Novel mining belt tearing segmentation model and degree detection method

Abstract

The invention relates to a novel mining belt tearing segmentation model and degree detection method. The system comprises a mining belt image processing module which is used for collecting and processing a tearing image of a belt in the operation process in real time; the novel lightweight feature extraction algorithm is used for enhancing the recognition capability of a crack in a belt image and extracting tearing features of different shapes; large selection kernel high-efficiency channel attention is used for capturing tearing characteristics of different scale changes and fully extracting semantic information of a belt context; the pixel attention guidance fusion module learns useful crack features from context branches, and the useful crack features cannot be submerged by massive detail information; and the tearing degree detection module is used for calculating the tearing area of the mining belt tearing image and determining the tearing degree. The method has the advantages that the belt tearing and segmenting accuracy is improved, and the method is more intelligent.

Description

Technical Field

[0001] The invention relates to the field of mine vision, and in particular to a novel mining belt tearing segmentation model and degree detection method. Background Art

[0002] During mining transportation, the belt is prone to tearing due to the large size of ore, the impact of sharp objects such as gangue and metal foreign matter. When a large area of ​​the belt is torn, it will not only cause material spillage and blockage, but also reduce production efficiency and economic costs. In serious cases, it will endanger the lives of production personnel. However, in the field of traditional belt tear detection, it mainly relies on on-site visual inspection by workers. This method has high labor costs, takes a long time, is prone to misdetection, and interferes with belt production. In addition to manual visual inspection, other traditional methods include mechanical switches, ultrasonic waves, radio frequency identification, and X-ray detection, but these methods have the limitations of low intelligence and poor accuracy. The above defects are what technical personnel in this field hope to overcome. Summary of the invention

[0003] 1. Technical issues to be resolved

[0004] In view of the existing technical problems, the present invention provides a new mining belt tearing and segmentation model and degree detection method, which has the advantages of high accuracy, intelligence and labor cost saving.

[0005] (II) Technical solution

[0006] In order to achieve the above-mentioned purpose, the main technical solutions adopted by the present invention are as follows.

[0007] A new mining belt tear segmentation model and degree detection method, including:

[0008] A mining belt image processing module (1) collects and processes the tearing images of the belt in real time during operation;

[0009] A new lightweight feature extraction algorithm (2) that identifies the tear in the belt image and extracts tear features of different shapes while reducing the computational cost;

[0010] Large selection kernel efficient channel attention (3), analyzes and captures tear features of different scales and fully understands the semantic information of the belt context;

[0011] The pixel attention-guided fusion module (4) selectively learns useful gap features from the context branch without being overwhelmed by massive details;

[0012] The tearing degree detection module (5) calculates the tearing area of ​​the mining belt tearing image and determines the tearing degree.

[0013] A novel mining belt tear segmentation model and degree detection method as described above comprises the following steps:

[0014] A. The mining belt image processing module (1) collects and processes the torn images of the mining belt in real time, and sends the processed torn images of the mining belt to the novel lightweight feature extraction algorithm (2);

[0015] B. A novel lightweight feature extraction algorithm (2) performs feature processing on the torn images of mining belts, including identifying the cracks in the belt images and extracting tear features of different shapes, and sends the generated feature maps to the large selection kernel efficient channel attention (3);

[0016] C. Large selection kernel efficient channel attention (3) analyzes the feature map of the mining belt tear image, including capturing the tear features of different scales and fully understanding the semantic information of the belt context, and sends the semantic information to the pixel attention guided fusion module (4);

[0017] D. The pixel attention guided fusion module (4) receives the semantic information from the large selection kernel efficient channel attention (3), selectively learns useful tear features from the semantic information, generates a prediction map of tear segmentation, and sends the prediction map to the tear degree detection module (5);

[0018] E. The tearing degree detection module (5) counts the number of pixel points of each category in the tear segmentation prediction image, calculates the tearing area, and determines the tearing degree by the tearing area ratio;

[0019] Furthermore, step A specifically includes:

[0020] A1. Annotate the extracted mining belt tear images and create corresponding ground truth labels;

[0021] A2. Randomly rotate the extracted mining belt tear images and the corresponding ground truth labels simultaneously;

[0022] A3. Noise is added to the extracted mining belt torn image, and the corresponding ground truth label remains unchanged;

[0023] A4. Blurring the extracted mining belt tearing image, and keeping the corresponding ground truth label unchanged;

[0024] A5. The extracted mining belt tear image is subjected to brightness changes, and the corresponding ground truth labels remain unchanged;

[0025] The A2-A5 regimens can be performed together or separately.

[0026] Further, step B specifically includes:

[0027] B1. Use the first convolutional layer to transform the channel of the torn image of the mining belt to form a primary feature map;

[0028] B2. Use a two-layer Mamba-like encoder to perform feature transformation on the primary feature map generated in step B1 to extract local features and form a secondary feature map;

[0029] B3. Use the BN layer to normalize and linearly transform the secondary feature map generated in step B2 to achieve faster reasoning;

[0030] B4. Use a nonlinear activation function, such as ReLU, on the information passed in step B3 to increase nonlinearity.

[0031] Further, step C specifically includes:

[0032] C1. The feature map generated by the new lightweight feature extraction algorithm (2) is used to extract the channel weight and spatial weight in parallel through the efficient channel mechanism and the large selection kernel mechanism;

[0033] C2. The channel weight and spatial weight of step C1 are fused and scaled to the interval (0,1) by Sigmoid as the final feature weight;

[0034] C3. Perform element-wise dot multiplication with the input features of step C1 to obtain the output features of the belt torn image.

[0035] Further, step D specifically includes:

[0036] D1. Use detail branches to parse and preserve detailed information of high-resolution feature maps output from the previous decoder block;

[0037] D2. Use context branches to aggregate local and global information to resolve long-distance dependencies;

[0038] D3. Let the detail branch selectively learn useful semantic tearing features from the context branch without being overwhelmed by massive detail information.

[0039] Further, step E specifically includes:

[0040] E1. When the torn area is larger than the untorn area, that is, P degree When >1, we can assume that the belt in the image is severely torn and it is recommended to replace the belt immediately;

[0041] E2. When the torn area is less than or equal to the untorn area, that is, 0.5 <P degree When ≤1, we can assume that the belt in the image has a moderate tear, and it is recommended to repair the belt immediately;

[0042] E3. When the torn area is slightly smaller than the untorn area, that is, 0 <P degree When the value is ≤0.5, we believe that the belt in the image is slightly torn and we recommend that the belt be repaired at a later time.

[0043] Among them, the tearing degree P degree The calculation formula is as follows:

[0044]

[0045] In the formula, S 撕裂的 Refers to the torn belt area, S 完好的 Refers to the intact belt area.

[0046] The present invention also provides a storage medium, which includes a stored program, wherein when the program is run, the mining belt tear segmentation model and degree detection method is executed.

[0047] (III) Beneficial effects

[0048] The beneficial effects of the present invention are as follows: the present invention proposes a mining belt tear segmentation model and degree detection method. The novel lightweight feature extraction algorithm uses a two-layer Mamba-like encoder to improve the segmentation efficiency of the tear area. The large selection kernel efficient channel attention provides the most suitable receptive field for large convolution kernels of different sizes from the two dimensions of channel and space to capture the local and global features of belt tears with different scales. The pixel attention guided fusion module alleviates the problem of receptive field mismatch between shallow convolution and deep convolution to a certain extent and reduces spatial redundancy. At the same time, compared with other detection methods, this method has high segmentation accuracy, fast detection speed and is more intelligent. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a schematic diagram of a specific implementation mode of the present invention. DETAILED DESCRIPTION

[0050] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation modes in conjunction with the accompanying drawings.

[0051] Reference Figure 1 , a specific embodiment of the present invention includes,

[0052] A new mining belt tear segmentation model and degree detection method, including:

[0053] A mining belt image processing module (1) collects and processes the tearing images of the belt in real time during operation;

[0054] A new lightweight feature extraction algorithm (2) that identifies the tear in the belt image and extracts tear features of different shapes while reducing the computational cost;

[0055] Large selection kernel efficient channel attention (3), analyzes and captures tear features of different scales and fully understands the semantic information of the belt context;

[0056] The pixel attention-guided fusion module (4) selectively learns useful gap features from the context branch without being overwhelmed by massive details;

[0057] The tearing degree detection module (5) calculates the tearing area of ​​the mining belt tearing image and determines the tearing degree through the tearing area ratio.

[0058] A novel mining belt tear segmentation model and degree detection method as described above comprises the following steps:

[0059] A. The mining belt image processing module (1) collects and processes the torn images of the mining belt in real time, and sends the processed torn images of the mining belt to the novel lightweight feature extraction algorithm (2);

[0060] B. A novel lightweight feature extraction algorithm (2) performs feature processing on the torn images of mining belts, including identifying the cracks in the belt images and extracting tear features of different shapes, and sends the generated feature maps to the large selection kernel efficient channel attention (3);

[0061] C. Large selection kernel efficient channel attention (3) analyzes the feature map of the mining belt tear image, including capturing the tear features of different scales and fully understanding the semantic information of the belt context, and sends the semantic information to the pixel attention guided fusion module (4);

[0062] D. The pixel attention guided fusion module (4) receives the semantic information from the large selection kernel efficient channel attention (3), selectively learns useful tear features from the semantic information, generates a prediction map of tear segmentation, and sends the prediction map to the tear degree detection module (5);

[0063] E. The tearing degree detection module (5) counts the number of pixel points in the tear segmentation prediction image, calculates the tearing area, and determines the tearing degree by the tearing area ratio.

[0064] In specific implementation, as a preferred embodiment of the present invention, step A specifically includes:

[0065] A1. Annotate the extracted mining belt tear images and create corresponding ground truth labels;

[0066] A2. Randomly rotate the extracted mining belt tear images and the corresponding ground truth labels simultaneously;

[0067] A3. Noise is added to the extracted mining belt torn image, and the corresponding ground truth label remains unchanged;

[0068] A4. Blurring the extracted mining belt tearing image, and keeping the corresponding ground truth label unchanged;

[0069] A5. The extracted mining belt tear image is subjected to brightness changes, and the corresponding ground truth labels remain unchanged;

[0070] The A2-A5 regimens can be performed together or separately.

[0071] In step B, the novel lightweight feature extraction algorithm (2) performs feature processing on the torn image of the mining belt, and further comprises the following steps:

[0072] B1. Use the first convolutional layer to transform the channel of the torn image of the mining belt to form a primary feature map;

[0073] B2. Use a two-layer Mamba-like encoder to perform feature transformation on the primary feature map generated in step B1 to extract local features and form a secondary feature map;

[0074] B3. Use the BN layer to normalize and linearly transform the secondary feature map generated in step B2 to achieve faster reasoning;

[0075] B4. Use a nonlinear activation function, such as ReLU, on the information passed in step B3 to increase nonlinearity.

[0076] In specific implementation, as a preferred embodiment of the present invention, step C specifically includes:

[0077] In step C, the large selection kernel efficient channel attention (3) analyzes the feature map of the mining belt tearing image and further includes the following steps:

[0078] C1. The feature map generated by the new lightweight feature extraction algorithm (2) is used to extract the channel weight and spatial weight in parallel through the efficient channel mechanism and the large selection kernel mechanism;

[0079] C2. The channel weight and spatial weight of step C1 are fused and scaled to the interval (0,1) by Sigmoid as the final feature weight;

[0080] C3. Perform element-wise dot multiplication with the input features of step C1 to obtain the output features of the belt torn image.

[0081] In the specific implementation, as a preferred embodiment of the present invention, step D specifically includes:

[0082] D1. Use detail branches to parse and preserve detailed information of high-resolution feature maps output from the previous decoder block;

[0083] D2. Use context branches to aggregate local and global information to resolve long-distance dependencies;

[0084] D3. Let the detail branch selectively learn useful semantic tearing features from the context branch without being overwhelmed by massive detail information.

[0085] In specific implementation, as a preferred embodiment of the present invention, step E specifically includes:

[0086] E1. When the torn area is larger than the untorn area, that is, P degree When >1, we can assume that the belt in the image is severely torn and it is recommended to replace the belt immediately;

[0087] E2. When the torn area is less than or equal to the untorn area, that is, 0.5 <P degree When ≤1, we can assume that the belt in the image has a moderate tear, and it is recommended to repair the belt immediately;

[0088] E3. When the torn area is slightly smaller than the untorn area, that is, 0 <P degree When the value is ≤0.5, we believe that the belt in the image is slightly torn and we recommend that the belt be repaired at a later time.

[0089] Among them, the tearing degree P degree The calculation formula is as follows:

[0090]

[0091] In the formula, S 撕裂的 Refers to the torn belt area, S 完好的 Refers to the intact belt area.

[0092] The embodiment of the present application also discloses a computer-readable storage medium, which stores a computer instruction set. When the computer instruction set is executed by a processor, the mining belt tear segmentation model and degree detection method provided in any of the above embodiments is implemented.

[0093] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0094] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0095] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0096] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied 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 computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A new mining belt tear segmentation model and degree detection method, characterized in that: include: A mining belt image processing module (1) collects and processes the tearing images of the belt in real time during operation; A new lightweight feature extraction algorithm (2) that identifies the tear in the belt image and extracts tear features of different shapes while reducing the computational cost; Large selection kernel efficient channel attention (3), analyzes and captures tear features of different scales and fully understands the semantic information of the belt context; The pixel attention-guided fusion module (4) selectively learns useful gap features from the context branch without being overwhelmed by massive details; The tearing degree detection module (5) calculates the tearing area of ​​the mining belt tearing image and determines the tearing degree through the tearing area ratio.

2. The novel mining belt tearing and segmentation model and degree detection method as claimed in claim 1 is characterized in that The following steps are involved: A. The mining belt image processing module (1) collects and processes the torn images of the mining belt in real time, and sends the processed torn images of the mining belt to the novel lightweight feature extraction algorithm (2); B. A novel lightweight feature extraction algorithm (2) performs feature processing on the torn images of mining belts, including identifying the cracks in the belt images and extracting tear features of different shapes, and sends the generated feature maps to the large selection kernel efficient channel attention (3); C. Large selection kernel efficient channel attention (3) analyzes the feature map of the mining belt tear image, including capturing the tear features of different scales and fully understanding the semantic information of the belt context, and sends the semantic information to the pixel attention guided fusion module (4); D. The pixel attention guided fusion module (4) receives the semantic information from the large selection kernel efficient channel attention (3), selectively learns useful tear features from the semantic information, generates a prediction map of tear segmentation, and sends the prediction map to the tear degree detection module (5); E. The tearing degree detection module (5) counts the number of pixel points in the tear segmentation prediction image, calculates the tearing area, and determines the tearing degree by the tearing area ratio.

3. The novel mining belt tearing and segmentation model and degree detection method as claimed in claim 2 is characterized in that: In step A, the mining belt image processing module (1) processes the image of the torn mining belt including the following schemes: A1. Annotate the extracted mining belt tear images and create corresponding ground truth labels; A2. Randomly rotate the extracted mining belt tear images and the corresponding ground truth labels simultaneously; A3. Noise is added to the extracted mining belt torn image, and the corresponding ground truth label remains unchanged; A4. Blurring the extracted mining belt tearing image, and keeping the corresponding ground truth label unchanged; A5. The extracted mining belt tear image is subjected to brightness changes, and the corresponding ground truth labels remain unchanged; The A2-A5 regimens can be performed together or separately.

4. The novel mining belt tearing and segmentation model and degree detection method as claimed in claim 2 is characterized in that: In step B, the novel lightweight feature extraction algorithm (2) performs feature processing on the torn image of the mining belt, and further comprises the following steps: B1. Use the first convolutional layer to transform the channel of the torn image of the mining belt to form a primary feature map; B2. Use a two-layer Mamba-like encoder to perform feature transformation on the primary feature map generated in step B1 to extract local features and form a secondary feature map; B3. Use the BN layer to normalize and linearly transform the secondary feature map generated in step B2 to achieve faster reasoning; B4. Use a nonlinear activation function, such as ReLU, on the information passed in step B3 to increase nonlinearity.

5. The novel mining belt tearing and segmentation model and degree detection method as claimed in claim 2 is characterized in that: In step C, the large selection kernel efficient channel attention (3) analyzes the feature map of the mining belt tearing image and further includes the following steps: C1. The feature map generated by the new lightweight feature extraction algorithm (2) is used to extract the channel weight and spatial weight in parallel through the efficient channel mechanism and the large selection kernel mechanism; C2. The channel weight and spatial weight of step C1 are fused and scaled to the interval (0,1) by Sigmoid as the final feature weight; C3. Perform element-wise dot multiplication with the input features of step C1 to obtain the output features of the belt torn image.

6. The novel mining belt tearing and segmentation model and degree detection method as claimed in claim 2 is characterized in that: In step D, the pixel attention guided fusion module (4) receives the semantic information from the large selection kernel efficient channel attention (3), selectively learns useful tear features from the semantic information, generates a prediction map of tear segmentation, and sends the prediction map to the tear degree detection module (5), which includes the following steps: D1. Use detail branches to parse and preserve detailed information of high-resolution feature maps output from the previous decoder block; D2. Use context branches to aggregate local and global information to resolve long-distance dependencies; D3. Let the detail branch selectively learn useful semantic tearing features from the context branch without being overwhelmed by massive detail information.

7. The novel mining belt tearing and segmentation model and degree detection method as claimed in claim 2 is characterized in that: In step E, the tearing degree detection module (5) counts the number of pixel points of each category in the tear segmentation prediction image and calculates the tearing area. Determining the tearing degree by the tearing area ratio includes the following steps: E1. When the torn area is larger than the untorn area, that is, P degree When >1, we can assume that the belt in the image is severely torn and it is recommended to replace the belt immediately; E2. When the torn area is less than or equal to the untorn area, that is, 0.5 <P degree When ≤1, we can assume that the belt in the image has a moderate tear, and it is recommended to repair the belt immediately; E3. When the torn area is slightly smaller than the untorn area, that is, 0 <P degree When the value is ≤0.5, we believe that the belt in the image is slightly torn and we recommend that the belt be repaired at a later time. Among them, the tearing degree P degree The calculation formula is as follows: In the formula, S 撕裂的 Refers to the torn belt area, S 完好的 Refers to the intact belt area.

8. A storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is run, the method according to any one of claims 1 to 7 is executed.

Images