calculating the gradients of the image in the horizontal and vertical directions. It performs convolution operations on the image in the horizontal and vertical directions respectively, and then determines the edge pixels based on a certain threshold. The Canny algorithm is a more complex and accurate edge detection algorithm, which includes multiple steps such as noise suppression, gradient calculation, non-maximum suppression, and dual threshold detection.

[0124] Binarization: Convert the image into a binary image (crack area is white and background is black) through threshold segmentation.

[0125] Sub-step 6.2 of the server 6 performing longitudinal crack detection on the pre-processed image preferably includes the following operations:

[0126] Meshing: Divide the binary image into an N*N grid, where each grid cell is k*k pixels in size. The choice of the grid size (k*k) should be based on the typical width of the crack.

[0127] Node initialization: Calculate the grayscale mean of each grid cell. If the mean is greater than a preset threshold (T), it is marked as a crack node (value 1), otherwise it is marked as a background node (value 0). Set the percolation threshold (T) based on the connectivity of the crack, usually 0.5-0.8.

[0128] Node Connectivity Calculation: Node connectivity is calculated using the 8-connectivity rule to simulate the crack propagation path. Starting from the top of the image, the crack propagation process is simulated. For each crack node, its connectivity with adjacent nodes is checked. If connected, it is marked as a crack region.

[0129] Crack morphology analysis: calculate the length and width of the crack.

[0130] Crack detection: Calculate the ratio of the crack length to the crack width. When the ratio of the crack length to the width is greater than a threshold (U) and the crack length is greater than a threshold (V), it is determined that the belt conveyor has longitudinal tearing.

[0131] Example 4:

[0132] Based on the solution of the above embodiment, the node connectivity calculation in step 6 further includes the following detailed sub-steps:

[0133] Traverse each node and check its connectivity with adjacent nodes using the 8-connectivity rule. Adjacent nodes include: up, down, left, and right. Also, there are four diagonal directions: upper left, upper right, lower left, and lower right.

[0134] For the current node ( (i, j) ), check its eight adjacent nodes ( (i-1, j), (i+1, j), (i,j-1), (i, j+1), (i-1, j-1), (i-1, j+1), (i+1, j-1), (i+1, j+1) ). If the adjacent node is a crack node (value is 1), the current node is considered connected to the adjacent node.

[0135] Use the Region Growing algorithm or the Connected Component Labeling algorithm to label the nodes belonging to the same crack region. The specific steps are as follows:

[0136] Initialize a label matrix to record the crack region to which each node belongs.

[0137] Traverse all nodes and assign a new label to the unlabeled crack nodes.

[0138] Recursively or iteratively check its neighboring nodes, and if the neighboring node is a crack node and is unlabeled, assign the same label.

[0139] Example 5:

[0140] Based on the solution of the above embodiment, the crack morphology analysis in step 6 further includes the following detailed steps:

[0141] Perform a global scan on the connected domains obtained.

[0142] Scan the connected domain from top to bottom and record the position of the first white pixel as (x1, y1).

[0143] Scan the connected domain from bottom to top and record the position of the first white pixel as (x2, y2).

[0144] Scan the connected domain from left to right and record the position of the first white pixel as (x3, y3).

[0145] Scan the connected domain from right to left and record the position of the first white pixel as (x4, y4);

[0146] Xleft =min{x1, x2, x3, x4},

[0147] Xright =max{x1, x2, x, x},

[0148] Yup = min{y1, y 2, y 3, y 4},

[0149] Ydown =max{y1, y2, y3, y4};

[0150] Finally, the length of the crack = Ydown-Yup, and the width of the crack = Xright-Xleft.

[0151] Example 6:

[0152] Based on the solution of the aforementioned embodiment, step 6 further includes the following detailed sub-steps: before performing image comparison, the server 6 preprocesses the images of the area with obvious changes and uses a deep learning model to perform longitudinal tear detection to improve the accuracy and efficiency of longitudinal tear detection.

[0153] The server 6 performs a sub-step 6.1 of pre-processing the image of the region with obvious changes, preferably including the following operations:

[0154] Normalize the size of the collected images, preferably resizing all images to a fixed size to facilitate input into the deep learning model for processing;

[0155] Normalize the image and map the pixel value range to the [0, 1] interval to improve the training effect and convergence speed of the model;

[0156] Enhance the image by performing operations such as random rotation, flipping, cropping, and adding noise to increase data diversity and prevent model overfitting.

[0157] Sub-step 6.2 of the server 6 performing longitudinal crack detection on the pre-processed image preferably includes the following operations:

[0158] To build a deep learning model, we chose a convolutional neural network (CNN) as the basic architecture because CNN performs well in image classification and feature extraction. The network structure of a convolutional neural network (CNN) includes the following model structures:

[0159] Input layer: receives preprocessed image data.

[0160] Convolutional layer: This layer extracts features by sliding multiple convolution kernels across the image, capturing local texture and pattern information. For example, you can set up multiple convolutional layers, each using convolution kernels of different sizes, to extract features at different scales.

[0161] Pooling layer: Downsamples the output of the convolutional layer to reduce the spatial dimension of the data while retaining important feature information, reducing the amount of computation and the number of parameters. Common pooling operations include max pooling and average pooling.

[0162] Fully connected layer: Flattens the output of the pooling layer into a one-dimensional vector and performs a linear transformation to further extract global feature information.

[0163] Output layer: Based on the task requirements, two output nodes are set to predict the crack type (vertical) and crack severity (minor or severe). For crack type prediction, a binary cross-entropy loss function can be used; for crack severity prediction, a binary cross-entropy loss function or other appropriate loss function (such as the mean squared error loss function if the severity is quantized as a continuous value) can be used.

[0164] The pre-collected conveyor belt images are divided into training and test sets according to a certain ratio (such as 8:2) for model training and evaluation.

[0165] The training set is fed into the deep learning model. The model's output is calculated through forward propagation, and the loss function is then calculated based on the difference between the output and the true label. Next, the gradient of the loss function with respect to the model parameters is calculated through backpropagation, and the model parameters are updated using an optimization algorithm. This process is repeated until the model's accuracy and loss on the training set meet the preset requirements or the number of training rounds is reached.

[0166] Images requiring longitudinal tear detection are preprocessed using the same preprocessing methods as the training set and then fed into the trained deep learning model. The model extracts and analyzes features from the input image and outputs a prediction of whether a longitudinal tear is present and the tear severity (mild or severe).

[0167] Example 7:

[0168] Based on the scheme of the aforementioned embodiment, as a preferred scheme, an identifier is set on the non-load-bearing surface of the belt near each reflective strip, which can be a number, letter, QR code or other easily recognizable symbol. When selecting the identifier, its readability and durability should be considered, and the identifier should be made of wear-resistant and weather-resistant materials, such as wear-resistant paint, durable stickers or engraved marks. The size of the identifier should be large enough to be clearly visible in the image, but not affect the normal use of the belt. The specific size can be adjusted according to the resolution and installation distance of the camera. The identifier should be kept at a certain distance from the reflective strip to avoid reflection interference. The spacing of the identifier should match the spacing of the reflective strip so that the system can recognize all identifiers in each detection cycle.

[0169] Numerical identifiers, such as those placed on the conveyor belt, are used to distinguish different areas. Each identifier corresponds to a specific area. Image processing algorithms (such as OCR) automatically identify the identifiers in the image to determine the specific area corresponding to the current image. Based on the identified identifiers, the conveyor belt is divided into several independent inspection zones. By assigning identifiers to each zone, the system can more efficiently perform image comparison and anomaly detection. This approach not only improves detection accuracy and speed but also facilitates fault location and maintenance.

[0170] The above description only describes specific embodiments of the present invention. Various examples do not limit the essential content of the present invention. After reading the description, ordinary technicians in the relevant technical field can make modifications or variations to the specific embodiments described above without departing from the essence and scope of the present invention.

Claims

1. A method for detecting longitudinal tearing of a belt conveyor based on computer vision, characterized in that: The following steps are involved: Step 1: Reflective strips are arranged at equal intervals on the non-load-bearing surface of the conveyor belt to divide the non-load-bearing surface into multiple areas; Step 2: A light source, a CCD camera, a photodiode, and a roller brush are arranged on the frame of the belt conveyor. The field of view of the CCD camera can completely cover the non-load-bearing surface between two adjacent reflective strips 1 to capture a complete image of an area of ​​the non-load-bearing surface. The light emitted by the light source is reflected by the reflective strips and then received by the photodiode. During the movement, the non-load-bearing surface is first cleaned by the roller brush and then illuminated by the planar light spot projected by the light source 2. Step 3: Use a photodiode to detect the reflection of the reflective strip to trigger a CCD camera to photograph the non-load-bearing surface of the conveyor belt. The CCD camera photographs the image of the non-load-bearing surface of the conveyor belt and transmits it to an on-site computer for processing. Step 4: Initial image acquisition and storage: When the system is first run, the initial image of each area on the non-load-bearing surface of the conveyor belt is collected as a reference for subsequent comparisons; Step 5: During the subsequent operation of the belt conveyor, image acquisition and real-time image comparison are performed. The on-site computer uses the image comparison algorithm to compare the real-time acquired regional image with the initial image to detect whether there are obvious changes. If there are obvious changes, proceed to step 6, otherwise continue to step 5; Step 6: Perform abnormal identification on the image of the area with obvious changes. The on-site computer transmits the image of the area with obvious changes to the server. The server identifies the image of the area with obvious changes and determines whether it is a longitudinal tear fault. If a longitudinal tear fault is found, proceed to step 7; otherwise, proceed to step 5. Step 7: Longitudinal tear alarm. When the belt conveyor is identified to have a longitudinal tear fault, the server sends an alarm signal and records the fault area and time so that maintenance personnel can handle it in time.

2. The method for detecting longitudinal tearing of a belt conveyor based on computer vision according to claim 1, characterized in that: The step 5 includes the following sub-steps: Sub-step 5.1: Before performing image comparison, pre-process the collected non-load-bearing surface image of the conveyor belt to improve the accuracy and efficiency of the comparison; Sub-step 5.2: Compare the real-time acquired image with the stored initial image to detect whether there is any change.

3. The method for detecting longitudinal tearing of a belt conveyor based on computer vision according to claim 2, characterized in that: The sub-step 5.2 includes the following operations: Image alignment: Calculate transformation parameters based on the reflection strips in the image, and transform the image according to the calculated transformation parameters so that the two images being compared are aligned in space. Calculate image difference: Calculate the difference between the two aligned images. Each pixel value in the difference image reflects the degree of change of the two image pixels at the corresponding position, which can directly show the difference between the two images. Thresholding: Thresholding is performed on the difference image to extract the areas with significant changes. By setting the change threshold, the pixel values ​​in the difference image that are less than the change threshold are set to 0, and the pixel values ​​that are greater than the change threshold are set to 1. Count different pixels: Count the number of pixels in the thresholded image with a median value of 1, that is, the number of different pixels. If this number exceeds the pre-set different pixel threshold, it is determined that the conveyor belt is in an abnormal state.

4. The method for detecting longitudinal tearing of a belt conveyor based on computer vision according to claim 1, characterized in that: Said step 6 comprises the following operations: a sub-step 6.1 of pre-processing the image of the area with obvious changes, and a sub-step 6.2 of detecting longitudinal cracks on the pre-processed image; The sub-step 6.1 includes the following operations: Grayscale: Convert the image output by the CCD camera into a grayscale image; Noise removal: Use Gaussian filtering and median filtering techniques to remove noise from the image, enhance image clarity, and provide a clearer image for subsequent edge detection and recognition operations; Edge detection: Use edge detection algorithms to extract edge information from images, including noise suppression, gradient calculation, non-maximum suppression, and dual threshold detection; Binarization: Convert the image into a binary image through threshold segmentation.

5. The method for detecting longitudinal tearing of a belt conveyor based on computer vision according to claim 4, characterized in that: The sub-step 6.2 includes the following operations: Grid division: Divide the binary image into an N*N grid, where the size of each grid unit is k*k pixels; Node initialization: Calculate the grayscale mean of each grid cell. If the mean is greater than the preset threshold T, it is marked as a crack node with a value of 1, otherwise it is marked as a background node with a value of 0; Node connectivity calculation: Calculate the connectivity of the nodes using the 8-connectivity rule. For each crack node, check its connectivity with adjacent nodes. If connected, mark it as a crack region. Crack morphology analysis: calculate the length and width of the crack in the crack area; Crack detection: Calculate the ratio of crack length to crack width. When the ratio of crack length to width is greater than a threshold value U and the crack length is greater than a threshold value V, it is determined that the belt conveyor has longitudinal tearing. The node connectivity calculation includes the following operations: Initialize a label matrix to record the crack area to which each node belongs; Traverse all nodes and assign a new label to unlabeled crack nodes; Recursively or iteratively check its neighboring nodes, and if the neighboring node is a crack node and is unlabeled, assign the same label.

6. The method for detecting longitudinal tearing of a belt conveyor based on computer vision according to claim 5, characterized in that: The crack morphology analysis includes the following detailed steps: Scan the connected domain from top to bottom and record the position of the first white pixel as (x1, y1). Scan the connected domain from bottom to top and record the position of the first white pixel as (x2, y2). Scan the connected domain from left to right and record the position of the first white pixel as (x3, y3). Scan the connected domain from right to left and record the position of the first white pixel as (x4, y4); Xleft =min{x1, x2, x3, x4}, Xright =max{x1, x2, x, x}, Yup = min{y1, y 2, y 3, y 4}, Ydown =max{y1, y2, y3, y4}; The length of the crack = Ydown-Yup, and the width of the crack = Xright-Xleft.

7. The method for detecting longitudinal tearing of a belt conveyor based on computer vision according to claim 4, characterized in that: The sub-step 6.1 includes: Resize the image to a fixed size to facilitate input into the deep learning model for processing; Normalize the image and map the pixel value range to the [0, 1] interval to improve the training effect and convergence speed of the model; Enhance the image by performing operations such as random rotation, flipping, cropping, and adding noise to increase data diversity and prevent model overfitting.

8. The method for detecting longitudinal tearing of a belt conveyor based on computer vision according to claim 4, characterized in that: The sub-step 6.2 includes: A convolutional neural network (CNN) was selected as the basic architecture to build a deep learning model. Its output layer was configured with two output nodes to predict whether there was a longitudinal crack and the severity of the crack. The pre-collected conveyor belt images are divided into training sets and test sets according to a certain ratio for model training and evaluation; The training set is input into the deep learning model, and the output of the model is calculated through forward propagation. The loss function value is then calculated based on the difference between the output and the true label. The gradient of the loss function with respect to the model parameters is then calculated through the backpropagation algorithm, and the model parameters are updated using the optimization algorithm. This process is repeated until the accuracy and loss value of the model on the training set meet the preset requirements or the number of training rounds is completed. The images that need to be detected for longitudinal tearing are processed in the same preprocessing method as the training set, and then input into the trained deep learning model to output the prediction results of whether there is a longitudinal tear and the prediction results of the tear severity.

9. A belt conveyor longitudinal tear detection device for use in the method according to any one of claims 1 to 8, comprising: Reflective strips, light sources, CCD cameras, photodiodes, on-site computers, servers, roller brushes; A reflective strip is provided on the non-load-bearing surface of the conveyor belt of the belt conveyor at intervals of L, wherein the distance L is the same as the width of the conveyor belt; The light source adopts a high-brightness LED lamp, which can project a planar light spot and illuminate the non-load-bearing surface of the conveyor belt of the belt conveyor. The planar light spot can evenly illuminate the non-load-bearing surface between two adjacent reflective strips; The CCD camera uses an industrial camera with high resolution. The installation position of the CCD camera is adjusted so that the field of view of the CCD camera can completely cover the non-bearing surface between two adjacent reflective strips. The light source and photodiode are installed on the left and right sides of the CCD camera respectively, so that the light emitted by the light source can be reflected by the reflective strips and then received by the photodiode; The roller brush is set on one side of the light source, so that the non-load-bearing surface of the conveyor belt is first cleaned by the roller brush during movement and then illuminated by the planar light spot projected by the light source, so as to prevent dust and other particles from adhering to the non-load-bearing surface of the conveyor belt and interfering with the subsequent longitudinal tear detection; The on-site computer is electrically connected to the light source, CCD camera, and photodiode. The on-site computer can control the on and off of the light source through software program settings, trigger the CCD camera to shoot, and receive images output by the CCD camera; The on-site computer is connected to the server via the network.

10. The belt conveyor longitudinal tear detection device according to claim 9, characterized in that: The light emitted by the light source is diffusely reflected on the non-load-bearing surface of the conveyor belt and mirror-reflected on the reflective strip. When the reflective strip passes near the photodiode, the electrical signal output by the photodiode increases. The on-site computer detects the rising edge of the received electrical signal output by the photodiode to obtain a trigger signal, and controls the CCD camera to take pictures based on the trigger signal.

Images