A belt tear detection method, device, apparatus and storage medium

Abstract

The application provides a belt tear detection method, device, equipment and storage medium, the method comprises the following steps: obtaining a target belt sample image, extracting first feature information of the target belt sample image through a target feature coding main network, the first feature information comprising spatial domain feature information; extracting second feature information of the target belt sample image through a target frequency domain branch network, the second feature information comprising frequency domain feature information; fusing the first feature information and the second feature information to obtain fused feature information; training the fused feature information in an initial classification network to obtain a final classification network; inputting target fused feature information of a to-be-detected belt image into the final classification network to output a tear detection result of the to-be-detected belt image. The application adopts a deep learning visual algorithm to automatically detect the tearing condition of the belt, and uses frequency domain information of the image to enhance model feature representation, so that accurate detection effect can be obtained.

Description

Technical Field

[0001] This invention relates to the field of machine vision technology, and in particular to a method, apparatus, equipment and storage medium for detecting belt tearing. Background Technology

[0002] As a primary piece of equipment in large industrial and mining enterprises, belt conveyors are extremely vulnerable to damage. Tear or breakage accidents can severely disrupt normal production and easily lead to safety accidents and significant property losses. Therefore, timely detection of belt tears and the implementation of emergency measures are crucial. Existing belt tear detection methods generally fall into three categories. The first is manual inspection, requiring workers to inspect belts one by one in multiple production environments, or necessitating multiple workers, resulting in high labor costs and low efficiency. The second method uses auxiliary detectors, which require additional hardware costs. These detectors are typically rod-shaped, linear, or leak-type, designed to detect belt tears. However, auxiliary detectors are often simple in structure, unreliable, and prone to false positives and false negatives. The third method uses traditional visual processing, but due to significant differences in image quality between cameras in different production environments, traditional visual recognition has poor robustness and cannot be universally applied across multiple production environments. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to overcome the shortcomings of the prior art and provide a belt tear detection method, apparatus, device and storage medium.

[0004] This invention provides the following technical solution:

[0005] In a first aspect, this application provides a method for detecting belt tearing, including:

[0006] A target belt sample image is acquired, and the first feature information of the target belt sample image is extracted through a target feature encoding main network. The first feature information includes spatial domain feature information.

[0007] The second feature information of the target belt sample image is extracted by the target frequency domain branch network, and the second feature information includes frequency domain feature information.

[0008] The first feature information and the second feature information are fused to obtain fused feature information;

[0009] The fused feature information is used to train the initial classification network to obtain the final classification network;

[0010] The target fusion feature information of the belt image to be tested is input into the final classification network, and the tear detection result of the belt image to be tested is output.

[0011] In one embodiment, acquiring the target belt sample image includes:

[0012] A torn belt image and a normal belt image are acquired, and the torn belt image and the normal belt image are superimposed using a mask to obtain the target belt sample image.

[0013] In one embodiment, the step of superimposing the torn belt image and the normal belt image using a mask to obtain the target belt sample image includes:

[0014] Function to construct initial belt sample image:

[0015]

[0016] Among them, I d For images of torn belts, M d I is the crack mask corresponding to the torn belt image. n This is a normal belt image. The initial belt sample image is the torn image, and ⊙ represents the dot product calculation.

[0017] The initial belt sample image is transformed in position and size to obtain the target belt sample image:

[0018] I = αT p T s (I d ⊙M d )+I n ⊙(1-T p T s (M d )),

[0019] Among them, T p T s Let represent the position transformation function and the size transformation function, respectively; α∈(0,1) represents a random transparency parameter; and I is the target belt sample image.

[0020] In one embodiment, the extraction of second feature information from the target belt sample image via a target frequency domain branch network, the second feature information including frequency domain feature information, includes:

[0021] The low-pass and high-pass frequency components of the target belt sample image are calculated by using a one-dimensional wavelet transform in the horizontal direction.

[0022] The frequency domain feature information is calculated by using another one-dimensional wavelet transform in the vertical direction of the target belt sample image.

[0023] In one embodiment, fusing the first feature information and the second feature information to obtain fused feature information includes:

[0024] Obtain the spatial domain feature information Z from the output of the (l-1)th layer feature encoding main network. l-1 And obtain the frequency domain feature information Z′ of the k-th layer frequency domain branch network output. k Among them, Z l-1 With Z′ k They have the same width and height dimensions;

[0025] Z l-1 With Z′ k After the splicing operation, a convolution operation f is performed. M Finally with Z l-1 Perform residual operations to obtain the final output Z of the next layer. l :

[0026]

[0027] In one implementation, the target feature encoding main network and the target frequency domain branch network are trained in the following manner:

[0028] Obtain training images, assuming that the set of torn image samples in the training images is D and the set of normal image samples is N;

[0029] The corresponding mask image is M d The feature obtained after the encoder is F d Then its normal region characteristics are It is made by M d The feature mask obtained by downsampling, wherein the encoder is the target feature encoding main network or the target frequency domain branch network;

[0030] The feature obtained after the encoder is F. n B d and F n After global average pooling, b is obtained respectively. d and f n ;

[0031] The calculated background regularization loss is:

[0032]

[0033] In one implementation, training the initial classification network with the fused feature information to obtain the final classification network includes:

[0034] Step 1: Input the fused feature information into the initial classification network to obtain the initial output result;

[0035] Step 2: Determine whether the initial output result is a preset result, where the preset result includes torn and untorn;

[0036] Step 3: If the initial output result is not the preset result, adjust the parameters of the initial classification network to obtain the first classification network;

[0037] Step 4: Input the fused feature information into the first classification network, and repeat steps 1 to 3 until the output of the Nth classification network is the preset result. Then, the Nth classification network is used as the final classification network.

[0038] Secondly, this application provides a belt tear detection device, comprising:

[0039] The first extraction module is used to acquire a target belt sample image and extract first feature information of the target belt sample image through a target feature encoding main network. The first feature information includes spatial domain feature information.

[0040] The second extraction module is used to extract second feature information of the target belt sample image through a target frequency domain branch network, wherein the second feature information includes frequency domain feature information.

[0041] The fusion module is used to fuse the first feature information and the second feature information to obtain fused feature information;

[0042] The training module is used to train the initial classification network with the fused feature information to obtain the final classification network;

[0043] The output module is used to input the target fusion feature information of the belt image to be tested into the final classification network and output the tear detection result of the belt image to be tested.

[0044] Thirdly, this application provides an electronic device including a memory and at least one processor, the memory storing a computer program, and the processor executing the computer program to implement the belt tear detection method as described in the first aspect.

[0045] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed, implements the belt tear detection method as described in the first aspect.

[0046] The embodiments of the present invention have the following beneficial effects:

[0047] The belt tear detection method provided by this invention uses a deep learning visual algorithm to automatically detect belt tears. By utilizing the frequency domain information of the image to enhance the model feature representation, accurate detection results can be obtained.

[0048] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 A schematic diagram of a belt tear detection method is shown.

[0051] Figure 2 A schematic diagram of a feature encoding network structure is shown;

[0052] Figure 3 A schematic diagram of a method for calculating frequency domain feature information is shown.

[0053] Figure 4 A schematic diagram of an autoencoder network structure is shown;

[0054] Figure 5 A schematic diagram of the overall framework structure of a belt tear detection network is shown.

[0055] Figure 6 A schematic diagram of the frame structure of a belt tear detection device is shown.

[0056] Explanation of key component symbols:

[0057] 600. Belt tear detection device; 601. First extraction module; 602. Second extraction module; 603. Fusion module; 604. Training module; 605. Output module. Detailed Implementation

[0058] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0059] It should be noted that when an element is said to be "fixed" to another element, it can be directly on the other element or there may be an intervening element. When an element is said to be "connected" to another element, it can be directly connected to the other element or there may be an intervening element. Conversely, when an element is said to be "directly" on another element, there is no intervening element. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0060] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0061] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0062] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the template description is for the purpose of describing particular embodiments only and is not intended to limit the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0063] Example 1

[0064] See Figure 1 , Figure 1 This embodiment provides a schematic flowchart of a belt tear detection method. This method can be used to detect whether a belt in a belt conveyor is torn, and can also be used to detect other belts that are the same as or similar to the belt in the belt conveyor. The method includes:

[0065] S101. Obtain a target belt sample image, and extract the first feature information of the target belt sample image through the target feature encoding main network. The first feature information includes spatial domain feature information.

[0066] In belt tear detection scenarios, the number of actual production belt tear images is limited, insufficient to support the training of neural network models. Directly using raw data to train the model will make it difficult for the model to learn enough tear sample features and patterns.

[0067] In one embodiment, this application provides a direct and efficient data augmentation scheme for belt tear samples, namely, mask-based sample augmentation to obtain a target belt sample image, including:

[0068] A torn belt image and a normal belt image are acquired, and the torn belt image and the normal belt image are superimposed using a mask to obtain the target belt sample image.

[0069] Specifically, this includes: a function for constructing the initial belt sample image.

[0070]

[0071] Among them, I d For images of torn belts, M d I is the crack mask corresponding to the torn belt image. n This is a normal belt image. The image represents the initial belt sample image, i.e., the torn image. ⊙ indicates dot product calculation.

[0072] Data augmentation can introduce more variants of torn samples, increase the diversity of the dataset, reduce the model's overfitting to the training set, and improve its generalization ability.

[0073] The enhancement method described above uses a mask to overlay the cracked portion of the torn sample with a normal sample to generate a fake torn sample, but it still suffers from the problem of monotonous crack variation. Therefore, in the above enhancement operation, transformations such as position and size are introduced to enrich the fake torn sample.

[0074] Specifically, this includes: performing position and size transformations on the initial belt sample image function to obtain the target belt sample image.

[0075] I = αT p T s (I d ⊙M d )+I n ⊙(1-T p T s (M d )),

[0076] Among them, T p T s Let represent the position transformation function and the size transformation function, respectively; α∈(0,1) represents a random transparency parameter; and I is the target belt sample image.

[0077] The above transformations ensure that the augmented data remains consistent and reliable with the real data, while enriching the number and diversity of torn samples, thereby improving the model's performance and generalization ability.

[0078] See Figure 2 , Figure 2 This is a schematic diagram of a feature coding network structure provided in this embodiment;

[0079] Feature coding networks consist of a main feature coding network and a frequency domain branch network. Larger squares (odd-numbered squares, such as the first and third squares) represent network layers, while smaller squares (even-numbered squares, such as the second and fourth squares) represent the output features of those layers. The numbers represent the dimensions of the data and features. The main feature coding network primarily performs convolutional operations, while the frequency domain branch network includes both transform and convolutional operations.

[0080] S102. Extract the second feature information of the target belt sample image through the target frequency domain branch network. The second feature information includes frequency domain feature information.

[0081] The transformation operation of the network layer in the frequency domain branch network mainly transforms the spatial domain information of the image into the frequency domain information through Discrete Wavelet Transform (DWT). In implementation, Haar wavelet transform is used, and two one-dimensional wavelet transforms can be performed sequentially in different directions for the image data.

[0082] See Figure 3 Step S102 further includes:

[0083] S1021. Using a one-dimensional wavelet transform in the horizontal direction of the target belt sample image, the low-pass and high-pass frequency components of the target belt sample image are calculated.

[0084] S1022. Using another one-dimensional wavelet transform in the vertical direction of the target belt sample image, the frequency domain feature information is calculated.

[0085] After wavelet transform, four images containing diagonal high-frequency information, horizontal high-frequency information, vertical high-frequency information, and low-frequency information are obtained, with a resolution half that of the original size. The transformed images are stacked according to the channel dimension, and then the frequency domain information of the images is extracted by convolution operation.

[0086] See Figure 4 , Figure 4 This is a schematic diagram of a self-encoding network structure provided in this embodiment.

[0087] In one implementation, an additional decoder can be designed on the basis of the original feature encoding network to form an autoencoder network for image reconstruction. The unsupervised image reconstruction pre-training task can effectively improve the feature encoder's ability to extract image features.

[0088] Since the encoder comprises two branches—a main network and a frequency domain branch network—the decoder also reconstructs the image from both the frequency and spatial domains during reconstruction. The main network layers of the encoder primarily perform deconvolution operations, while the frequency domain branch network layers include transform and deconvolution operations; the transform operation is the inverse of the discrete wavelet transform.

[0089] To improve the network's attention to cracks in the image, a region-sensitive reconstruction loss is adopted. A weighting matrix is ​​introduced into the mean squared error loss in the reconstruction task, making the network focus more on cracked regions in the torn image. The loss function is:

[0090]

[0091] Wherein, the reconstructed image is R, the original input image is I, and the weight matrix is ​​W. This indicates element-wise multiplication (positional multiplication), (RI) 2 This represents the square of the difference between corresponding pixels, where N is the number of pixels in the image. The weight matrix W is a matrix of the same size as the image, used to specify the weight of each pixel, assigning higher weights to cracked areas and lower weights to the normal background.

[0092] This embodiment constructs an autoencoder network for image reconstruction, reconstructs the encoded features, and outputs the reconstructed image. By comparing the difference between the reconstructed image and the input image, i.e. calculating the loss function value, it can be determined whether the extraction of the encoded features is accurate.

[0093] S103. The first feature information and the second feature information are fused to obtain fused feature information.

[0094] Obtain the spatial domain feature information Z from the output of the (l-1)th layer feature encoding main network. l-1 And obtain the frequency domain feature information Z′ of the k-th layer frequency domain branch network output. k Among them, Z l-1 With Z′ k They have the same width and height dimensions;

[0095] Z l-1 With Z′ k After the splicing operation, a convolution operation f is performed. M Finally with Z l-1 Perform residual operations to obtain the final output Z of the next layer. l :

[0096]

[0097] After extracting features in the frequency domain network, the feature information is fused with features of the same size in the main network. This allows frequency domain information to be incorporated into the main network, helping the network learn to extract more comprehensive features.

[0098] In one implementation, the feature encoding network should produce a relatively consistent depth representation of the normal regions in the torn images as it does in the normal training images, because they have similar appearances. Given a large number of normal belt training images but few torn belt images, the encoder learns data containing a large number of normal regions.

[0099] Therefore, background regularization can be used to maximize the similarity between normal regions in torn images and abundant normal training samples. Specifically, background regularization promotes a unified representation of normal regions in the image by maximizing the similarity between normal regions within the torn image and the global region of the normal training image. This aligns the non-tear areas of normal and torn samples, thus better describing the features of normal regions and making the system more sensitive to cracked areas in belt images, generating more differentiated crack features.

[0100] Therefore, the target feature encoding main network and the target frequency domain branch network are trained in the following manner:

[0101] Obtain training images, assuming that the set of torn image samples in the training images is D and the set of normal image samples is N;

[0102] The corresponding mask image is M d The feature obtained after the encoder is F d Then its normal region characteristics are It is made by M d The feature mask obtained by downsampling, wherein the encoder is the target feature encoding main network or the target frequency domain branch network;

[0103] The feature obtained after the encoder is F. n B d and F n After global average pooling, b is obtained respectively. d and f n ;

[0104] The calculated background regularization loss is:

[0105]

[0106] In this way, the characteristics of normal areas can be better described, and the cracked areas in belt images can be more sensitive, producing more differentiated crack features.

[0107] S104. The fused feature information is used to train the initial classification network to obtain the final classification network.

[0108] The training process of the final classification network specifically includes:

[0109] Step 1: Input the fused feature information into the initial classification network to obtain the initial output result;

[0110] Step 2: Determine whether the initial output result is a preset result, where the preset result includes torn and untorn;

[0111] Step 3: If the initial output result is not the preset result, adjust the parameters of the initial classification network to obtain the first classification network;

[0112] Step 4: Input the fused feature information into the first classification network, and repeat steps 1 to 3 until the output of the Nth classification network is the preset result. Then, the Nth classification network is used as the final classification network.

[0113] By training the initial classification network with fused feature information, the final classification network can determine whether a belt image is torn based on the fused feature information.

[0114] S105. Input the target fusion feature information of the belt image to be tested into the final classification network, and output the tear detection result of the belt image to be tested.

[0115] See Figure 5 , Figure 5 This is a schematic diagram of the overall framework structure of the belt tear detection network provided in this embodiment.

[0116] Before an image is input, it can be preprocessed, such as through image enhancement preprocessing.

[0117] After obtaining the final classification network, in addition to performing tear detection on the belt image under test, the trained encoding network and the final classification network can be used to train the belt tear detection task to improve the accuracy of the entire belt tear detection network. This task is a binary classification task, determining whether there is a tear in the belt in the image. The loss function is as follows:

[0118] L = L CE +L BR ,

[0119] in, L CE This represents the loss of the classification network. By calculating the loss function, we can determine whether the network has reached the preset accuracy.

[0120] Example 2

[0121] See Figure 6 This application also provides a belt tear detection device 600, comprising:

[0122] The first extraction module 601 is used to acquire a target belt sample image and extract first feature information of the target belt sample image through a target feature encoding main network. The first feature information includes spatial domain feature information.

[0123] The second extraction module 602 is used to extract second feature information of the target belt sample image through a target frequency domain branch network, wherein the second feature information includes frequency domain feature information.

[0124] The fusion module 603 is used to fuse the first feature information and the second feature information to obtain fused feature information;

[0125] Training module 604 is used to train the initial classification network with the fused feature information to obtain the final classification network;

[0126] The output module 605 is used to input the target fusion feature information of the belt image to be tested into the final classification network and output the tear detection result of the belt image to be tested.

[0127] The belt tear detection device provided in this embodiment uses a deep learning vision algorithm to automatically detect belt tears. By utilizing the frequency domain information of the image to enhance the model feature representation, accurate detection results can be obtained.

[0128] It is understood that the implementation method of belt tear detection described in Embodiment 1 above is also applicable to this embodiment and can achieve the same technical effect. To avoid redundancy, it will not be described again here.

[0129] Example 3

[0130] This application also provides a computer device, which may be, but is not limited to, a desktop computer, a laptop, etc., and its form is not limited, mainly depending on whether it needs to support the interface display function of a browser webpage. Exemplarily, the computer device includes a memory and at least one processor. The memory stores a computer program, and the processor executes the computer program to implement the belt tear detection method described in Embodiment 1 above.

[0131] The processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including at least one of a Central Processing Unit (CPU), Graphics Processing Unit (GPU), Network Processor (NP), Digital Signal Processor (DSP), Application-Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application.

[0132] The memory can be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), and Electrically Erasable Programmable Read-Only Memory (EEPROM). The memory stores computer programs, and the processor, upon receiving execution instructions, can execute the computer programs accordingly.

[0133] Furthermore, the memory may include a stored program area and a stored data area, wherein the stored program area may store the operating system and application programs required for at least one function; the stored data area may store data created based on the use of the computer device (such as iterative data, version data, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0134] Example 4

[0135] This application also provides a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are called and run by a processor, the computer-executable instructions cause the processor to run the belt tear detection method described in Embodiment 1 above.

[0136] It is understood that the implementation method of belt tear detection described in Embodiment 1 above is also applicable to this embodiment, so it will not be described again here.

[0137] The computer-readable storage medium can be either a non-volatile storage medium or a volatile storage medium. For example, the computer-readable storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0138] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that, in alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0139] In addition, the functional modules or units in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0140] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a smartphone, personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0141] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

[0142] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not as limitations; therefore, other examples of exemplary embodiments may have different values.

[0143] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0144] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A belt tear detection method characterized by, The method comprises the following steps: obtaining a target belt sample image, extracting first feature information of the target belt sample image through a target feature coding main network, wherein the first feature information comprises spatial domain feature information; extracting second feature information of the target belt sample image through a target frequency domain branch network, wherein the second feature information comprises frequency domain feature information; fusing the first feature information and the second feature information to obtain fused feature information; training the fused feature information in an initial classification network to obtain a final classification network; inputting target fused feature information of a to-be-tested belt image into the final classification network to output a belt tearing detection result of the to-be-tested belt image; the step of obtaining a target belt sample image comprises the following steps: obtaining a belt tearing image and a normal belt image, and superimposing the belt tearing image and the normal belt image through a mask to obtain the target belt sample image; the step of superimposing the belt tearing image and the normal belt image through a mask to obtain the target belt sample image comprises the following steps: constructing an initial belt sample image function: ， wherein, is a torn belt image, is a mask image corresponding to the torn belt image, is a normal belt image, is an initial belt sample image, i.e. a torn image, denotes a point-wise multiplication. performing position and size transformation on the initial belt sample image function to obtain a target belt sample image: ， wherein, respectively denote a position transformation function and a size transformation function, denotes a random transparency parameter, I a target belt sample image.

2. The belt tear detection method of claim 1, wherein the step of extracting second feature information of the target belt sample image through a target frequency domain branch network, wherein the second feature information comprises frequency domain feature information, comprises the following steps: utilizing a one-dimensional wavelet transform in a horizontal direction of the target belt sample image to calculate low-pass and high-pass frequency components of the target belt sample image; utilizing another one-dimensional wavelet transform in a vertical direction of the target belt sample image to calculate the frequency domain feature information.

3. The belt tear detection method of claim 1, wherein the step of fusing the first feature information and the second feature information to obtain fused feature information comprises the following steps: Get the Spatial domain feature information output by the layer feature encoding main network and obtain the first Frequency domain feature information output by the layer frequency domain branch network ,in, and They have the same width and height dimensions; concatenation operation with convolution operation , and finally residual operation to obtain the final output of the next layer : 。 4. The belt tear detection method of claim 1, wherein the target feature coding main network and the target frequency domain branch network are trained through the following method: obtaining training images, assuming that a tearing image sample set in the training images is D and a normal image sample set is N; The corresponding mask image is The feature obtained after the encoder is The normal region feature is , The feature mask obtained by Downsampling, wherein the encoder is a target feature encoding main network or the target frequency domain branch network; , the features obtained after the encoder are , the features obtained after the encoder are , the features obtained after the encoder are , the features obtained after the encoder are , the features obtained after the encoder are ; calculating a background regularization loss as: 。 5. The belt tear detection method of claim 4, wherein, the step of training the fused feature information in an initial classification network to obtain a final classification network comprises the following steps: step one, inputting the fused feature information into the initial classification network to obtain an initial output result; step two, judging whether the initial output result is a preset result, wherein the preset result comprises tearing and non-tearing; step three, if the initial output result is not the preset result, adjusting parameters of the initial classification network to obtain a first classification network; step four, inputting the fused feature information into the first classification network, and repeatedly executing steps one to three until an output result of an Nth classification network is the preset result, and then taking the Nth classification network as the final classification network.

6. A belt tear detection apparatus characterized by, The belt tearing detection device is used for executing the belt tearing detection method according to any one of claims 1-5, and the belt tearing detection device comprises: a first extraction module, configured to obtain a target belt sample image, extract first feature information of the target belt sample image through a target feature coding main network, wherein the first feature information comprises spatial domain feature information; a second extraction module configured to extract second feature information of the target belt sample image through a target frequency domain branch network, the second feature information including frequency domain feature information; a fusion module configured to fuse the first feature information and the second feature information to obtain fused feature information; a training module configured to train the fused feature information in an initial classification network to obtain a final classification network; an output module configured to input target fused feature information of a to-be-tested belt image into the final classification network to output a tearing detection result of the to-be-tested belt image.

7. An electronic device, comprising: A computer readable storage medium stores a computer program, and the computer program is executed to implement the belt tearing detection method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed to implement the belt tearing detection method according to any one of claims 1 to 5.

Images