Coal and gangue recognition method based on cross-stage partial network improved target detection algorithm

Abstract

The coal and gangue recognition method based on the cross-stage partial network improved target detection algorithm is disclosed, and the steps comprise: collecting coal and gangue images in the underground working face for pretreatment to obtain a coal and gangue data set; the coal and gangue data set is expanded through a data enhancement method, and the data set is divided to obtain a training set and a test set; a cross-stage partial network improved YOLOv7 target detection algorithm is used to construct a coal and gangue recognition model; the model is trained through three times of migration learning; and the trained model is used for classification recognition and position detection of the coal and gangue test set. Through the improved YOLOv7 target detection algorithm, a new type of coal and gangue recognition model is realized, the parameter quantity of the coal and gangue recognition model is smaller, the accuracy of the coal and gangue recognition is higher, and a good balance between the recognition accuracy and speed of the model is achieved.

Description

Technical Field

[0001] This invention relates to the field of deep learning-based coal gangue image recognition technology, specifically to a coal gangue recognition method based on a cross-stage partial network improved target detection algorithm. Background Technology

[0002] Traditional coal and gangue sorting relies mainly on manual and equipment sorting. Manual sorting is not only time-consuming and labor-intensive, but also has significant uncertainties in the results, while equipment sorting is prone to environmental pollution. With the rapid development of artificial intelligence technology, object detection algorithms in the field of computer vision have provided technical support for the accurate and rapid identification of coal and gangue. Image-based coal and gangue identification is a non-contact identification method. After the coal mining machine cuts the coal face, the falling coal is transported by scraper conveyors or belt conveyors. By acquiring real-time images of the coal and gangue on the conveyors, the coal and gangue can be tracked and identified.

[0003] The low illumination and high dust environment of underground coal mine working faces easily leads to the problem of high similarity and low recognition of coal and gangue features. Commonly used target detection algorithms have low accuracy in recognizing unclear coal and gangue images and lack in-depth extraction of feature maps at different scales, making it difficult to effectively identify small target coal and gangue. Summary of the Invention

[0004] To address the aforementioned technical shortcomings, the purpose of this invention is to provide a coal gangue identification method based on a cross-stage partial network improved target detection algorithm. This method improves the network structure of existing target detection algorithms to enhance the model's feature extraction capability while reducing the number of algorithm parameters, thereby achieving more accurate and faster identification of coal gangue.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0006] This invention provides a coal gangue identification method based on an improved target detection algorithm using a cross-stage partial network, comprising:

[0007] S100: Coal and gangue images collected from the underground working face are preprocessed to obtain a coal and gangue dataset;

[0008] S200: Expand the coal gangue dataset using data augmentation methods, and divide the dataset to obtain training and test sets;

[0009] S300: A coal gangue identification model is constructed by improving the YOLOv7 target detection algorithm with a cross-stage partial network;

[0010] S400: The model is trained using three rounds of transfer learning;

[0011] S500: Uses a trained model to classify, identify, and detect the location of coal gangue on a test set.

[0012] Preferably, the coal gangue image acquisition and preprocessing method in step S100 is as follows:

[0013] Coal and gangue images were collected using a 1:1 ratio of coal and gangue, with different placement positions, shooting distances, and light intensities.

[0014] Obtain the height and width of the captured image, and resize the image to 640 pixels × 640 pixels.

[0015] Preferably, the data augmentation method and the annotation of the coal gangue dataset in step S200 include expanding the coal gangue dataset and enhancing the expression of coal gangue features by using image size scaling, rotation, pixel translation, contrast adjustment, perspective transformation, and noise addition, and using Labelimg image annotation software to annotate the coal gangue, and dividing the training set and test set according to a 9:1 ratio.

[0016] Preferably, step S300 employs a cross-stage partial network to improve the YOLOv7 object detection algorithm, specifically as follows:

[0017] Design the CSPEM module to replace the ELAN and MP modules in the YOLOv7 target detection algorithm backbone network;

[0018] The input image is processed by four CBS modules to extract features at different scales and expand the number of channels to obtain the basic feature map;

[0019] The CSPEM module performs feature segmentation on the basic feature map at a segmentation ratio of 0.25 to obtain Seg-one and Seg-two feature maps;

[0020] The Seg-one feature map is subjected to deep feature extraction by the ELAN-MP module and then gradient segmentation by the Transition module.

[0021] Finally, it is fused with the Seg-two feature map to achieve cross-stage splitting and merging;

[0022] A recursive pyramid module is used to replace upsampling and downsampling in the neck feature extraction network;

[0023] A switchable convolution module is used to replace the ordinary convolution in the neck feature extraction network to deeply extract features at different scales and perform sufficient feature fusion.

[0024] Preferably, in steps S400 and S500, three transfer learning operations are used to train the model, including a public dataset, a laboratory-made coal gangue dataset, an underground working face coal gangue dataset, and classification, identification, and location detection of the coal gangue in the test set.

[0025] The beneficial effects of this invention are as follows: This method implements a novel coal gangue identification model by improving the YOLOv7 target detection algorithm. This coal gangue identification model has fewer parameters while achieving higher accuracy in coal gangue identification, thus achieving a good balance between the model's identification accuracy and speed. The advantages of this application and the problems it can solve are clearly stated in light of the shortcomings and existing problems mentioned in the background art. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 A flowchart illustrating a coal gangue identification method based on a cross-stage partial network improved target detection algorithm provided in an embodiment of the present invention;

[0028] Figure 2 This is a schematic diagram of the structure of a cross-stage partial network provided in an embodiment of the present invention;

[0029] Figure 3 This is a schematic diagram of the backbone network of the YOLOv7 target detection algorithm based on a cross-stage partial network provided in an embodiment of the present invention;

[0030] Figure 4 A flowchart illustrating the improved YOLOv7 target detection algorithm based on a cross-stage partial network, provided in an embodiment of the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] like Figures 1 to 3 As shown, a coal gangue identification method based on a cross-stage partial network-based improved target detection algorithm includes the following steps:

[0033] S100: Coal and gangue images collected from the underground working face are preprocessed to obtain a coal and gangue dataset;

[0034] S200: Expand the coal gangue dataset using data augmentation methods, and divide the dataset to obtain training and test sets;

[0035] S300: A coal gangue identification model is constructed by improving the YOLOv7 target detection algorithm with a cross-stage partial network;

[0036] S400: The model is trained using three rounds of transfer learning;

[0037] S500: Uses a trained model to classify, identify, and detect the location of coal gangue on a test set.

[0038] In this embodiment of the invention, the image enhancement processing specifically includes:

[0039] Coal and gangue images were collected using a 1:1 ratio of coal to gangue, with different placement positions, shooting distances, and light intensities.

[0040] Obtain the height and width of the captured image, and resize the image to 640 pixels × 640 pixels;

[0041] In this embodiment of the invention, the data augmentation method and the annotation of the coal gangue dataset specifically include using methods such as image size scaling, rotation, pixel translation, contrast adjustment, perspective transformation, and noise addition to expand the coal gangue dataset and enhance the expression of coal gangue features, and using Labelimg image annotation software to annotate the coal gangue, and dividing the training set and test set according to a 9:1 ratio.

[0042] like Figure 2-4 As shown, in this embodiment, a cross-stage partial network is used to improve the YOLOv7 target detection algorithm, specifically as follows:

[0043] Design the CSPEM module to replace the ELAN and MP modules in the YOLOv7 target detection algorithm backbone network;

[0044] The input image is processed by four CBS modules to extract features at different scales and expand the number of channels to obtain the basic feature map;

[0045] The CSPEM module performs feature segmentation on the basic feature map at a segmentation ratio of 0.25 to obtain Seg-one and Seg-two feature maps;

[0046] The Seg-one feature map is subjected to deep feature extraction by the ELAN-MP module and then gradient segmentation by the Transition module.

[0047] Finally, it is fused with the Seg-two feature map to achieve cross-stage splitting and merging;

[0048] A recursive pyramid module is used to replace upsampling and downsampling in the neck feature extraction network;

[0049] A switchable convolution module is used to replace the ordinary convolution in the neck feature extraction network to deeply extract features at different scales and perform sufficient feature fusion.

[0050] The model was trained using three transfer learning methods, including a public dataset, a laboratory-made coal gangue dataset, an underground working face coal gangue dataset, and a test set of coal gangue for classification, identification, and location detection.

[0051] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A coal and gangue recognition method based on a cross-stage partial network improved target detection algorithm, characterized in that, The method comprises the following steps: S100: collecting coal gangue images at a working face underground to obtain a coal gangue dataset through preprocessing; S200: expanding the coal gangue dataset through a data enhancement method, and dividing the dataset to obtain a training set and a test set; S300: constructing a coal gangue recognition model by improving a YOLOv7 target detection algorithm through a cross-stage partial network; S400: training the model through three times of transfer learning; S500: classifying and recognizing the test set of coal gangue and detecting the position of the test set of coal gangue by using the trained model; In step S300, the YOLOv7 target detection algorithm is improved through a cross-stage partial network, specifically as follows: A CSPEM module is designed to replace the ELAN module and the MP module in the backbone network of the YOLOv7 target detection algorithm; An input image is subjected to feature extraction at different scales and expansion of the number of channels through four CBS modules to obtain a basic feature map; The CSPEM module performs feature segmentation on the basic feature map at a segmentation ratio of 0.25 to obtain Seg-one and Seg-two feature maps; The Seg-one feature map is subjected to deep feature extraction through an ELAN-MP module and gradient segmentation through a Transition module; Finally, the Seg-one feature map is fused with the Seg-two feature map to realize cross-stage splitting and merging; A recursive pyramid module is used to replace up-sampling and down-sampling in the neck feature extraction network; A switchable convolution module is used to replace ordinary convolution in the neck feature extraction network to deeply extract features at different scales and sufficiently fuse the features.

2. The coal and gangue identification method based on cross-stage partial network improved target detection algorithm according to claim 1, characterized in that, The coal gangue image collection and preprocessing method in step S100 is specifically as follows: Coal and gangue images are collected in a 1:1 ratio by using different combinations of placement positions, photographing distances and light intensities; The height and width of the collected images are obtained, and the size of the images is adjusted to 640 pixels x 640 pixels.

3. The coal and gangue identification method based on cross-stage partial network improved target detection algorithm according to claim 1, characterized in that, The data enhancement method in step S200 and the labeling of the coal gangue dataset comprise the following steps:

4. The coal and gangue identification method based on cross-stage partial network improved target detection algorithm according to claim 1, characterized in that, The coal gangue dataset is expanded and the expression of the coal gangue features is enhanced through image size scaling, rotation, pixel translation, contrast change, perspective transformation and noise addition, and the coal gangue is labeled by using a Labelimg image labeling software, and the training set and the test set are divided in a ratio of 9:

1. In steps S400 and S500, the model is trained through three times of transfer learning, including a public dataset, a laboratory self-made coal gangue dataset, a coal gangue dataset at a working face underground and classification and recognition of the test set of coal gangue and detection of the position of the test set of coal gangue.

Images