Belt tearing detection method and device, storage medium and electronic equipment

Abstract

The invention relates to a belt tearing detection method and device, a storage medium and electronic equipment, which can effectively improve the accuracy of a belt tearing detection result and ensure the safe and stable operation of a system. The method comprises the following steps: acquiring belt image data; preprocessing the belt image data to obtain a target grayscale image; determining a target detection result according to the target grayscale image and a preset convolutional neural network; and determining an actual tearing position according to the target detection result.

Description

Technical Field

[0001] The present disclosure relates to the field of electric power technology, and in particular to a belt tear detection method, device, storage medium and electronic equipment. Background Art

[0002] In the power generation process of power plants, the coal transportation system is one of the key links. The coal conveyor belt conveyor is an important part of the coal transportation system, which is mainly responsible for transporting coal from the coal storage yard to the boiler raw coal bunker. However, the coal conveyor belt conveyor may encounter various problems in actual operation. Among them, belt tearing is a serious fault that may occur during the operation of the coal conveyor belt conveyor. Once the belt tear occurs, it will affect the normal operation of the coal transportation system. In severe cases, it may also cause the entire system to paralyze, which will have a serious impact on the power generation of the power plant. Therefore, in order to ensure the normal operation of the coal conveyor belt conveyor, it is necessary to detect and prevent possible belt tearing.

[0003] In the related art, belt tear detection is mainly performed by installing a laser above or below the belt to project straight laser stripes onto the bottom surface of the belt, and judging whether there is a tear on the belt surface by determining whether the center of the light strip is distorted, or by manual patrolling to detect belt tear. This will result in low accuracy of belt tear detection results, affecting the normal operation of the coal transportation system. Summary of the invention

[0004] The purpose of the present disclosure is to provide a belt tear detection method, device, storage medium and electronic device to solve the problems existing in the related art.

[0005] In order to achieve the above objectives, in a first aspect, the present disclosure provides a belt tear detection method, the method comprising: Get belt image data; Preprocessing the belt image data to obtain a target grayscale image; Determine a target detection result according to the target grayscale image and a preset convolutional neural network; The actual tearing position is determined according to the target detection result.

[0006] Optionally, the preprocessing of the belt image data to obtain a target grayscale image includes: Performing color space transformation on the belt image data to obtain a corresponding grayscale image; Sum each pixel value in the grayscale image and divide the sum by the total number of pixels to obtain a mean value; Calculating a standard deviation of the grayscale image according to each pixel value and the mean value; Subtract the mean value from each pixel value, and divide the result by the standard deviation to obtain a target pixel value corresponding to each pixel value; Each of the target pixel values ​​is linearly mapped into a first preset range to obtain the target grayscale image.

[0007] Optionally, determining the target detection result according to the target grayscale image and a preset convolutional neural network includes: The preset convolutional neural network is used to perform the following operations on the target grayscale image: Downsampling the target grayscale image to determine a first feature map; Upsampling the first feature map to determine a second feature map; Adding elements at each corresponding position in the first feature map and the second feature map to obtain a feature pyramid structure; The target detection result is determined according to the feature pyramid structure.

[0008] Optionally, determining the target detection result according to the feature pyramid structure includes: Determining a target heat map head and a target bounding box head according to the feature pyramid structure; Determine the target center point position according to the target heat map head; The target bounding box range is calculated according to the target center point position and the target bounding box head to determine the target detection result.

[0009] Optionally, the method further comprises: Determining a target semantic head according to the feature pyramid structure; Determining the actual tearing position according to the target detection result includes: Adjusting the target bounding box range to a second preset range; Determine, according to a result of semantic segmentation performed by the target semantic head on each pixel within the second preset range, a pixel position characterized as a torn pixel category; The actual tearing position is determined according to the pixel position.

[0010] Optionally, the method further comprises: When the actual tearing position is determined, warning information representing the actual tearing position is output.

[0011] In a second aspect, the present disclosure further provides a belt tear detection device, the device comprising: An acquisition module, used for acquiring belt image data; A preprocessing module, used for preprocessing the belt image data to obtain a target grayscale image; A target detection module, used to determine a target detection result based on the target grayscale image and a preset convolutional neural network; The determination module is used to determine the actual tearing position according to the target detection result.

[0012] Optionally, the preprocessing module is used to: Performing color space transformation on the belt image data to obtain a corresponding grayscale image; Sum each pixel value in the grayscale image and divide the sum by the total number of pixels to obtain a mean value; Calculating a standard deviation of the grayscale image according to each pixel value and the mean value; Subtract the mean value from each pixel value, and divide the result by the standard deviation to obtain a target pixel value corresponding to each pixel value; Each of the target pixel values ​​is linearly mapped into a first preset range to obtain the target grayscale image.

[0013] In a third aspect, the present disclosure further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the methods described in the first aspect.

[0014] In a fourth aspect, the present disclosure further provides an electronic device, including: a memory having a computer program stored thereon; A processor is used to execute the computer program in the memory to implement the steps of any one of the methods in the first aspect.

[0015] Through the above technical solution, the acquired belt image data is preprocessed to obtain the target grayscale image, and the target detection result is determined based on the target grayscale image and the preset convolutional neural network, and the actual tearing position is further determined based on the target detection result. Among them, the use of convolutional neural network for target detection can more accurately determine the tearing position, avoiding the uncertainty and error caused by simple optical methods or human factors in related technologies, thereby effectively improving the accuracy of belt tearing detection results and ensuring safe and stable operation of the system.

[0016] Other features and advantages of the present disclosure will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings are used to provide a further understanding of the present disclosure and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present disclosure but do not constitute a limitation of the present disclosure. In the accompanying drawings: Figure 1 The figure is a flow chart of a belt tear detection method according to an exemplary embodiment of the present disclosure.

[0018] Figure 2 The figure is a schematic diagram showing an arrangement of a conveyor belt and an image acquisition device according to an exemplary embodiment of the present disclosure.

[0019] Figure 3 is a schematic diagram of a polygonal outline of an actual tearing position according to an exemplary embodiment of the present disclosure.

[0020] Figure 4 The figure is a schematic diagram of an image target detection network structure according to an exemplary embodiment of the present disclosure.

[0021] Figure 5 It is a block diagram of a belt tear detection device according to an exemplary embodiment of the present disclosure.

[0022] Figure 6 It is a block diagram of an electronic device according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0023] The specific implementation of the present disclosure is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the present disclosure, and is not used to limit the present disclosure.

[0024] As mentioned in the background technology, in the related art, belt tear detection is mainly performed by installing a laser above or below the belt to project straight laser stripes onto the bottom surface of the belt, and judging whether there is a tear on the belt surface by determining whether the center of the light strip is distorted, or by manual patrolling to detect belt tear. This will result in low accuracy of belt tear detection results, affecting the normal operation of the coal transportation system.

[0025] In view of this, the present disclosure provides a belt tear detection method, device, storage medium and electronic device, which can effectively improve the accuracy of belt tear detection results and ensure safe and stable operation of the system.

[0026] Figure 1 FIG. 1 is a flow chart of a belt tear detection method according to an exemplary embodiment of the present disclosure. Figure 1 , the method comprises the following steps: Step S101, acquiring belt image data.

[0027] It should be understood that a CCD (Charge-Coupled Device) industrial camera may be used to collect belt image data. A CCD industrial camera is a digital camera device specifically used in the industrial field, and has high performance, high stability, and high reliability.

[0028] For example, a CCD industrial camera can be installed under the belt conveyor. There can be one or more cameras. If multiple cameras are used, they need to be arranged in parallel to ensure that the field of view of different cameras overlaps and is arranged along the width of the belt. If there is only one camera, it is necessary to ensure that the camera's field of view covers the entire width of the belt.

[0029] Figure 2 is a schematic diagram of an arrangement of a conveyor belt and an image acquisition device according to an exemplary embodiment of the present disclosure, referring to Figure 2 , numbers 1 to 7 are all CCD industrial cameras, number 8 is a belt conveyor, and the CCD industrial cameras can be fixed on the upper and lower layers of the belt conveyor with a bracket to monitor the tearing conditions on both sides of the belt in an oblique upward direction. Multiple CCD industrial cameras are arranged at a preset interval, and the number of cameras is set according to the length of the belt.

[0030] Step S102, preprocessing the belt image data to obtain a target grayscale image.

[0031] Step S103, determining the target detection result according to the target grayscale image and the preset convolutional neural network.

[0032] Step S104: determining the actual tearing position according to the target detection result.

[0033] In the above manner, the acquired belt image data is preprocessed to obtain a target grayscale image, and the target detection result is determined based on the target grayscale image and the preset convolutional neural network, and the actual tearing position is further determined based on the target detection result. Among them, the use of convolutional neural network for target detection can more accurately determine the tearing position, avoiding the uncertainty and error caused by simple optical methods or human factors in related technologies, thereby effectively improving the accuracy of belt tearing detection results and ensuring safe and stable operation of the system.

[0034] In order to enable those skilled in the art to better understand the belt tear detection method in the above embodiment of the present disclosure, the above steps are described below with examples.

[0035] In a possible manner, step S102 may be: Perform color space transformation on the belt image data to obtain a corresponding grayscale image; Sum the values ​​of each pixel in the grayscale image and divide it by the total number of pixels to get the mean; Calculate the standard deviation of the grayscale image based on each pixel value and the mean; Subtract the mean from each pixel value and divide it by the standard deviation to obtain the target pixel value corresponding to each pixel value; Each target pixel value is linearly mapped into a first preset range to obtain a target grayscale image.

[0036] It should be understood that the image data obtained from devices such as CCD industrial cameras is an RGB image, which is a common type of color image. It consists of three color channels: red, green, and blue. In an RGB image, each pixel uses the values ​​of these three colors to represent its color. Three 8-bit integers (0-255) are usually used to represent the color intensity of each channel. Therefore, the pixels of an RGB image are usually composed of three 8-bit integers, that is, each pixel has 24 bits.

[0037] Among them, the RGB image is converted into a grayscale image, that is, the color image is converted into a single-channel grayscale image. The specific steps can be as follows: for each pixel, according to the values ​​of its three RGB channels, it can be converted into a grayscale value using the following formula: Gray = 0.299 * R + 0.587 * G + 0.114 * B, where R, G, and B represent the pixel values ​​of the red, green, and blue channels respectively; then the grayscale value calculated above is used to replace the RGB value of the corresponding pixel in the original image to obtain a grayscale image.

[0038] It should also be understood that the process of standardizing grayscale images can make the pixel values ​​of the image evenly distributed within a certain range, which helps to improve the robustness and accuracy of image processing and analysis. The specific steps can be as follows: calculate the mean and standard deviation of the grayscale image, that is, find the mean and standard deviation of all pixels; for each pixel, subtract the mean from its original grayscale value and then divide it by the standard deviation to get the standardized pixel value. The new pixel value range can usually be [-1, 1] or [0, 1]; finally, the standardized pixel value can be mapped to the range of [0, 1] to better perform subsequent image processing and analysis.

[0039] In the above manner, the belt image data can be converted into a grayscale image, and the grayscale image can be standardized to meet specific data distribution and range requirements, thereby improving the effect and accuracy of image processing.

[0040] In a possible manner, step S103 may be: The following operations are performed on the target grayscale image through the preset convolutional neural network: Down-sampling the target grayscale image to determine a first feature map; Upsampling the first feature map to determine a second feature map; Add the elements of each corresponding position in the first feature map and the second feature map to obtain a feature pyramid structure; According to the feature pyramid structure, the target detection result is determined.

[0041] It should be understood that downsampling (Pooling) is to compress the pixel values ​​in each area by performing a pooling operation on the local area of ​​the input image, thereby reducing the size of the feature map. In this step, the target grayscale image is subjected to a downsampling operation to obtain a first feature map, which is usually small in size but retains certain feature information. Upsampling (Upsampling) is to expand the size of the feature map by interpolation and other methods to restore the high-resolution feature representation. In the embodiment of the present disclosure, the input image data after standardization is input, and the size is h*w*1, where h is the height of the image, w is the width of the image, and 1 is the feature of 1 channel. By presetting the convolution layer and pooling layer of the convolutional neural network model, the input image can be downsampled to extract the feature information in the image. The downsampling operation helps to reduce the size of the feature map while retaining important feature information, thereby obtaining a higher level of abstract feature representation. The feature map obtained by downsampling can be subjected to an upsampling operation to expand the size of the feature map, restore the resolution of the original image, and increase the detail information of the image, which helps to better locate and identify the target.

[0042] It should also be understood that the feature pyramid structure refers to a structure composed of feature maps obtained by upsampling and downsampling operations at different scales. In the disclosed embodiment, the first feature map and the second feature map are fused, for example, by element-wise addition, splicing or other methods, to achieve the combination of multi-level feature information, and obtain a feature pyramid structure that contains both low-level detail texture information and high-level semantic information. In the feature pyramid, the feature map at each scale can contain different levels of feature information, which is helpful for detecting targets of different sizes.

[0043] It should also be understood that object detection can be performed based on the overall framework of the CenterNet algorithm. The core idea of ​​the CenterNet algorithm is to achieve object detection by predicting the position and size of the center point of the target. Specifically, the CenterNet algorithm includes three main components: the backbone network, the center point prediction network, and the size prediction network.

[0044] (a) Backbone network: responsible for extracting features from the input image.

[0045] (b) Center point prediction network: used to predict the location of the target center point. This network maps the center point location of each target to a Gaussian heatmap through regression operation. In this Gaussian heatmap, the location of the target center point corresponds to the peak of the Gaussian distribution.

[0046] (c) Size prediction network: used to predict the width and height of the target. The network predicts the width and height of the bounding box corresponding to each point by performing regression operations on each point on the feature map.

[0047] During the training process, CenterNet uses Gaussian heatmap as the objective function to accurately locate the center point of the target by maximizing the similarity between the predicted heatmap and the true heatmap. At the same time, the size prediction network predicts the size of the target by regressing the width and height of the bounding box.

[0048] In the inference phase, low-quality predicted bounding boxes are first filtered out based on the confidence threshold, and then non-maximum suppression (NMS) is performed on the remaining bounding boxes to eliminate bounding boxes with more overlap. Finally, CenterNet can output the bounding box coordinates, category, and confidence of each object.

[0049] In summary, CenterNet can transform the target detection task into a regression problem and realize target detection by predicting the center point and width and height of the target. This design simplifies the target detection task and improves the detection accuracy.

[0050] For example, regarding the preset convolutional neural network, when designing the network architecture, the network structure can be determined according to actual needs, including the number, size, and connection method of components such as convolutional layers, pooling layers, and fully connected layers, etc., and the embodiments of the present disclosure are not limited to this.

[0051] In the above way, the feature pyramid structure can be used to more comprehensively characterize image features, which is suitable for processing target detection tasks of different scales and complexities, thereby improving the accuracy of target detection.

[0052] In one possible approach, determining the target detection result based on the feature pyramid structure may include: According to the feature pyramid structure, determine the target heat map head and the target bounding box head; According to the target heat map head, determine the target center point position; According to the target center point position and the target bounding box head, the target bounding box range is calculated to determine the target detection result.

[0053] It should be understood that the convolutional layer in the feature pyramid structure can be used to generate a heatmap head and a bounding box head from the fused features. Among them, the heatmap head can be used to generate a heatmap of the center point of the tearing location. Specifically, the values ​​of all points on the heatmap are between 0-1. When the data at a certain position is greater than the preset threshold, it can be considered that there is a torn target here. The bounding box head can usually contain four channels, which correspond to the position and size information of the bounding box in the target detection. These four channels represent the x-direction offset, y-direction offset, bounding box width and bounding box height from the center point heatmap. Through these four parameters and the position on the heatmap, the specific position information of the bounding box can be calculated.

[0054] Through the above method, the target center point position can be determined according to the target heat map header, and the target bounding box range can be calculated in combination with the target bounding box header as the target detection result, which can effectively reduce false detection and missed detection and improve the accuracy and stability of target detection.

[0055] In one possible embodiment, the method may further include: Determine the target semantic head based on the feature pyramid structure; According to the target detection results, the actual tearing position is determined, which may include: Adjusting the target bounding box range to a second preset range; Determine the pixel position characterized as a torn pixel category according to the result of semantic segmentation performed by the target semantic head on each pixel within the second preset range; Based on the pixel position, the actual tearing position is determined.

[0056] It should be understood that the target bounding box range is adjusted to the second preset range because the detected bounding box position may have a certain deviation, so the range of the detection box can be appropriately enlarged to reduce errors and improve detection efficiency.

[0057] It should also be understood that a semantic head refers to a part of a deep learning neural network, which is usually used to perform high-level semantic understanding and representation of input data. In the disclosed embodiment, the convolutional layer in the feature pyramid structure is used to generate a semantic head with fused features, which can be used to accurately segment all pixel points at the tear location and obtain the precise pixel-level position of the tear.

[0058] Specifically, according to the segmentation results corresponding to all pixels in the second preset range in the target semantic header, if the segmentation results contain pixels characterized as torn pixel categories, it is considered that there is actual tearing, and a polygonal contour can be generated for the area corresponding to the torn pixels, and an alarm message is output, wherein the polygonal contour of the actual tearing position can be referenced Figure 3 If the segmentation result does not contain pixels characterized as torn pixels, it is considered that there is no actual tearing, and the detection results of the target heat map head and the target bounding box head in the previous steps are unreliable and are false detections. The detection results representing the range of this target bounding box are deleted.

[0059] For example, the second preset range can be set according to actual conditions, and the embodiments of the present disclosure are not limited to this.

[0060] By combining the target semantic head with the target detection result in the above way, the actual belt tearing position can be determined more accurately, the error can be reduced, and the detection accuracy and efficiency can be improved.

[0061] Figure 4 is a schematic diagram of an image target detection network structure according to an exemplary embodiment of the present disclosure, referring to Figure 4 In this structure, the input grayscale image can be downsampled using the preset convolutional neural network (CNN) to extract features, and then the features are upsampled and fused with the downsampled features to form a feature pyramid structure that fuses multiple layers of information. Finally, the fused features are used to generate three heads using the convolutional layer, namely the heatmap head, the bounding box head, and the semantic head.

[0062] In one possible embodiment, the method further includes: When the actual tearing position is determined, warning information indicating the actual tearing position is output.

[0063] For example, the alarm information can be output in the form of voice reminder or text reminder. The voice reminder can send corresponding voice through a voice transmission device such as a buzzer, a speaker, etc. to remind the staff to pay attention to the detected tearing position. The text reminder device can also send texts such as text reminders, email reminders, etc. to specific staff to remind them to check the corresponding tearing position.

[0064] Through the above method, relevant staff can be notified in a timely and accurate manner of the specific location where maintenance is required, thereby improving maintenance efficiency and ensuring the safety and stability of equipment operation.

[0065] Figure 5is a block diagram of a belt tear detection device 500 according to an exemplary embodiment. Figure 5 , the belt tearing detection device 500 may include: An acquisition module 501 is used to acquire belt image data; A preprocessing module 502 is used to preprocess the belt image data to obtain a target grayscale image; The target detection module 503 is used to determine the target detection result according to the target grayscale image and the preset convolutional neural network; The determination module 504 is used to determine the actual tearing position according to the target detection result.

[0066] Optionally, the preprocessing module 502 is used to: Perform color space transformation on the belt image data to obtain a corresponding grayscale image; Sum the values ​​of each pixel in the grayscale image and divide it by the total number of pixels to get the mean; Calculate the standard deviation of the grayscale image based on each pixel value and the mean; Subtract the mean from each pixel value and divide it by the standard deviation to obtain the target pixel value corresponding to each pixel value; Each target pixel value is linearly mapped into a first preset range to obtain a target grayscale image.

[0067] Optionally, the target detection module 503 is used to: The following operations are performed on the target grayscale image through the preset convolutional neural network: Down-sampling the target grayscale image to determine a first feature map; Upsampling the first feature map to determine a second feature map; Add the elements of each corresponding position in the first feature map and the second feature map to obtain a feature pyramid structure; According to the feature pyramid structure, the target detection result is determined.

[0068] Optionally, the target detection module 503 is used to: According to the feature pyramid structure, determine the target heat map head and the target bounding box head; According to the target heat map head, determine the target center point position; According to the target center point position and the target bounding box head, the target bounding box range is calculated to determine the target detection result.

[0069] Optionally, the target detection module 503 is further used for: Determine the target semantic head based on the feature pyramid structure; The determining module 504 is used for: Adjusting the target bounding box range to a second preset range; Determine the pixel position characterized as a torn pixel category according to the result of semantic segmentation performed by the target semantic head on each pixel within the second preset range; Based on the pixel position, the actual tearing position is determined.

[0070] Optionally, the device further comprises an alarm module, configured to: When the actual tearing position is determined, warning information indicating the actual tearing position is output.

[0071] Through the above device, the acquired belt image data is preprocessed to obtain a target grayscale image, and the target detection result is determined based on the target grayscale image and the preset convolutional neural network, and the actual tearing position is further determined based on the target detection result. Among them, the use of convolutional neural network for target detection can more accurately determine the tearing position, avoiding the uncertainty and error caused by simple optical methods or human factors in related technologies, thereby effectively improving the accuracy of belt tearing detection results and ensuring safe and stable operation of the system.

[0072] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0073] In summary, the belt tear detection method and device in the embodiment of the present disclosure have the following advantages compared with the related art: 1) High detection accuracy: The surface condition of the belt can be detected with high accuracy, including tiny cracks, flaws, etc., which makes it easier to find and locate problems and reduce missed inspections.

[0074] 2) Strong real-time performance: It can monitor the running status of the belt in real time, detect and deal with abnormal situations in time, avoid problems such as belt tearing, and help ensure the stable operation of the production line.

[0075] 3) High degree of automation: It adopts non-contact detection method, which can perform high-precision detection without touching the belt. The degree of automation is high, which reduces the difficulty and workload of manual detection.

[0076] 4) High reliability: It has the characteristics of high efficiency and stability, can work continuously for a long time, and is not affected by external factors such as environment and climate, ensuring the stability and reliability of the test results.

[0077] 5) Strong scalability: It can meet the detection needs of belts of different types and specifications through different configurations and expansion methods, and has strong adaptability.

[0078] Figure 6FIG. 6 is a block diagram of an electronic device 600 according to an exemplary embodiment. Figure 6 As shown, the electronic device 600 may include: a processor 601 , a memory 602 . The electronic device 600 may also include one or more of a multimedia component 603 , an input / output (I / O) interface 604 , and a communication component 605 .

[0079] The processor 601 is used to control the overall operation of the electronic device 600 to complete all or part of the steps in the belt tear detection method described above. The memory 602 is used to store various types of data to support the operation of the electronic device 600, and these data may include, for example, instructions for any application or method used to operate on the electronic device 600, and application-related data. The memory 602 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk. The multimedia component 603 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving an external audio signal. The received audio signal may be further stored in the memory 602 or sent through the communication component 605. The audio component also includes at least one speaker for outputting an audio signal. The I / O interface 604 provides an interface between the processor 601 and other interface modules, and the above-mentioned other interface modules may be a keyboard, a mouse, a button, etc. These buttons may be virtual buttons or physical buttons. The communication component 605 is used for wired or wireless communication between the electronic device 600 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, 4G, NB-IOT, eMTC, or other 5G, etc., or a combination of one or more of them, is not limited here. Therefore, the corresponding communication component 605 may include: a Wi-Fi module, a Bluetooth module, an NFC module, etc.

[0080] In an exemplary embodiment, the electronic device 600 can be implemented by one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components to perform the above-mentioned belt tear detection method.

[0081] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the belt tear detection method described above are implemented. For example, the computer-readable storage medium may be the memory 602 including the program instructions, and the program instructions may be executed by the processor 601 of the electronic device 600 to implement the belt tear detection method described above.

[0082] The preferred embodiments of the present disclosure are described in detail above in conjunction with the accompanying drawings; however, the present disclosure is not limited to the specific details in the above embodiments. Within the technical concept of the present disclosure, a variety of simple modifications can be made to the technical solution of the present disclosure, and these simple modifications all fall within the protection scope of the present disclosure.

[0083] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present disclosure will not further describe various possible combinations.

[0084] In addition, various embodiments of the present disclosure may be arbitrarily combined, and as long as they do not violate the concept of the present disclosure, they should also be regarded as the contents disclosed by the present disclosure.

Claims

1. A belt tear detection method, characterized in that: The method comprises: Get belt image data; Preprocessing the belt image data to obtain a target grayscale image; Determine a target detection result according to the target grayscale image and a preset convolutional neural network; The actual tearing position is determined according to the target detection result.

2. The method according to claim 1, characterized in that The preprocessing of the belt image data to obtain a target grayscale image includes: Performing color space transformation on the belt image data to obtain a corresponding grayscale image; Sum each pixel value in the grayscale image and divide the sum by the total number of pixels to obtain a mean value; Calculating a standard deviation of the grayscale image according to each pixel value and the mean value; Subtract the mean value from each pixel value, and divide the result by the standard deviation to obtain a target pixel value corresponding to each pixel value; Each of the target pixel values ​​is linearly mapped into a first preset range to obtain the target grayscale image.

3. The method according to claim 1, characterized in that Determining the target detection result according to the target grayscale image and the preset convolutional neural network includes: The preset convolutional neural network is used to perform the following operations on the target grayscale image: Downsampling the target grayscale image to determine a first feature map; Upsampling the first feature map to determine a second feature map; Adding elements at each corresponding position in the first feature map and the second feature map to obtain a feature pyramid structure; The target detection result is determined according to the feature pyramid structure.

4. The method according to claim 3, characterized in that The determining the target detection result according to the feature pyramid structure includes: Determining a target heat map head and a target bounding box head according to the feature pyramid structure; Determine the target center point position according to the target heat map head; The target bounding box range is calculated according to the target center point position and the target bounding box head to determine the target detection result.

5. The method according to claim 4, characterized in that The method further comprises: Determining a target semantic head according to the feature pyramid structure; Determining the actual tearing position according to the target detection result includes: Adjusting the target bounding box range to a second preset range; Determine, according to a result of semantic segmentation performed by the target semantic head on each pixel within the second preset range, a pixel position characterized as a torn pixel category; The actual tearing position is determined according to the pixel position.

6. The method according to claim 1, characterized in that The method further comprises: When the actual tearing position is determined, warning information representing the actual tearing position is output.

7. A belt tear detection device, characterized in that: The device comprises: An acquisition module, used for acquiring belt image data; A preprocessing module, used for preprocessing the belt image data to obtain a target grayscale image; A target detection module, used to determine a target detection result based on the target grayscale image and a preset convolutional neural network; The determination module is used to determine the actual tearing position according to the target detection result.

8. The device according to claim 7, characterized in that The pre-processing module is used for: Performing color space transformation on the belt image data to obtain a corresponding grayscale image; Sum each pixel value in the grayscale image and divide the sum by the total number of pixels to obtain a mean value; Calculating a standard deviation of the grayscale image according to each pixel value and the mean value; Subtract the mean value from each pixel value, and divide the result by the standard deviation to obtain a target pixel value corresponding to each pixel value; Each of the target pixel values ​​is linearly mapped into a first preset range to obtain the target grayscale image.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

10. An electronic device, characterized in that: include: a memory having a computer program stored thereon; A processor, configured to execute the computer program in the memory to implement the steps of the method according to any one of claims 1 to 6.

Images