Mine hoisting steel wire rope damage identification method and device based on improved YOLOv8

Abstract

The embodiment of the invention provides a mine hoisting steel wire rope damage identification method and device based on improved YOLOv8. The method comprises the steps that an initial damage image of a mine steel wire rope is obtained; performing denoising processing on the initial damaged image by adopting a C-BM3D denoising method to obtain a denoised damaged image; performing enhancement processing on the denoised damaged image by adopting a homomorphic filtering enhancement method to obtain an enhanced damaged image; performing data expansion processing on the enhanced damaged image to obtain a target damaged image; and based on an improved YOLOv8 model, carrying out damage identification on the target damage image to obtain a damage identification result of the mine steel wire rope. Therefore, image noise can be effectively removed, and important structure information can be reserved; the homomorphic filtering technology is combined to further enhance the contrast and definition of the image, so that the damaged area is more obvious; the accuracy of a subsequent recognition algorithm is improved, damage can be effectively detected under a complex background, and faster real-time detection is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of visual damage recognition, and in particular to, but not limited to, a mine hoisting wire rope damage recognition method and device based on improved YOLOv8. Background Art

[0002] Wire rope is the key load-bearing component of mine hoist, and its stability and reliability are crucial to the safety of hoist. Due to the harsh working conditions of wire rope, long-term heavy-load operation, curling and other physical collisions caused by wire breakage, wear and other damage, when the wire breakage and wear increase to a certain amount, it is easy to cause the entire wire rope to break, causing immeasurable production accidents.

[0003] At present, wire rope damage detection has become a research hotspot. Domestic and foreign scholars have mainly studied detection methods such as x-ray detection, acoustic detection, eddy current detection, electromagnetic detection and visual detection. Among them, electromagnetic detection and visual detection are detection methods with relatively high feasibility for wire rope damage. The electromagnetic detection method can realize the detection of internal and external damage of wire ropes, but this method is easily affected by the swing of wire ropes, the magnetization of wire ropes causes malfunction of switches, and is difficult to implement. It is usually used for online detection of high-speed running state of wire ropes greater than 6m / s.

[0004] In view of the problems existing in electromagnetic detection of wire ropes, with the development of computer vision technology and deep learning technology, image recognition has become a research hotspot for online monitoring of high-speed operation status of wire ropes. Therefore, studying the intelligent recognition method of damaged wire rope images in mine hoisting is of great significance to ensure the safety of mine hoisting. The existing single-stage recognition algorithms such as YOLO for wire rope damage recognition have a relatively fast recognition speed but a low accuracy rate. The two-stage recognition algorithms such as Fast RCNN have a relatively high accuracy rate but a slow detection speed. Summary of the invention

[0005] Based on the problems in the related art, an embodiment of the present invention provides a method and device for identifying damage of a mine hoisting wire rope based on improved YOLOv8.

[0006] The technical solution of the embodiment of the present invention is achieved as follows:

[0007] In a first aspect, an embodiment of the present invention provides a mine hoisting wire rope damage identification method based on improved YOLOv8, the method comprising:

[0008] Obtain the initial damage image of mine wire rope;

[0009] Using a C-BM3D denoising method to denoise the initial damaged image to obtain a denoised damaged image;

[0010] Using a homomorphic filtering enhancement method to enhance the denoised damaged image to obtain an enhanced damaged image;

[0011] Performing data expansion processing on the enhanced damage image to obtain a target damage image;

[0012] Based on the improved YOLOv8 model, damage identification is performed on the target damaged image to obtain the damage identification result of the mine wire rope; wherein the improved YOLOv8 model is constructed in the following manner: introducing depthwise separable convolution into the initial YOLOv8 model to obtain a YOLOv8 model with improved damage identification speed; introducing a multi-scale attention mechanism into the backbone extraction network in the YOLOv8 model with improved damage identification speed to obtain the improved YOLOv8 model.

[0013] In a second aspect, an embodiment of the present invention provides a mine hoisting wire rope damage identification device based on improved YOLOv8, the device comprising:

[0014] An acquisition module, used for obtaining an initial damage image of a mine wire rope;

[0015] A denoising module, used for denoising the initial damaged image by using a C-BM3D denoising method to obtain a denoised damaged image;

[0016] An enhancement module, used for performing enhancement processing on the denoised damaged image by adopting a homomorphic filtering enhancement method to obtain an enhanced damaged image;

[0017] An expansion module, used for performing data expansion processing on the enhanced damage image to obtain a target damage image;

[0018] The recognition module is used to perform damage recognition on the target damage image based on an improved YOLOv8 model to obtain a damage recognition result of the mine wire rope; wherein the improved YOLOv8 model is constructed in the following manner: introducing a depthwise separable convolution into an initial YOLOv8 model to obtain a YOLOv8 model with improved damage recognition speed; introducing a multi-scale attention mechanism into a backbone extraction network in the YOLOv8 model with improved damage recognition speed to obtain the improved YOLOv8 model.

[0019] In some embodiments, the denoising module is further used to group and stack similar 2D image blocks in the initial damaged image to form a 3D array by adopting a block matching combination method; sequentially performing collaborative fixed threshold filtering, noise separation and inverse 3D transformation on the 3D array to obtain a signal image estimate of the 3D array; calculating a weighted value of each signal image estimate, and mapping it to a corresponding position pixel in a one-to-one correspondence to obtain the denoised damaged image.

[0020] In some embodiments, the enhancement module is further used to assume that the denoised damaged image s1(x,y) is decomposed into an incident component i1(x,y) and a reflected component r1(x,y), and perform logarithmic transformation on the denoised damaged image based on a logarithmic change formula to obtain a logarithmically transformed damaged image; the logarithmically transformed damaged image Ins1(x,y) is expressed as: Ins1(x,y)=Ini1(x,y)+Inr1(x,y); perform Fourier transformation on the logarithmically transformed damaged image to obtain a first-transformed damaged image; the first-transformed damaged image s1′(x,y) is expressed as: s1′(x,y)=i1′(x,y)+r1′(x,y); wherein i1′(x,y) represents the incident component of the first-transformed damaged image, and r1′(x,y) represents the reflected component of the first-transformed damaged image; perform frequency filtering on the first-transformed damaged image to obtain a frequency The damaged image after frequency filtering is expressed as: H(x,y)s1′(x,y)=H(x,y)i1′(x,y)+H(x,y)r1′(x,y); the damaged image after frequency filtering is subjected to inverse Fourier transform to obtain a quadratically transformed image; the quadratically transformed image s1″(x,y) is expressed as: s1″(x,y)=i1″(x,y)+r1″(x,y); wherein, i1″(x ,y) represents the incident component of the damaged image after the secondary transformation, r1″(x,y) represents the reflected component of the damaged image after the secondary transformation; the image after the secondary transformation is subjected to exponential power calculation to obtain a homomorphic filtered image t(x,y); the homomorphic filtered image is expressed as: t(x,y)=exp|i1″(x,y)|exp|r1″(x,y)|; the homomorphic filtered image is normalized and enhanced to obtain the enhanced damaged image.

[0021] In some embodiments, the recognition module is also used for deep convolution: Assuming that the size of the input sample feature map is H×W and the number of channels is C in ; Convolution kernel setting: Each input channel corresponds to a convolution kernel, and the size of the convolution kernel is K×K; Convolution operation: Perform convolution operation on each of the input channels, and each convolution kernel is only calculated with its corresponding input channel to generate the corresponding channel Q of the output feature map i ;Q i =X i *K i (i=1,2,...,C in ), the number of channels of the generated output feature map is still C in , the size is H′×W′, and the output feature map shape is (N,C in,H′,W′); point-by-point convolution: the input is the output feature map of the depth convolution, and the number of channels is C in ; Convolution kernel setting: Use 1×1 convolution kernel, output channel is C out ; Convolution operation: performing a convolution operation on each of the input channels; each of the convolution kernels is only calculated on the input channel corresponding to it to generate the corresponding channel of the output feature map; The output feature map shape is (N,C out ,H′,W′).

[0022] In some embodiments, the recognition module is further used to assume that the input feature map is F, and the shape is C×H×W; where C is the number of channels, H is the height, and W is the width; channel attention mechanism: perform global average pooling and global maximum pooling on the input feature map F to obtain a pooling result; average pooling: F avg =AvgPool(F), maximum pooling: F max = MAXPool(F); Through the fully connected layer: the pooling result is processed through the fully connected layer to generate a channel attention weight; the channel attention weight is expressed as: W C =σ(FC2(ReLU(FC1(F avg ))))+FC2(ReLU(FC1(F max ))); FC1 and FC2 are two fully connected layers respectively; Apply channel attention: The channel attention weight W C Applied to the input feature map F, the channel attention output F is obtained c ; The channel attention output is expressed as: F c =W c ⊙F; spatial attention mechanism: output F for the channel attention c Aggregate the channel dimension to obtain the spatial feature map; Average pooling: F 1_avg =AvgPool(F1), maximum pooling: F 1_max =MaxPool(F1); The spatial feature maps of average pooling and maximum pooling are spliced ​​in the channel dimension to obtain a spliced ​​spatial feature map F2; The spliced ​​spatial feature map is expressed as: F2 = [F 1_avg ,F 1_max ]; Through the convolution layer, a 7×7 convolution operation is performed on the spliced ​​feature map F2 to generate the spatial attention weight W s ; The spatial attention weight is expressed as: W s =σ(Conv(F s ));where σ is the Sigmoid activation function;The spatial attention weight is applied to the channel attention output F cThe final feature output graph F is obtained out ; The final feature output graph is represented as: F out =W s eF1; where e represents element-by-element multiplication; output the final output feature map F out , the shape is C×H×W.

[0023] In a third aspect, an embodiment of the present invention provides an electronic device, comprising: a memory for storing executable instructions; and a processor for implementing the above-mentioned mine hoisting wire rope damage identification method based on improved YOLOv8 when executing the executable instructions stored in the memory.

[0024] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium storing executable instructions for causing a processor to execute the executable instructions to implement the above-mentioned mine hoisting wire rope damage identification method based on improved YOLOv8.

[0025] The invention provides a mine hoisting wire rope damage identification method and device based on improved YOLOv8. First, the initial damage image of the mine wire rope is obtained; the C-BM3D denoising method is used to denoise the initial damage image to obtain a denoised damage image; the homomorphic filtering enhancement method is used to enhance the denoised damage image to obtain an enhanced damage image; the enhanced damage image is subjected to data expansion processing to obtain a target damage image; based on the improved YOLOv8 model, the target damage image is subjected to damage identification to obtain a damage identification result of the mine wire rope. In this way, in view of the problems of harsh acquisition environment and poor quality of shooting details, the invention introduces the C-BM3D denoising algorithm, which can effectively remove image noise and retain important structural information on the one hand; it has significant advantages in image clarity and detail retention, and provides high-quality input for subsequent identification; on the other hand, the image contrast and clarity are further enhanced in combination with homomorphic filtering technology, making the damaged area more obvious; this enhancement processing helps to improve the accuracy of the subsequent identification algorithm and ensure that damage can be effectively detected even in complex backgrounds. In view of the problem of low wire rope damage recognition speed, the present invention improves the YOLOv8 algorithm and improves the model's computing efficiency; the optimized algorithm can achieve faster real-time detection while ensuring accuracy, and meet the needs of mine hoisting operations; the proposed algorithm solves the problem of low recognition accuracy caused by oil coverage on the wire rope surface and large rope strand gaps; through specially designed feature extraction and detection strategies, it can better cope with complex actual environments and improve recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flow chart of a method for obtaining an initial damage image of a mine wire rope provided by the present invention;

[0027] Figure 2 is a schematic diagram of an image acquisition system provided by the present invention;

[0028] Figure 3 It is a schematic diagram of the overall scheme provided by the present invention;

[0029] Figure 4 It is a structural diagram of the improved YOLOv8 model provided by the present invention;

[0030] Figure 5 It is a structural diagram of the depth-separable convolution module provided by the present invention;

[0031] Figure 6 It is a structural diagram of the CSAM module provided by the present invention;

[0032] Figure 7 is the improved detection layer diagram provided by the present invention;

[0033] Figure 8 It is a result diagram of the homomorphic filtering algorithm provided by the present invention;

[0034] FIG9 is a diagram showing the recognition results of different models of steel wire ropes provided by the present invention;

[0035] Fig.10 It is a field experiment diagram provided by the present invention;

[0036] Fig.11 It is a schematic diagram of the composition structure of a mine hoisting wire rope damage identification device based on improved YOLOv8 provided in an embodiment of the present invention;

[0037] Fig.12 A schematic diagram of the composition structure of a mine hoisting wire rope damage identification device based on improved YOLOv8 provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present invention. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present invention.

[0039] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict. Unless otherwise defined, all technical and scientific terms used in the embodiments of the present invention have the same meaning as those commonly understood by those skilled in the art to which the embodiments of the present invention pertain. The terms used in the embodiments of the present invention are only for the purpose of describing the embodiments of the present invention and are not intended to limit the present invention.

[0040] The following describes an exemplary application of a mine hoisting wire rope damage identification device based on improved YOLOv8 according to an embodiment of the present invention. The mine hoisting wire rope damage identification device based on improved YOLOv8 provided by an embodiment of the present invention can be implemented as a terminal or a server. In one implementation, the mine hoisting wire rope damage identification device based on improved YOLOv8 provided by an embodiment of the present invention can be implemented as various types of terminals such as laptops, tablet computers, desktop computers, and mobile devices; in another implementation, the mine hoisting wire rope damage identification device based on improved YOLOv8 provided by an embodiment of the present invention can also be implemented as a server, wherein the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks (CDN, Content Delivery Network), and big data and artificial intelligence platforms. The terminal and the server can be directly or indirectly connected by wired or wireless communication, which is not limited in the embodiment of the present invention. Next, an exemplary application of the mine hoisting wire rope damage identification device based on improved YOLOv8 when implemented as a server will be described.

[0041] The embodiment of the present invention provides a mine hoisting wire rope damage identification method based on improved YOLOv8, see Figure 1 , Figure 1 This is a flowchart of a mine hoisting wire rope damage identification method based on improved YOLOv8 provided by an embodiment of the present invention. Figure 1 The steps shown are explained.

[0042] Step S110, obtaining an initial damage image of the mine wire rope.

[0043] In some embodiments, the initial damage image of the mine wire rope is obtained by arranging two cameras symmetrically on both sides of the wire rope and then collecting them through a reasonable light source arrangement. Here, the initial damage image collected by the camera is transmitted to the image collection location via Ethernet and finally stored in the computer memory or hard disk.

[0044] Step S120, using the C-BM3D denoising method to denoise the initial damaged image to obtain a denoised damaged image.

[0045] In some embodiments, block matching and three-dimensional transformation techniques are used to denoise the initial damaged image of the wire rope, reduce the noise in the initial damaged image and retain the image details, so that the denoised damaged image is clearer and the details and edge features are more obvious.

[0046] Step S130, using a homomorphic filtering enhancement method to enhance the denoised damaged image to obtain an enhanced damaged image.

[0047] In some embodiments, a logarithmic transformation is used to separate the brightness component and the reflection component of the image, and a high-pass filter is used to suppress the illumination change (low-frequency component) while enhancing the details of the image (high-frequency component). Finally, the processed image is converted back to the spatial domain through an inverse transformation to obtain an enhanced damage image.

[0048] Step S140, performing data expansion processing on the enhanced damage image to obtain a target damage image.

[0049] In some embodiments, the target damage image refers to a damage image after denoising, homomorphic filtering and data expansion. In the present invention, the wire rope damage data set is geometrically transformed, such as flipping, cropping, deformation, scaling, brightness adjustment, etc., to expand the data set and obtain the target damage image.

[0050] Step S150, based on the improved YOLOv8 model, damage identification is performed on the target damage image to obtain a damage identification result of the mine wire rope; wherein the improved YOLOv8 model is constructed in the following manner: introducing a depthwise separable convolution into the initial YOLOv8 model to obtain a YOLOv8 model with improved damage identification speed; introducing a multi-scale attention mechanism into the backbone extraction network in the YOLOv8 model with improved damage identification speed to obtain the improved YOLOv8 model.

[0051] The invention provides a mine hoisting wire rope damage identification method and device based on improved YOLOv8. First, the initial damage image of the mine wire rope is obtained; the C-BM3D denoising method is used to denoise the initial damage image to obtain a denoised damage image; the homomorphic filtering enhancement method is used to enhance the denoised damage image to obtain an enhanced damage image; the enhanced damage image is subjected to data expansion processing to obtain a target damage image; based on the improved YOLOv8 model, the target damage image is subjected to damage identification to obtain a damage identification result of the mine wire rope. In this way, in view of the problems of harsh acquisition environment and poor quality of shooting details, the invention introduces the C-BM3D denoising algorithm, which can effectively remove image noise and retain important structural information on the one hand; it has significant advantages in image clarity and detail retention, and provides high-quality input for subsequent identification; on the other hand, the image contrast and clarity are further enhanced in combination with homomorphic filtering technology, making the damaged area more obvious; this enhancement processing helps to improve the accuracy of the subsequent identification algorithm and ensure that damage can be effectively detected even in complex backgrounds. In view of the problem of low wire rope damage recognition speed, the present invention improves the YOLOv8 algorithm and improves the model's computing efficiency; the optimized algorithm can achieve faster real-time detection while ensuring accuracy, and meet the needs of mine hoisting operations; the proposed algorithm solves the problem of low recognition accuracy caused by oil coverage on the wire rope surface and large rope strand gaps; through specially designed feature extraction and detection strategies, it can better cope with complex actual environments and improve recognition accuracy.

[0052] The wire rope damage online identification system designed by the present invention integrates an improved algorithm to realize real-time monitoring of wire rope damage. The system can timely feedback the detection results, help on-site workers to respond quickly, and reduce potential safety hazards. The system of the present invention can be widely used in the mining field to promote the improvement of production safety and efficiency. By real-time monitoring and accurate identification of wire rope damage, the risk of accidents is reduced and the safety of operators is guaranteed.

[0053] The following describes an exemplary application of an embodiment of the present invention in a practical application scenario.

[0054] This embodiment also provides a mine hoisting wire rope damage identification method based on an improved YOLOv8, which specifically includes the following steps:

[0055] Step 1: Damage image acquisition:

[0056] Step 101: an image acquisition system, such as Figure 2 As shown, Figure 2The image acquisition system camera placement diagram is shown in Figure 1. The main components include the main sheave, auxiliary sheave, two cameras, main bucket, auxiliary bucket, drum, network switch, and information processing platform. The two cameras are symmetrically arranged on both sides of the wire rope, and the camera optical axis is kept perpendicular to the wire rope axis. After a reasonable light source arrangement, the image can be acquired.

[0057] Step 102: The images captured by the camera are transmitted to the data processing platform of the ground control center via the Ethernet optical cable and the network switch; and finally stored in the computer memory or hard disk.

[0058] Step 103: Ensure a stable network connection to avoid data loss.

[0059] Step 104: regularly back up the transmitted image data to prevent data damage or loss.

[0060] In step 101 of this embodiment, two cameras are symmetrically arranged on both sides of the wire rope: wherein, a single camera can collect 180° image information of the wire rope, and two cameras can cover the circumferential area of ​​the wire rope at a time. At the same time, when the wire rope is in a cyclic operation, it will twist, so that the next time it is collected, a new area on the circumference of the wire rope can be scanned, which can meet the needs of wire rope image collection.

[0061] In the present invention, two cameras are symmetrically arranged on both sides of the wire rope, so that images can be obtained from different angles, which improves the comprehensiveness and accuracy of damage detection. This method effectively reduces the blind spots caused by a single viewing angle and improves the reliability of detection. The use of a reasonable light source configuration can significantly improve the shooting environment, reduce image noise, and improve image quality. This provides a good foundation for subsequent image processing and analysis, ensuring that details are clearly visible.

[0062] Step 2: Damage image preprocessing:

[0063] Step 201, C-BM3D denoising: The wire rope damage image is obtained according to step 102. The C-BM3D denoising method uses block matching and three-dimensional transformation technology to denoise the wire rope image, reduce the noise in the image and retain the image details, so that the image is clearer and the details and edge features are more obvious.

[0064] Step 202, homomorphic filtering: Use logarithmic transformation to separate the brightness component and reflection component of the image, suppress the illumination change (low-frequency component) through a high-pass filter, and enhance the details of the image (high-frequency component). Finally, the processed image is converted back to the spatial domain through inverse transformation to obtain an enhanced image.

[0065] Step 203, constructing a data set: Data sample source: On-site collection at coal mines, collecting images of hoisting wire ropes in normal operation to ensure the authenticity and representativeness of the samples. Laboratory simulation, simulating and photographing different types of wire rope damage based on actual damage conditions.

[0066] In this embodiment, the C-BM3D algorithm image denoising in step 201 is specifically performed as follows:

[0067] Step 211, convert the image format of the wire rope, and convert the noisy RGB image into a YUV image. The YUV color space is transformed according to the matrix of the following formula, where Y, U, and V are brightness signals and chrominance signals respectively;

[0068]

[0069] Step 212, image block grouping: extract a fixed size N×N target image block R from the wire rope noise image, search for similar blocks in the M×M area, and stack them into a 3D array. The similarity between the blocks depends on the Euclidean distance. After sorting the blocks from small to large in distance, take n blocks and combine them into a three-dimensional array. noisy (Z Xg ,Z x ) is the Euclidean distance between the target block and the search block, as follows:

[0070]

[0071] In the formula, x is the pixel, X is the image, and Z is Xg is the target block, Z x To search for blocks, is the size of the image block.

[0072] Step 213: extract the pixel combination array at the same position from the image group formed by the target block and the surrounding search blocks; take a normalized two-dimensional linear transform (2D-DCT) for both the target block and the search block, and if the transformed result is less than the hyperparameter λ 2D All components of (threshold) are set to zero using hard threshold filtering; finally, the Euclidean distance of the non-zero part is calculated as one of the subsequent weights; the calculation formula is as follows:

[0073]

[0074] Among them, γ′ is a hard threshold operation, and the threshold is λ 2Dσ where γ (x) For hard thresholding:

[0075]

[0076] For the group of graphics whose Euclidean distance is less than or equal to the threshold, first perform a two-dimensional linear transform (2D-DCT), and then use a one-dimensional Hadamard transform; the part less than the hyperparameter is set to 0 using a hard threshold filter. Hard threshold filtering And finally, the processed image block is obtained by performing 3D inverse transformation The formula is as follows:

[0077]

[0078] Step 2014, the aggregation process, using the Euclidean distance d(Z Xg ,Z x ) and block-by-block estimates The weighted number of the block estimation value is divided by the weighted number of the pixel point to obtain the basic estimated image. The second part can better clear the noise in the color image. The block group is recombined after 3D inverse transformation.

[0079] Step 2015: Wiener filtering is used in collaborative filtering instead of hard threshold filtering, see formula; six 3D arrays will be obtained in the aggregation process: three three-dimensional matrices ZY, ZU, ZV formed by the noise image and three three-dimensional matrices YY, YU, YV formed by the basic estimation;

[0080]

[0081] Step 2016: For the array of tiles that have completed the basic estimation, the array of pixels at the same position of these tiles is formed by using the coefficient Perform three-dimensional linear transformation to obtain the processed image block The formula is as follows:

[0082]

[0083] Weight The following formula:

[0084]

[0085] Step 2017: Use defined weights By the block-by-block estimate obtained in step 2013 The two are weighted and divided by the pixel xx m (x) and weight The weighted value of obtains the global basic estimate The formula is as follows:

[0086]

[0087] Complete the weighted average calculation of the tiles to obtain the global basic estimate Compared with the image after completing the basic estimation, the final estimation will better restore the details of the original image.

[0088] Step 2018: Convert the wire rope image that has finally been denoised from the YUV image format to the RGB image format:

[0089]

[0090] The specific process of homomorphic filtering image enhancement in step 202 of this embodiment is as follows:

[0091] Step 2021: Assuming that an image s1(x,y) can be decomposed into an incident component i1(x,y) and a reflected component r1(x,y), perform a logarithmic transformation on the input image: Ins1(x,y)=Ini1(x,y)+Inr1(x,y).

[0092] Step 2022, perform Fourier transform on both sides of the logarithmic transformation formula to obtain the relationship between the transformed image and the incident component and the reflected component: s1′(x,y)=i1′(x,y)+r1′(x,y).

[0093] Step 2023: Use filter H(x,y) to perform frequency filtering on (x,y): H(x,y)s1′(x,y)=H(x,y)i1′(x,y)+H(x,y)r1′(x,y).

[0094] Step 2024, perform inverse Fourier transform on equation 303 to obtain the relationship between the transformed image and the incident component and the reflected component: s1″(x,y)=i1″(x,y)+r1″(x,y).

[0095] Step 2025, take the exponential power of e on both sides of equation 304 to obtain the homomorphic filtering result t(x,y): t(x,y)=exp|i1″(x,y)|exp|r1″(x,y)|.

[0096] Step 2026: normalize and enhance the final image to improve the contour clarity.

[0097] In this embodiment, step 203 constructs a data set, and the specific process is as follows:

[0098] Step 2031, classify the damage types: Normal: 480 images, showing the appearance of the wire rope under normal operating conditions; Broken: 613 images, showing the situation of broken wires on the surface of the wire rope; Wear: 286 images, showing the characteristics of surface wear of the wire rope; Broken & Wear: 301 images, combining the situation of broken wires and wear.

[0099] Step 2032, count the number of each type of images and organize the data: classify the collected images according to the damage type to ensure that the images of each type are evenly distributed; normal: 480 images, broken wire: 613 images, worn: 286 images, broken wire and worn: 301 images, total: 1680 images.

[0100] Step 2033, data set division: according to the ratio of 8:1:1, the data set is divided into training set, validation set and test set. Specific division: training set 1344 pictures (80%), including 384 normal pictures, 490 broken wire pictures, 229 worn pictures, and 241 broken wire and worn pictures; validation set: 168 pictures (10%), including 48 normal pictures, 61 broken wire pictures, 29 worn pictures, and 30 broken wire and worn pictures; test set: 168 pictures (10%), including 48 normal pictures, 61 broken wire pictures, 29 worn pictures, and 30 broken wire and worn pictures.

[0101] Step 2034, data processing: performing uniform preprocessing steps on each type of image, including resizing, adjusting all images to the same resolution (such as 256×256 pixels); data normalization, normalizing pixel values ​​to between 0 and 1.

[0102] Step 2035, data set evaluation and verification: randomly check samples in each data set to ensure image quality and annotation accuracy. Confirm the balance of data distribution to avoid too many or too few samples of a certain type.

[0103] Step 2036, data storage and documentation: Store each type of image in different folders and name them according to the damage type. Record the detailed information of the data set, including sample source, damage type, quantity and division, for subsequent use and reference.

[0104] Step 3: Design and improve the YOLOv8 algorithm model:

[0105] Due to the problems of oil coverage and large rope gap in mine hoisting wire rope damage image recognition, resulting in low recognition accuracy and low recognition speed, this paper proposes an improved YOLOv8 model.

[0106] Step 301, deep separable convolution: To meet the requirement of real-time damage identification of mine hoisting wire ropes, DSConv is used to improve the damage identification speed. DSConv reduces the number of parameters and computational costs, and extracts more feature information with fewer parameters.

[0107] Step 302: Integrate the CSAM attention mechanism into the backbone network to strengthen the focus on key areas and targets, effectively eliminate irrelevant interference factors, improve the recognition and extraction capabilities of features, and thus improve the accuracy of wire rope damage recognition;

[0108] Step 303: Based on YOLOv8, a small target detection head with a size of 160×160 is introduced to improve detection sensitivity and reduce missed detection and false detection.

[0109] In this implementation case, the depthwise separable convolution is introduced in step 301 to improve the wire rope damage recognition speed. The specific process is as follows:

[0110] Step 3011, depth convolution: input feature map, assuming that the input feature map size is H×W (height×width), and the number of channels is C in .

[0111] Step 3012: Convolution kernel setting: each input channel corresponds to an independent convolution kernel, and the size of the convolution kernel is K×K.

[0112] Step 3013: Convolution operation: perform convolution operation on each input channel. Each convolution kernel is calculated only on its corresponding input channel to generate the corresponding channel of the output feature map.

[0113] Specific formula: Q i =X i *K i (i=1,2,...,C in ),

[0114] The number of channels of the generated output feature map is still C in , the size is H′×W′, and the output feature map shape is (N,C in ,H′,W′).

[0115] Step 3014, point-by-point convolution: Input feature map, input is the output feature map of depth convolution, the number of channels is C in .

[0116] Step 3015, convolution kernel setting, using 1×1 convolution kernel, output channel is C out .

[0117] Step 3016: Convolution operation, performing convolution operation on each input channel. Each convolution kernel is calculated only on its corresponding input channel to generate the corresponding channel of the output feature map;

[0118] Specific formula:

[0119] Step 3017, the output feature map shape is (N,C out ,H′,W′).

[0120] In this implementation case, in step 302, since the surface of the mine hoisting wire rope is covered with oil stains and the rope strand gap is large, the CSAM attention mechanism is integrated into the backbone network. The specific process is as follows:

[0121] Assume that the input feature map is F, and its shape is C×H×W, where C is the number of channels, H is the height, and W is the width.

[0122] Step 3021, channel attention mechanism: global pooling, perform global average pooling and global maximum pooling on the input feature map F to obtain two feature vectors:

[0123] Average pooling: F avg =AvgPool(F); Maximum pooling: F max =MAXPool(F).

[0124] Step 3022, through the fully connected layer: the two pooling results are processed through a series of fully connected layers (FC) to generate channel attention weights:

[0125] W C =σ(FC2(ReLU(FC1(F avg ))))+FC2(ReLU(FC1(F max ))),

[0126] Here, FC1 and FC2 are two fully connected layers, and σ is the Sigmoid activation function.

[0127] Step 3023, apply channel attention: set the channel attention weight W C Applied to the input feature map F: F1 = W c eF, where e represents element-wise multiplication.

[0128] Step 3024, spatial attention mechanism: generate spatial features and output F for channel attention c Aggregate the channel dimension to obtain the spatial feature map: Average pooling: F 1_avg =AvgPool(F1);

[0129] Max Pooling: F 1_max =MaxPool(F1).

[0130] Step 3025, concatenate feature maps, concatenate the results of average pooling and maximum pooling in the channel dimension: F2 = [F 1_avg ,F 1_max ].

[0131] Step 3026: Through the convolution layer, a 7×7 convolution operation is performed on the concatenated feature map F2 to generate the spatial attention weight: W s =σ(Conv(F s )).

[0132] Step 3027: Apply spatial attention and apply the spatial attention weight to the channel attention output F c Upper: F out =W s e F1.

[0133] Step 3028: Output feature map, and finally output feature map F out It is the feature map processed by the CSAM module and has a shape of C×H×W.

[0134] Step 4: Build a real-time detection system for wire rope damage identification:

[0135] Step 401, design and develop a damage identification and detection system. The acquisition end of the mine hoisting wire rope damage identification system includes a visual inspection system prototype, which is mainly responsible for the collection of wire rope images, and the visual inspection system terminal includes an industrial computer, a memory, a display, etc., and its main functions are to realize image processing, damage identification and information release, etc. The host computer system is responsible for the organization and coordination of the system software.

[0136] Step 402, operate the upper computer interface, the entire interface body is decorated with technological blue, and the interface can realize functions such as real-time display of wire rope surface status, camera acquisition control, damage detection control, damage information display, information release, video playback, etc.

[0137] In this embodiment, step 402 operates the upper computer interface, and the specific process is as follows:

[0138] Step 4021, first start the software system, then set the camera parameters and the initial position parameters of the wire rope, and then the detection system initializes the detection model. Finally, after the initialization is completed, the entire system will be in a state of waiting for the next instruction.

[0139] Step 4022, click the image acquisition start button to acquire the wire rope surface image information. At this time, not only can the real-time wire rope image information be displayed on the interface, but the underlying algorithm also transfers the wire rope image information to the memory, and then to the two established folders, and finally stores it in the hard disk.

[0140] Step 4023, select the video to be detected. When the detection algorithm does not detect the video, the algorithm detection ends. When the video is detected, the damage detection program based on the improved YOLOv8 model is called to carry out the damage detection test. In the process of detecting the damage, the position information will be called, and the position information, confidence and category will be marked together at the damage position of the image. The detailed information of the damage detection, such as the category, is also saved in the database. Information release is to use statistical methods to count the damage detection results, draw a bar chart of the damage results in the underlying logic of the algorithm, and display it in the entire interface. The horizontal axis of the bar chart is the wire rope damage type, and the vertical axis is the number of mine hoist wire rope damage.

[0141] Step 4024, video playback, select the video that needs to be played back and reviewed, and play it to meet the needs of manual review, post-processing and other operations.

[0142] In this implementation case, the camera used is an IDS Imaging UI-3860CP industrial camera with a resolution of 3840x2160 pixels (4K), about 8.3 million pixels, and a maximum frame rate of 60fps. The data collected by the camera is transferred to the computer via USB.

[0143] In this embodiment, the experimental environment is configured using Windows 10 operating system, Intel (R) Core (TM) i7-12700 processor, GPU using NVIDIA GeForce RTX 3080, deep learning framework PyTorch 1.12, and CUDA version 11.6.

[0144] In this example, in order to verify the feasibility of the improved method, the improved YOLOv8 is compared with the YOLOv5s, YOLOv5-CBAM, and YOLOv8 models, and the recognition results are visualized as follows: Figure 8 In the identification of broken wires in wire ropes, the YOLOv8 algorithm misdetects broken wires, while the YOLOv5s and YOLOv5-CBAM algorithms can correctly identify broken wires. The improved algorithm can not only correctly identify broken wire damage, but also has a higher recognition accuracy than the other two algorithms, with an accuracy of 0.926 and a recognition speed of 43.5 frames per second.

[0145] In order to verify the actual recognition effectiveness of the algorithm of the present invention, the improved YOLOv8 model in this paper is used to conduct recognition experiments on the actual collected mine hoisting wire ropes. Since the wire ropes at the coal mine site are currently in good condition, there is no damaged data source on site, but the recognition effect can be judged by identifying whether there is a false detection in the normal state of the wire rope. The test was conducted on 100 pictures of data collected at the coal mine site, and the recognition results are shown in Table 1. The test results show that the recognition accuracy rates of the three groups of wire ropes are 92%, 88%, and 90% respectively, and the average recognition accuracy of the method in this paper for actual mine hoisting wire ropes is 90%.

[0146] Table 1 Actual wire rope identification results in mines

[0147]

[0148] The mine hoisting wire rope damage image recognition method based on improved YOLOv8 proposed in the present invention provides an efficient and reliable new method for damage monitoring of mine hoisting wire ropes.

[0149] The present invention relates to a mine hoisting wire rope damage identification method and system based on YOLOv8, such as Figure 3 As shown, it includes four steps: data set image acquisition, image preprocessing, designing algorithm training data and building system online detection; the present invention proposes to use C-BM3D algorithm denoising and homomorphic filtering image enhancement to clarify the wire rope, and then uses labelimg to annotate the data to construct a data set; in view of the problems of oil coverage and large rope strand gap in mine hoisting wire rope damage identification resulting in low recognition accuracy and low recognition speed, a wire rope damage recognition algorithm is proposed, and its performance on a self-built data set is better than that of the mainstream target detection model and the existing wire rope damage recognition algorithm; in combination with the actual application needs of coal mines, the present invention designs a wire rope damage online recognition system, which integrates the damage recognition algorithm into the online detection software and deploys it on the detection terminal, realizes high-reliability online recognition, and overcomes the problems of low recognition efficiency.

[0150] Since the damage identification of mine hoisting wire ropes has problems such as oil coverage and large rope gaps, resulting in low recognition accuracy and low recognition speed, this paper proposes an improved YOLOv8 model. The ordinary convolution Conv in the YOLOv8 network is replaced with a more streamlined depth-separable convolution DSConv, and the CSAM module is introduced in the Backbonne part to improve the YOLOv8 model. Figure 4 shown.

[0151] like Figure 5 As shown, Figure 5Depthwise separable convolution first performs depthwise convolution on each channel, and then merges all channels into the output feature map through point-wise convolution. This method reduces the amount of computation and improves computational efficiency.

[0152] The CSAM attention mechanism structure is as follows Figure 6 As shown in the figure, Channel Attention represents channel attention, which replaces the original channel attention of CBAM with multi-scale convolutional attention, so that channel attention has multi-scale performance. First, the input image is subjected to maximum pooling and average pooling respectively to learn the salient features and overall degree information of the image, and two 1×1×C feature maps are obtained, which are then input into a two-layer neural network (MLP). Then, the features output by the MLP are summed element by element, and the channel attention feature F1 is generated through the sigmoid activation function. Finally, F1 is multiplied element by element with the input feature map F to obtain the input features required by the Spatial attention module.

[0153] The improved detection layer is as follows Figure 7 As shown in Figure 1, the model starts with the backbone network to perform preliminary feature extraction, and finally outputs the small target detection result in the fourth detection head at layer 18. The structure of the small target detection head is the same as the original detection head, which includes a detection head for target detection and a classification head for classification, which generate the accuracy and category probability of target detection respectively.

[0154] Experimental setup Figure 8 As shown, it includes a camera, a host computer, a switch, and a power supply. The host computer is used to collect images of the mine hoisting wire rope. The host computer is responsible for image processing and data storage. The switch is used for data transmission between components to ensure high-speed connection and stability between the camera and the host computer. The power supply provides a stable power supply for the entire system to ensure the normal operation of the equipment.

[0155] In order to verify the recognition effectiveness of the algorithm in this paper, the recognition results are visualized using YOLOv5s, YOLOv5_CBAM, YOL Ov8 and the algorithm in this paper, as shown in Figure 9, (a) is the original image, (b) is the YOLOv5s algorithm, (c) is the YOLOv5-CBAM algorithm, (d) is the YOLOv8 algorithm, and (e) is the algorithm in this paper. Analyzing the visualization results in Figure 9, we can get: from the first column of Figure 9, in the identification of broken wires in wire ropes, the YOLOv8 algorithm misdetects broken wires, while the YOLOv5s, YOLOv5-CBAM and the algorithm in this paper can all correctly identify broken wires, and the algorithm in this paper can not only correctly identify broken wire damage, but also has a higher recognition accuracy than the other two algorithms; from the second column of Figure 9, in the identification of wire rope wear, the algorithm in this paper has the highest recognition accuracy for the worn part, which is better than the other three algorithms; from the third column of Figure 9, in the simultaneous identification of broken wires and wear in wire ropes, YOLOv5s, YOLOv5-CBAM, YOLOv8 and the algorithm in this paper can all correctly identify broken wires, and the recognition accuracy of the algorithm in this paper is higher than that of the other three algorithms. In summary, the improved yolov8 algorithm in this invention is more accurate and comprehensive in identifying wire ropes.

[0156] In order to verify the technical effect that the present invention can produce, a field test of the method of the present invention was carried out. Fig.10 As shown in the figure, an area array camera is used to obtain wire rope damage images. Two cameras are fixed one meter away from the mine hoisting wire rope to ensure that the wire rope can be covered in all directions. The collected wire rope damage images are connected to the computer through wired transmission. Figure 8 Visual recognition results,Experimental results show that the accuracy and recognition speed of the proposed method are better, with a recognition accuracy of 92.6% and a recognition speed of 43.5 frames per second.,The proposed mine hoisting wire rope damage image recognition method based on improved YOLOv8 provides an efficient and reliable new method for mine hoisting wire rope damage recognition.

[0157] Fig.11 : is a schematic diagram of the composition structure of a mine hoisting wire rope damage identification device based on improved YOLOv8 provided in an embodiment of the present invention, such as Fig.11As shown, the mine hoisting wire rope damage identification device 1100 based on improved YOLOv8 includes: an acquisition module 1101, used to obtain the initial damage image of the mine wire rope; a denoising module 1102, used to use the C-BM3D denoising method to denoise the initial damage image to obtain a denoised damage image; an enhancement module 1103, used to use the homomorphic filtering enhancement method to enhance the denoised damage image to obtain an enhanced damage image; an expansion module 1104, used to perform data expansion on the enhanced damage image to obtain a target damage image; an identification module 1105 is used to perform damage identification on the target damage image based on an improved YOLOv8 model to obtain a damage identification result of the mine wire rope; wherein the improved YOLOv8 model is constructed by: introducing a deep separable convolution into the initial YOLOv8 model to obtain a YOLOv8 model with improved damage identification speed; introducing a multi-scale attention mechanism into the backbone extraction network in the YOLOv8 model with improved damage identification speed to obtain the improved YOLOv8 model.

[0158] In some embodiments, the denoising module is further used to group and stack similar 2D image blocks in the initial damaged image to form a 3D array by adopting a block matching combination method; sequentially performing collaborative fixed threshold filtering, noise separation and inverse 3D transformation on the 3D array to obtain a signal image estimate of the 3D array; calculating a weighted value of each signal image estimate, and mapping it to a corresponding position pixel in a one-to-one correspondence to obtain the denoised damaged image.

[0159] In some embodiments, the enhancement module is further used to assume that the denoised damaged image s1(x,y) is decomposed into an incident component i1(x,y) and a reflected component r1(x,y), and perform logarithmic transformation on the denoised damaged image based on a logarithmic change formula to obtain a logarithmically transformed damaged image; the logarithmically transformed damaged image Ins1(x,y) is expressed as: Ins1(x,y)=Ini1(x,y)+Inr1(x,y); perform Fourier transformation on the logarithmically transformed damaged image to obtain a first-transformed damaged image; the first-transformed damaged image s1′(x,y) is expressed as: s1′(x,y)=i1′(x,y)+r1′(x,y); wherein i1′(x,y) represents the incident component of the first-transformed damaged image, and r1′(x,y) represents the reflected component of the first-transformed damaged image; perform frequency filtering on the first-transformed damaged image to obtain a frequency The damaged image after frequency filtering is expressed as: H(x,y)s1′(x,y)=H(x,y)i1′(x,y)+H(x,y)r1′(x,y); the damaged image after frequency filtering is subjected to inverse Fourier transform to obtain a quadratically transformed image; the quadratically transformed image s1″(x,y) is expressed as: s1″(x,y)=i1″(x,y)+r1″(x,y); wherein, i1″(x ,y) represents the incident component of the damaged image after the secondary transformation, r1″(x,y) represents the reflected component of the damaged image after the secondary transformation; the image after the secondary transformation is subjected to exponential power calculation to obtain a homomorphic filtered image t(x,y); the homomorphic filtered image is expressed as: t(x,y)=exp|i1″(x,y)|exp|r1″(x,y)|; the homomorphic filtered image is normalized and enhanced to obtain the enhanced damaged image.

[0160] In some embodiments, the recognition module is also used for deep convolution: Assuming that the size of the input sample feature map is H×W and the number of channels is C in ; Convolution kernel setting: Each input channel corresponds to a convolution kernel, and the size of the convolution kernel is K×K; Convolution operation: Perform convolution operation on each of the input channels, and each convolution kernel is only calculated with its corresponding input channel to generate the corresponding channel Q of the output feature map i ;Q i =X i *K i (i=1,2,...,C in ), the number of channels of the generated output feature map is still C in , the size is H′×W′, and the output feature map shape is (N,C in ,H′,W′); point-by-point convolution: the input is the output feature map of the depth convolution, and the number of channels is Cin ; Convolution kernel setting: Use 1×1 convolution kernel, output channel is C out ; Convolution operation: performing a convolution operation on each of the input channels; each of the convolution kernels is only calculated on the input channel corresponding to it to generate the corresponding channel of the output feature map; The output feature map shape is (N,C out ,H′,W′).

[0161] In some embodiments, the recognition module is further used to assume that the input feature map is F, and the shape is C×H×W; where C is the number of channels, H is the height, and W is the width; channel attention mechanism: perform global average pooling and global maximum pooling on the input feature map F to obtain a pooling result; average pooling: F avg =AvgPool(F), maximum pooling: F max = MAXPool(F); Through the fully connected layer: the pooling result is processed through the fully connected layer to generate a channel attention weight; the channel attention weight is expressed as: W C =σ(FC2(ReLU(FC1(F avg ))))+FC2(ReLU(FC1(F max ))); FC1 and FC2 are two fully connected layers respectively; Apply channel attention: The channel attention weight W C Applied to the input feature map F, the channel attention output F is obtained c ; The channel attention output is expressed as: F c =W c ⊙F; spatial attention mechanism: output F for the channel attention c Aggregate the channel dimension to obtain the spatial feature map; Average pooling: F 1_avg =AvgPool(F1), maximum pooling: F 1_max =MaxPool(F1); The spatial feature maps of average pooling and maximum pooling are spliced ​​in the channel dimension to obtain a spliced ​​spatial feature map F2; The spliced ​​spatial feature map is expressed as: F2 = [F 1_avg ,F 1_max ]; Through the convolution layer, a 7×7 convolution operation is performed on the spliced ​​feature map F2 to generate the spatial attention weight W s ; The spatial attention weight is expressed as: W s =σ(Conv(F s ));where σ is the Sigmoid activation function;The spatial attention weight is applied to the channel attention output F c The final feature output graph F is obtained out ; The final feature output graph is represented as: Fout =W s eF1; where e represents element-by-element multiplication; output the final output feature map F out , the shape is C×H×W.

[0162] It should be noted that the description of the device of the embodiment of the present invention is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment, so it is not repeated. For technical details not disclosed in the device embodiment, please refer to the description of the method embodiment of the present invention for understanding.

[0163] It should be noted that, in the embodiment of the present invention, if the above-mentioned mine hoisting wire rope damage identification method based on improved YOLOv8 is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a terminal to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), disk or optical disk, etc. Various media that can store program codes. In this way, the embodiment of the present invention is not limited to any specific combination of hardware and software.

[0164] Correspondingly, an embodiment of the present invention provides a mine hoisting wire rope damage identification device based on improved YOLOv8. Fig.12 : is a schematic diagram of the composition structure of a mine hoisting wire rope damage identification device based on improved YOLOv8 provided in an embodiment of the present invention, such as Fig.12 As shown, the mine hoisting wire rope damage identification device 1200 based on improved YOLOv8 at least includes: a processor 1201 and a computer-readable storage medium 1202 configured to store executable instructions, wherein the processor 1201 generally controls the overall operation of the mine hoisting wire rope damage identification device 1200 based on improved YOLOv8. The computer-readable storage medium 1202 is configured to store instructions and applications executable by the processor 1201, and can also cache data to be processed or processed by each module in the processor 1201 and the mine hoisting wire rope damage identification device 1200 based on improved YOLOv8, which can be implemented by flash memory (FLASH) or random access memory (Random Access Memory, RAM).

[0165] An embodiment of the present invention provides a storage medium storing executable instructions, wherein the executable instructions are stored. When the executable instructions are executed by a processor, the processor will be caused to execute the method provided by the embodiment of the present invention, for example, Figure 1 The method shown.

[0166] In some embodiments, the storage medium can be a computer-readable storage medium, for example, a ferroelectric random access memory (FRAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory, a magnetic surface memory, an optical disk, or a compact disk read-only memory (CD-ROM), etc.; it can also be various devices including one or any combination of the above memories.

[0167] In some embodiments, executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0168] As an example, executable instructions may, but need not necessarily, correspond to a file in a file system, may be stored as part of a file storing other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files storing one or more modules, subroutines, or code portions). As an example, executable instructions may be deployed to be executed on one electronic device, or on multiple electronic devices located at one location, or on multiple electronic devices distributed at multiple locations and interconnected by a communication network.

[0169] The above description is only an embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement and improvement made within the spirit and scope of the present invention are included in the protection scope of the present invention.

[0170] It should be understood that "one embodiment" or "an embodiment" mentioned throughout the specification means that specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present invention. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present invention, the size of the serial number of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention. The serial numbers of the above-mentioned embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments.

[0171] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method or device. In the absence of further restrictions, an element defined by the statement "comprises one..." does not exclude the presence of other identical elements in the process, method, article or device including the element. In several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed.

[0172] The above is only an embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A mine hoisting wire rope damage identification method based on improved YOLOv8, characterized in that: The method comprises: Obtain the initial damage image of mine wire rope; Using a C-BM3D denoising method to denoise the initial damaged image to obtain a denoised damaged image; Using a homomorphic filtering enhancement method to enhance the denoised damaged image to obtain an enhanced damaged image; Performing data expansion processing on the enhanced damage image to obtain a target damage image; Based on the improved YOLOv8 model, damage identification is performed on the target damaged image to obtain the damage identification result of the mine wire rope; wherein the improved YOLOv8 model is constructed in the following manner: introducing depthwise separable convolution into the initial YOLOv8 model to obtain a YOLOv8 model with improved damage identification speed; introducing a multi-scale attention mechanism into the backbone extraction network in the YOLOv8 model with improved damage identification speed to obtain the improved YOLOv8 model.

2. The method according to claim 1, characterized in that The method of using the C-BM3D denoising method to denoise the initial damaged image to obtain a denoised damaged image includes: Using a block matching combination method, similar 2D image blocks in the initial damage image are grouped and stacked to form a 3D array; performing coordinated fixed threshold filtering, noise separation and inverse 3D transformation on the 3D array in sequence to obtain a signal image estimation of the 3D array; The weighted value of each signal image estimation is calculated and mapped to the corresponding position pixel in a one-to-one correspondence to obtain the denoised damage image.

3. The method according to claim 1, characterized in that The method of using a homomorphic filtering enhancement method to enhance the denoised damaged image to obtain an enhanced damaged image includes: Assuming that the denoised damaged image s1(x, y) is decomposed into an incident component i1(x, y) and a reflected component r1(x, y), the denoised damaged image is logarithmically transformed based on a logarithmic change formula to obtain a logarithmically transformed damaged image; the logarithmically transformed damaged image Ins1(x, y) is expressed as: Ins1(x, y) = Ini1(x, y) + Inr1(x, y); Performing Fourier transformation on the damage image after logarithmic transformation to obtain a damage image after primary transformation; the damage image after primary transformation s1′(x, y) is expressed as: s1′(x, y)=i1′(x, y)+r1′(x, y); wherein i1′(x, y) represents the incident component of the damage image after primary transformation, and r1′(x, y) represents the reflected component of the damage image after primary transformation; Perform frequency filtering on the damaged image after the primary transformation to obtain a damaged image after frequency filtering; the damaged image after frequency filtering is expressed as: H(x,y)s1′(x,y)=H(x,y)i1′(x,y)+H(x,y)r1′(x,y); Performing inverse Fourier transform on the damaged image after frequency filtering to obtain a secondary transformed image; the secondary transformed image s1″(x, y) is expressed as: s1″(x, y)=i1″(x, y)+r1″(x, y); wherein i1″(x, y) represents the incident component of the damaged image after secondary transform, and r1″(x, y) represents the reflected component of the damaged image after secondary transform; Performing exponential power calculation on the image after the secondary transformation to obtain a homomorphic filtering image t(x, y); the homomorphic filtering image is expressed as: t(x, y) = exp|i1″(x, y)|exp|r1″(x, y)|; The homomorphic filtered image is normalized and enhanced to obtain the enhanced damaged image.

4. The method according to claim 1, characterized in that: The depthwise separable convolution is introduced into the backbone of the initial YOLOv8 model to obtain a YOLOv8 model with improved damage recognition speed, including: Deep convolution: Assume that the input sample feature map size is H×W and the number of channels is C in ; Convolution kernel setting: Each input channel corresponds to a convolution kernel, and the size of the convolution kernel is K×K; Convolution operation: Perform convolution operation on each input channel, and each convolution kernel is calculated only with its corresponding input channel to generate the corresponding channel Q of the output feature map i ;Q i =X i *K i (i=1,2,...,C in ), the number of channels of the generated output feature map is still C in , the size is H′×W′, and the output feature map shape is (N,C in ,H′,W′); Point-by-point convolution: The input is the output feature map of the depth convolution, and the number of channels is C in ; Convolution kernel settings: Use a 1×1 convolution kernel and output channels of C out ; Convolution operation: performing a convolution operation on each of the input channels; each of the convolution kernels is only calculated on the input channel corresponding to it to generate the corresponding channel of the output feature map; The output feature map shape is (N,C out ,H′,W′).

5. The method according to claim 1, characterized in that The multi-scale attention mechanism is introduced into the backbone extraction network in the YOLOv8 model with improved damage identification speed to obtain the improved YOLOv8 model, including: Assume that the input feature map is F, and its shape is C×H×W, where C is the number of channels, H is the height, and W is the width; Channel attention mechanism: perform global average pooling and global maximum pooling on the input feature map F to obtain the pooling result; average pooling: F avg =AvgPool(F), maximum pooling: F max =MAXPool(F); Through the fully connected layer: the pooling result is processed through the fully connected layer to generate a channel attention weight; the channel attention weight is expressed as: W C =σ(FC2(ReLU(FC1(F avg ))))+FC2(ReLU(FC1(F max ))); Among them, FC1 and FC2 are two fully connected layers respectively; Apply channel attention: The channel attention weight W C Applied to the input feature map F, the channel attention output F is obtained c ; The channel attention output is expressed as: F c =W c ⊙F; Spatial attention mechanism: Attention output F for the channel c Aggregate the channel dimension to obtain the spatial feature map; Average pooling: F 1_avg =AvgPool(F1), maximum pooling: F 1_max =MaxPool(F1); The spatial feature maps of average pooling and maximum pooling are spliced ​​in the channel dimension to obtain a spliced ​​spatial feature map F2; the spliced ​​spatial feature map is expressed as: F2 = [F 1_avg ,F 1_max ]; Through the convolution layer, a 7×7 convolution operation is performed on the spliced ​​feature map F2 to generate the spatial attention weight W s ; The spatial attention weight is expressed as: W s =σ(Conv(F s ));where σ is the Sigmoid activation function; Apply the spatial attention weights to the channel attention output F c The final feature output graph F is obtained out ; The final feature output graph is represented as: F out =W s e F1; where e represents element-by-element multiplication; Output the final output feature map F out , the shape is C×H×W.

6. A mine hoisting wire rope damage identification device based on improved YOLOv8, characterized in that: The device comprises: An acquisition module, used for obtaining an initial damage image of a mine wire rope; A denoising module, used for denoising the initial damaged image by using a C-BM3D denoising method to obtain a denoised damaged image; An enhancement module, used for performing enhancement processing on the denoised damaged image by adopting a homomorphic filtering enhancement method to obtain an enhanced damaged image; An expansion module, used for performing data expansion processing on the enhanced damage image to obtain a target damage image; The recognition module is used to perform damage recognition on the target damage image based on an improved YOLOv8 model to obtain a damage recognition result of the mine wire rope; wherein the improved YOLOv8 model is constructed in the following manner: introducing a depthwise separable convolution into an initial YOLOv8 model to obtain a YOLOv8 model with improved damage recognition speed; introducing a multi-scale attention mechanism into a backbone extraction network in the YOLOv8 model with improved damage recognition speed to obtain the improved YOLOv8 model.

Images