Conveyor belt damage detection method and system

Abstract

The invention provides a conveyor belt damage detection method and system, and relates to the field of artificial intelligence, and the method comprises the steps: collecting conveyor belt damage images, and making a data set for training; based on the YOLO model, utilizing a C2PSAACmix module to enhance the extraction capability of global features and local features of the YOLO model, and utilizing an iRMB module to optimize an information flow transmission mechanism of the YOLO model so as to obtain a conveyor belt damage detection model; inputting a to-be-detected conveyer belt image into the trained conveyer belt damage detection model to obtain a segmentation result; based on the segmentation result, determining that the damage category is epidermis wear, hole or fracture; judging the fracture direction of the conveying belt by using an ellipse fitting algorithm; and determining a damage area and damage pixels according to the detection frame of the segmentation result so as to carry out quantitative analysis, and determining the damage degree and / or the fracture expansion trend of the conveyor belt according to the result of the quantitative analysis. According to the invention, different types of conveyor belt damages can be automatically identified and evaluated, and the possibility of manual intervention and operation errors is reduced.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a conveyor belt damage detection method and system. Background Art

[0002] Belt conveyors are important continuous transportation equipment in modern industrial production and are widely used in coal, mining, ports, electricity, chemical and other fields. They have the advantages of long transportation distance, large transportation capacity and high efficiency. As an important component of belt conveyors, the condition of the conveyor belt directly affects the overall operating efficiency and safety performance of the equipment. However, due to the long-term operation of conveyor belts at high speeds, heavy loads and complex environments, they are very prone to damage, including surface wear, holes and cracks. If these damages are not detected and treated in time, they will not only accelerate the aging of the conveyor belt and affect the normal operation of the equipment, but also lead to serious consequences such as downtime for maintenance and production interruptions, causing economic losses to the company, and even causing safety accidents such as longitudinal tearing and belt breakage. In recent years, image processing methods based on deep learning have been widely used, especially convolutional neural networks (CNN) and target detection algorithms, which have made significant progress in many visual inspection fields. However, existing deep learning networks still face technical challenges when dealing with small target damage and complex backgrounds, such as insufficient information flow processing capabilities and insufficient ability to extract global and local features. Furthermore, existing detection methods can only output image segmentation or classification results, lacking quantitative evaluation mechanisms and unable to form directly applicable detection conclusions. Relying on manual judgment of segmentation results can lead to subjective errors. Therefore, a new and efficient damage detection method is urgently needed to improve the accuracy and robustness of conveyor belt damage detection. Summary of the Invention

[0003] The present invention aims to solve at least one of the technical problems existing in the prior art or related art, and discloses a conveyor belt damage detection method and system, which improves the accuracy and robustness of conveyor belt damage detection, can automatically identify and evaluate different types of conveyor belt damage, and reduce the possibility of manual intervention and operational errors.

[0004] The first aspect of the present invention discloses a conveyor belt damage detection method, comprising: preparing training materials: collecting conveyor belt damage images to prepare a training data set; constructing a conveyor belt damage detection model: based on the YOLO model, using the C2PSA_ACmix module to enhance the global feature and local feature extraction capabilities of the YOLO model, using the iRMB module to optimize the information flow transmission mechanism of the YOLO model, capturing long-distance dependencies while keeping the model lightweight, so as to obtain a conveyor belt damage detection model; inputting the training data set into the conveyor belt damage detection model for training to obtain a trained conveyor belt damage detection model; image segmentation: inputting the conveyor belt image to be detected into the trained conveyor belt damage detection model to obtain a segmentation result; damage assessment: based on the segmentation result, determining the damage category as surface wear, hole or fracture; using an ellipse fitting algorithm to determine the fracture direction of the conveyor belt; determining the damage area and damaged pixels according to the detection frame of the segmentation result for quantitative analysis, and determining the damage degree and / or fracture extension trend of the conveyor belt according to the result of the quantitative analysis.

[0005] In this technical solution, two major challenges must be overcome to effectively detect conveyor belt damage, especially the three different types of damage: fractures, surface wear, and holes. First, surface wear is a small target, usually with fewer pixels and weaker feature representation, resulting in limited representative feature information. Second, fractures and holes are larger targets, but in a complex background, especially in the presence of multiple other damages, the damage may be confused with the surrounding environment, thereby increasing the difficulty of extracting its unique feature information. To solve these problems, the attention mechanism can be integrated into the feature extraction process. The present invention improves the traditional YOLO network structure and uses the C2PSA_ACmix module to enhance the global and local feature extraction capabilities of the conveyor belt damage detection and segmentation network, thereby improving the detection and segmentation performance of small target damage and complex damage.

[0006] According to the conveyor belt damage detection method disclosed in the present invention, preferably, the quantitative analysis specifically includes:

[0007] The damage category is determined to be surface wear or holes, and the damaged pixels are surface wear pixels or hole pixels. The damage degree of the conveyor belt is determined based on the damage ratio. The damage ratio is the ratio of the number of damaged pixels to the number of pixels in the damaged area. The damaged area is a part of the conveyor belt image and includes the detection frame. The damage category is determined to be fracture. The fracture extension trend is determined based on the fracture direction. Based on the fracture direction and the damage naming convention, it is subdivided into longitudinal tear and transverse fracture. If the fracture direction is determined to be longitudinal, the damage degree is determined based on the damage height ratio. The damage height ratio is the ratio of the detection frame height at the fracture to the conveyor belt image height. If the fracture direction is determined to be transverse, the damage degree is determined based on the damage width ratio. The damage width ratio is the ratio of the detection frame width at the fracture to the conveyor belt image width.

[0008] According to the conveyor belt damage detection method disclosed in the present invention, preferably, it also includes: when the damage proportion is less than or equal to 10%, determining that the conveyor belt is slightly damaged; when the damage proportion is greater than 10% and less than or equal to 30%, determining that the conveyor belt is moderately damaged; when the damage proportion is greater than 30%, determining that the conveyor belt is severely damaged; when the damage height proportion is less than or equal to 10%, determining that the conveyor belt is slightly longitudinally torn; when the damage height proportion is greater than 10% and less than or equal to 30%, determining that the conveyor belt is moderately longitudinally torn; when the damage height proportion is greater than or equal to 30%, determining that the conveyor belt is severely longitudinally torn; when the damage width proportion is less than or equal to 10%, determining that the conveyor belt is slightly transversely broken; when the damage width proportion is greater than 10% and less than or equal to 30%, determining that the conveyor belt is moderately transversely broken; when the damage width proportion is greater than or equal to 30%, determining that the conveyor belt is severely transversely broken.

[0009] According to the conveyor belt damage detection method disclosed in the present invention, preferably, the step of determining the fracture direction of the conveyor belt using an ellipse fitting algorithm specifically includes: extracting the contour of the fracture area through edge detection; fitting an ellipse model according to the contour of the fracture area, and then calculating the inclination angle of the fracture; determining the direction of the fracture by judging the angle value: if the inclination angle is within the range of [45°, 135°], it is regarded as a longitudinal tear; if the inclination angle is outside the range of [45°, 135°], it is regarded as a transverse fracture.

[0010] According to the conveyor belt damage detection method disclosed in the present invention, preferably, the conveyor belt damage detection model includes the backbone network, neck network and detection head of the YOLO model, and the C2PSA_ACmix module and the iRMB module are added to the backbone network for feature extraction.

[0011] According to the conveyor belt damage detection method disclosed in the present invention, preferably, the C2PSA_ACmix module includes: a channel and position space attention module C2PSA and a self-attention and convolution hybrid module ACmix, wherein C2PSA is composed of a pyramid split attention PSA and a 1×1 convolution, and PSA is composed of a self-attention mechanism Attention and a feedforward neural network FNN; ACmix is combined with the feedforward neural network FNN in PSA, and the hybrid attention mechanism of ACmix is used to replace the self-attention mechanism Attention in PSA, so as to help the conveyor belt damage detection model focus on the damaged area, learn the distribution pattern of features, and recalibrate the focus according to the specific circumstances of the damage.

[0012] According to the conveyor belt damage detection method disclosed in the present invention, preferably, the iRMB module includes an extended window multiple self-focusing EW-MHSA and a depthwise separable convolution DWConv.

[0013] According to the conveyor belt damage detection method disclosed in the present invention, preferably, the YOLO model is a YOLOv11 model.

[0014] According to the conveyor belt damage detection method disclosed in the present invention, preferably, the step of preparing training materials further includes: performing Mosaic data enhancement on the images in the data set and adjusting the image size to a uniform size.

[0015] The second aspect of the present invention discloses a conveyor belt damage detection system, comprising: a memory for storing program instructions; a processor for calling the program instructions stored in the memory to implement the conveyor belt damage detection method as any of the above technical solutions.

[0016] Compared with the prior art, the beneficial effects of the present invention include at least:

[0017] (1) The traditional YOLO structure is optimized, and the network's ability to segment and detect small target damage and complex damage is improved. It can achieve efficient and accurate conveyor belt damage detection and is suitable for real-time monitoring in complex industrial environments;

[0018] (2) Combining the damage assessment method based on geometric features and size quantification, the fracture direction is determined by the ellipse fitting algorithm, and the damage morphology is quantitatively analyzed, providing a scientific basis for the subsequent fracture extension trend prediction and damage severity assessment;

[0019] (3) It can automatically identify and evaluate different types of conveyor belt damage, reducing the possibility of manual intervention and operational errors, and providing efficient and reliable technical support for intelligent detection and fault warning of conveyor belts, which has significant application prospects and economic value. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic flow chart of a conveyor belt damage detection method according to an embodiment of the present invention is shown.

[0021] Figure 2 FIG. 4 shows a schematic structural diagram of an iRMB module according to an embodiment of the present invention.

[0022] Figure 3 A schematic diagram of the C2PSA_ACmix module structure according to an embodiment of the present invention is shown.

[0023] Figure 4 A schematic diagram of ellipse fitting according to an embodiment of the present invention is shown.

[0024] Figure 5 FIG. 4 shows a schematic diagram of a damaged area according to an embodiment of the present invention.

[0025] Figure 6 A schematic block diagram of a conveyor belt damage detection system according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0026] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0027] like Figure 1 According to one embodiment of the present invention, a method for detecting conveyor belt damage is disclosed, comprising:

[0028] Step S1, preparing training materials: collecting conveyor belt damage images to prepare a training data set;

[0029] Step S2: Constructing a conveyor belt damage detection model: Based on the YOLO model, the C2PSA_ACmix module is used to enhance the YOLO model's ability to extract global and local features. The iRMB module is used to optimize the YOLO model's information flow transmission mechanism, capturing long-range dependencies while maintaining the model's lightweightness, thereby obtaining a conveyor belt damage detection model. The training dataset is input into the conveyor belt damage detection model for training, thereby obtaining a trained conveyor belt damage detection model.

[0030] Step S3, image segmentation: input the conveyor belt image to be detected into the trained conveyor belt damage detection model to obtain a segmentation result;

[0031] Step S4, damage assessment: Based on the segmentation results, determine the damage category as surface wear, hole or fracture; use the ellipse fitting algorithm to determine the fracture direction of the conveyor belt; determine the damaged area and damaged pixels based on the detection frame of the segmentation result for quantitative analysis, and determine the damage degree and / or fracture extension trend of the conveyor belt based on the results of the quantitative analysis.

[0032] like Figures 2 to 5 According to another embodiment of the present invention, the specific implementation process and working principle of the conveyor belt damage detection method provided in the above embodiment are also disclosed: This embodiment provides a conveyor belt damage detection segmentation network (conveyor belt damage detection model) and evaluation method based on an improved YOLOv11. The purpose is to improve the accuracy, real-time and adaptability of conveyor belt damage detection through an optimized network structure and a new damage assessment mechanism, and provide an effective solution for intelligent detection of conveyor belt damage in complex environments. Specifically, it includes:

[0033] The damaged conveyor belt images captured by the camera in actual working conditions are compiled into a dataset. Mosaic data enhancement is performed on the images in the dataset, and the images in the dataset are uniformly resized to 640*640 to obtain the preprocessed image I.

[0034] The conveyor belt damage detection model is based on an optimized and improved version of YOLOv11. In the feature extraction module of the conveyor belt damage detection and segmentation network N (conveyor belt damage detection model), the C2PSA_ACmix module leverages the advantages of traditional convolution and self-attention mechanisms to enhance the global and local feature extraction capabilities of the conveyor belt damage detection and segmentation network N, improving the detection and segmentation performance of small and complex damage targets. The iRMB module improves information flow processing and captures long-range dependencies while maintaining the network's lightweightness, thereby improving the efficiency and accuracy of segmentation tasks.

[0035] Conveyor belt damage includes fracture, surface wear and holes. Wear is a small target, while fracture is a large target. Traditional detection methods face many challenges in dealing with these types of damage, especially in dealing with long-distance dependencies and complex background environments. To solve this problem, this embodiment improves the original YOLOv11 network structure and uses iRMB modules (such as Figure 2As shown) improves the efficiency and accuracy of the segmentation task. iRMB consists of extended window multiple autofocus EW-MHSA and depthwise separable convolution DW convolution (DWConv), combining the advantages of global modeling and local information fusion. EW-MHSA enables the model to accurately capture the characteristics of different types of damage such as small wear and large-scale fractures in the image when processing conveyor belt damage, especially in complex backgrounds, effectively avoiding the confusion of small targets such as wear and background. DW convolution improves processing efficiency by reducing the amount of calculation, ensuring that the conveyor belt damage detection and segmentation network N can respond quickly in practical applications. The combined advantages of EW-MHSA and DW make iRMB particularly suitable for practical applications in conveyor belt damage detection. It can efficiently and accurately detect and segment small damage and large target damage, while solving the problem of long-distance dependence.

[0036] To effectively detect conveyor belt damage, particularly the three different types of damage: fractures, surface wear, and holes, two key challenges must be overcome. First, surface wear is a small object, typically with fewer pixels and weaker feature representation, resulting in limited representative feature information. Second, fractures and holes are larger objects, but in complex backgrounds, especially when there are multiple other types of damage, the damage may be confused with the surrounding environment, making it more difficult to extract their unique feature information. To address these issues, attention mechanisms can be integrated into the feature extraction process.

[0037] This embodiment improves the traditional YOLOv11 network structure and uses the C2PSA_ACmix module (such as Figure 3 (As shown) enhances the global and local feature extraction capabilities of the conveyor belt damage detection and segmentation network N, and improves the detection and segmentation performance of small target damage and complex damage. The C2PSA_ACmix module consists of channel and position space attention C2PSA and self-attention and convolution mixed module ACmix. The C2PSA module consists of pyramid split attention PSA and 1×1 convolution. The PSA module consists of self-attention mechanism Attention and feedforward neural network FNN. The ACmix module is combined with the feedforward neural network FNN in the PSA module. ACmix is divided into two stages. In the initial stage, the H×W×C features are projected three times with 1×1 convolution to reorganize the input features into N segments, thereby obtaining a feature subset consisting of 3×N feature maps. In the next stage, the feature subset obtained from the previous stage is input into two different branches. The upper branch adopts a convolution path with a kernel size of k to collect information from the local receptive field. A lightweight fully connected layer is then used to convert the features into k 2Feature maps. These generated features are shifted, aggregated, and convolved to generate H×W×C feature maps. At the same time, the lower-level branches use self-attention to incorporate global information. The intermediate features are divided into N groups, each consisting of 3 feature maps: query, key, and value. These grouped features are processed by a multi-head self-attention model. After that, the features are further processed by shifting, aggregation, and convolution to generate H×W×C feature maps. Finally, the outputs of the two branches are weighted and summed, and the weights are determined by two learnable scalars as shown in the following formula:

[0038] F out =αF conv +βF att

[0039] Among them, F out represents the final output of the path; F att represents the output of the self-attention branch; F conv Represents the output of the convolutional attention branch. α and β are both set to 1. The outputs of the two branches are merged together, considering both global and local features, thereby enhancing the detection and segmentation performance of the conveyor belt damage detection and segmentation network N for small objects.

[0040] In order to accurately assess the damage of the conveyor belt material or structure, this embodiment designs a damage assessment method based on geometric features and size quantification, with particular emphasis on the determination of the fracture direction and the extraction of relevant quantitative indicators. This method combines the ellipse fitting algorithm with the quantization operation based on the segmentation mask, and can efficiently and accurately perform damage detection, segmentation, and assessment. In the fracture morphology analysis, the accurate determination of the fracture direction is crucial for the subsequent damage assessment. To this end, the ellipse fitting algorithm is used to determine the fracture direction (e.g. Figure 4 Ellipse fitting is a geometric morphology-based algorithm that fits the pixel points along the fracture edge to determine the fracture's primary direction and geometric features. Specifically, the ellipse fitting algorithm first extracts the fracture's contour through edge detection. It then uses the least squares method to fit an ellipse model, which then calculates the fracture's tilt angle. The tilt angle is accurate to two decimal places, and the fracture direction is determined by determining the angle. If the angle is within the range [45°, 135°], it is considered a longitudinal tear; if the angle is outside this range, it is considered a transverse fracture.

[0041] Assume that the set of broken edge points in the image is P = {(x1, y1), (x2, y2), ..., (x n ,y n )}, where n is the number of edge points. The general equation of an ellipse is:

[0042]

[0043] Where (h, k) is the center of the ellipse, and a and b are the semi-axis lengths of the ellipse on the x-axis and y-axis, respectively.

[0044] In order to fit the ellipse P, the parameters (h, k, a, b) are optimized by the least squares method, and the objective function E is the sum of squared errors:

[0045]

[0046] Taking the partial derivative of the objective function and setting it equal to zero, we get the following system of equations:

[0047]

[0048] The Gauss-Seidel method is used to iteratively approximate the optimal solution. The fracture edge fitting in the image data is completed and the ellipse parameters (h, k, a, b) are obtained. The tilt angle θ of the ellipse can be calculated by the ratio of the semi-axis lengths a and b:

[0049]

[0050] The advantages of this method lie in its strong adaptability and accuracy. Especially for complex or irregular fracture morphologies, the ellipse fitting algorithm can provide a stable geometric model, thereby helping to determine the fracture propagation trend. Through this logic, the embodiment can effectively distinguish different types of fracture morphologies, providing a reliable basis for accurate damage assessment.

[0051] To further quantify the size and morphological characteristics of the damaged area on the conveyor belt, a quantization operation based on a segmentation mask is used to extract the damaged area and perform a series of calculation operations. Through the image segmentation algorithm, the precise area of the conveyor belt damage can be obtained, and the relevant geometric features can be calculated based on this. The quantitative indicators mainly include the following eight quantitative indicators: detection frame height, detection frame width, damage height ratio, damage width ratio, damage category pixels, damage area pixels, damage ratio, and fracture angle. The damage height ratio, damage width ratio, and damage ratio are as follows:

[0052] Damage height ratio = detection frame height / image height × 100%;

[0053] Damage width ratio = detection frame width / image height × 100%;

[0054] Damage ratio = damaged pixels / damaged area pixels × 100%.

[0055] like Figure 5 As shown, the damaged pixel is Figure 5 The damaged pixels are marked in the figure. The damage category is hole. The damaged area is the area obtained by cropping the image according to the dotted line (detection frame height) in the figure.

[0056] Combining the extracted geometric features and size quantification results with the set threshold, the severity of the damage is determined:

[0057] If the damage type is epidermal abrasion and hole, the damage degree is determined by the damage ratio. The damage ratio refers to the ratio of the number of pixels in the damage type to the number of pixels in the damage area, which is used to measure the extent of the damage:

[0058] Injury percentage ≤10% was defined as mild injury;

[0059] 10%<injury ratio≤20% is defined as moderate injury;

[0060] Injury ratio > 20% is defined as severe injury.

[0061] If the damage type is fracture, the damage extent is determined based on the damage height ratio / damage width ratio:

[0062] Longitudinal tear:

[0063] The injury height accounted for ≤10%, which was defined as mild longitudinal tear;

[0064] 10% < Injury height ≤ 30%, defined as moderate longitudinal tear;

[0065] The injury height ratio was greater than 30%, which was defined as severe longitudinal tear.

[0066] Transverse fracture:

[0067] The damage width accounts for ≤10%, which is defined as mild transverse fracture;

[0068] 10%<damage width accounts for ≤30%, which is defined as moderate transverse fracture;

[0069] The lesion width ratio was greater than 30%, which was defined as severe transverse fracture.

[0070] According to another embodiment of the present invention, a specific application scenario of the conveyor belt damage detection model disclosed in the above embodiment is also disclosed:

[0071] Step 1, data preparation: The conveyor belt damage images used are all collected from cameras in actual working conditions. The data set contains 2471 images, the training set includes 1722 images, the validation set includes 254 images, and the test set includes 495 images. It is divided into three categories: surface wear, holes and fractures. Use the Labelme annotation tool to annotate the damage in the image and store it in YOLO format. The stored content includes the label category, the coordinates of the center point of the target bounding box, and the width and height of the target bounding box. The data set is divided into training set, validation set and test set in a ratio of 7:1:2. Figure 5Shown are some training set images in the dataset.

[0072] Step 2, model training: Based on YOLOv11 as the basic architecture, the C2PSA_ACmix module is added to the feature extraction module to optimize the backbone network, and the iRMB module is introduced into the backbone network. The iRMB structure is as follows Figure 2 As shown, the C2PSA_ACmix structure is as follows Figure 3 As shown in the figure, the SGD optimizer is used for training. The batch size is adjusted to 32 based on the GPU memory and dataset size, and the number of training epochs is set to 500 to ensure that the model is fully trained. The initial learning rate is set to 0.01, the momentum is 0.937, and the weight decay is 0.0005.

[0073] Step 3, Model Evaluation: At the end of each training cycle, the model is evaluated using the validation set, and metrics such as precision (mAP), recall, and F1 score are calculated. Overfitting is checked by observing the change in loss between the training and validation sets.

[0074] Step 4: Model Testing and Results Visualization: Images from the test set are fed into the trained model. The model will output segmentation results for each image, including candidate bounding boxes, class labels, and confidence scores. The model's performance on the test set is quantified using the same evaluation metrics as the validation set. Model accuracy is evaluated using precision (P), recall (R), mean average precision (mAP), and the harmonic mean of precision and recall (F1). Model complexity and detection speed are assessed using parameter count (Params), computational load (FLOPs), and inference time (Times).

[0075] Step 5, damage assessment: The assessment is performed based on geometric features and dimensional quantification, with particular emphasis on the determination of fracture direction and the extraction of relevant quantitative indicators to obtain damage assessment results.

[0076] The conveyor belt damage detection and segmentation network N and evaluation method based on the improved YOLOv11 provided in the present invention not only have high detection accuracy and robustness, but also show good adaptability in complex industrial environments, providing an effective solution for the intelligent detection, segmentation and evaluation of conveyor belt damage.

[0077] like Figure 6 As shown, according to another embodiment of the present invention, a conveyor belt damage detection system 600 is also disclosed, including: a memory 601 for storing program instructions; a processor 602 for calling the program instructions stored in the memory to implement the conveyor belt damage detection method as described in the above embodiment.

[0078] In summary, the conveyor belt damage detection network constructed by the present invention is improved based on YOLOv11, and uses the C2PSA_ACmix module to enhance the extraction capabilities of global and local features, thereby improving the detection and segmentation performance of small target damage and complex damage. Through the iRMB module, information flow processing is improved, long-distance dependencies are captured, and the network is kept lightweight, thereby improving the efficiency and accuracy of segmentation tasks. The present invention also proposes a damage assessment method based on geometric features and size quantization. This method comprehensively evaluates the damage type of the conveyor belt damage area, uses an ellipse fitting algorithm to determine the fracture direction, uses size quantization technology to accurately analyze the damage morphology, and uses 8 quantitative indicators for comprehensive evaluation, providing a reliable assessment basis for the severity of the damage and the fracture extension trend, improving the accuracy, real-time and adaptability of conveyor belt damage detection, and providing an effective solution for intelligent detection of conveyor belt damage in complex environments.

[0079] All or part of the steps in the various methods of the above embodiments can be completed by controlling related hardware through a program. The program can be stored in a readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other readable medium that can be used to carry or store data.

[0080] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A conveyor belt damage detection method, characterized in that: include: Create training materials: Collect conveyor belt damage images and create training data sets; Constructing a conveyor belt damage detection model: Based on the YOLO model, the C2PSA_ACmix module is used to enhance the YOLO model's global and local feature extraction capabilities. The iRMB module is used to optimize the YOLO model's information flow transmission mechanism, capturing long-range dependencies while maintaining the model's lightweightness. This results in a conveyor belt damage detection model. The training dataset is input into the conveyor belt damage detection model for training, resulting in a trained conveyor belt damage detection model. Image segmentation: inputting the conveyor belt image to be detected into the trained conveyor belt damage detection model to obtain a segmentation result; Damage assessment: Based on the segmentation results, determining the damage type as epidermal abrasion, hole or fracture; The ellipse fitting algorithm is used to determine the fracture direction of the conveyor belt; the damaged area and damaged pixels are determined according to the detection frame of the segmentation result for quantitative analysis, and the damage degree and / or fracture extension trend of the conveyor belt are determined according to the results of the quantitative analysis.

2. The conveyor belt damage detection method according to claim 1, characterized in that: The quantitative analysis specifically includes: Determine the damage category as surface wear or hole, the damaged pixels as surface wear pixels or hole pixels, and determine the damage degree of the conveyor belt based on the damage ratio, where the damage ratio is the ratio of the number of damaged pixels to the number of pixels in the damaged area, where the damaged area is a portion of the conveyor belt image that includes the detection frame; Determine the damage type as fracture, and determine the fracture extension trend based on the fracture direction; If the fracture direction is determined to be longitudinal, the degree of damage is determined based on the damage height ratio, where the damage height ratio is the ratio of the detection frame height at the fracture to the conveyor belt image height; If the fracture direction is determined to be horizontal, the degree of damage is determined according to the damage width ratio, where the damage width ratio is the ratio of the detection frame width at the fracture to the conveyor belt image width.

3. The conveyor belt damage detection method according to claim 2, characterized in that: Also includes: When the damage percentage is less than or equal to 10%, the conveyor belt is determined to be slightly damaged; When the damage percentage is greater than 10% and less than or equal to 30%, the conveyor belt is determined to be moderately damaged; If the damage accounts for more than 30%, the conveyor belt is determined to be severely damaged; When the damage height accounts for less than or equal to 10%, it is determined that the conveyor belt has a slight longitudinal tear; When the damage height accounts for more than 10% and less than or equal to 30%, it is determined that the conveyor belt has moderate longitudinal tearing; When the damage height accounts for more than or equal to 30%, it is determined that the conveyor belt is severely longitudinally torn; When the damage width accounts for less than or equal to 10%, it is determined that the conveyor belt has a slight transverse fracture; When the damage width accounts for more than 10% and less than or equal to 30%, the conveyor belt is determined to have a moderate transverse fracture; When the damage width accounts for greater than or equal to 30%, it is determined that the conveyor belt has a severe transverse fracture.

4. The conveyor belt damage detection method according to claim 1, characterized in that: The step of determining the fracture direction of the conveyor belt by using an ellipse fitting algorithm specifically includes: The contour of the fracture area is extracted using an edge detection algorithm; An elliptical model is fitted according to the contour of the fracture area, and then the inclination angle of the fracture is calculated; The direction of the fracture is determined by judging the angle value: if the inclination angle is within the range of [45°, 135°], it is considered as a longitudinal tear; if the inclination angle is outside the range of [45°, 135°], it is considered as a transverse fracture.

5. The conveyor belt damage detection method according to claim 1, characterized in that: The conveyor belt damage detection model includes a backbone network, a neck network, and a detection head of a YOLO model, and the C2PSA_ACmix module and the iRMB module are both added to the backbone network for feature extraction.

6. The conveyor belt damage detection method according to claim 1, characterized in that: The C2PSA_ACmix module includes: a channel and position space attention module C2PSA and a self-attention and convolution hybrid module ACmix. Among them, C2PSA is composed of pyramid split attention PSA and 1×1 convolution, and PSA is composed of a self-attention mechanism Attention and a feedforward neural network FNN. ACmix is combined with the feedforward neural network FNN in PSA, and the hybrid attention mechanism of ACmix is used to replace the self-attention mechanism Attention in PSA. This helps the conveyor belt damage detection model focus on the damaged area, learn the distribution pattern of features, and recalibrate the focus point according to the specific situation of the damage.

7. The conveyor belt damage detection method according to claim 1, characterized in that: The iRMB module includes extended window multiple autofocus (EW-MHSA) and depthwise separable convolution (DWConv).

8. The conveyor belt damage detection method according to claim 1, characterized in that: The YOLO model is a YOLOv11 model.

9. The conveyor belt damage detection method according to claim 1, characterized in that: The step of preparing training materials further includes: Mosaic data augmentation is performed on the images in the dataset and the image dimensions are resized to a uniform size.

10. A conveyor belt damage detection system, characterized in that: include: a memory for storing program instructions; A processor, configured to call the program instructions stored in the memory to implement the conveyor belt damage detection method according to any one of claims 1 to 9.

Images