Coal gangue rapid detection method and device based on ADS-YOLOv11n model
By improving the structure of the YOLOv11n model, combined with AKConv, DySample and SimAM mechanisms, the automated and intelligent detection of coal gangue is realized, solving the problem of insufficient small object detection capabilities of the YOLO model in complex scenarios, and improving the sorting accuracy and efficiency.
Patent Information
- Application Number
- CN202510564132.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-30
- Publication Date
- 2025-08-19
AI Technical Summary
The existing YOLO model has weak detection capabilities for small targets in complex industrial scenarios, and it is difficult to accurately identify and locate coal gangue when they are blocked by each other, resulting in low efficiency and insufficient accuracy of coal gangue sorting.
The ADS-YOLOv11n model was used for improvement, including using the AKConv module to replace the ordinary convolution module in the backbone network, using the DySample dynamic upsampling module in the neck network, and introducing the SimAM attention mechanism to build a coal gangue detection system, including image acquisition, detection and result output modules.
The automated and intelligent detection of coal gangue has been realized, with the sorting accuracy exceeding 95%, which greatly reduces labor workload and reduces labor costs by 60%, and improves detection accuracy and sorting efficiency.
Smart Images
Figure CN120510099A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of computer vision and mine resource processing, and in particular to a method and device for rapid detection of coal gangue based on an ADS-YOLOv11n model. Background Art
[0002] Coal occupies a fundamental position in the global energy supply system and is an indispensable energy source and raw material for numerous industries, including power, steel, and chemicals. The effective identification and separation of coal gangue remains a critical and challenging task during coal mining and subsequent processing.
[0003] Gangue is a low-carbon, high-hardness rock deposited alongside coal seams during the coal formation process. Its composition is complex and diverse, including clay minerals, quartz, and carbonate minerals. After coal mining, raw coal is often mixed with large amounts of gangue. While gangue differs from coal in appearance, physical properties, and chemical composition, distinguishing the two in actual raw coal mixtures is challenging. Traditionally, gangue identification and separation relies primarily on manual sorting, a method with significant drawbacks. Firstly, it is inefficient. Given the large-scale coal mining volume, manual operations are unable to meet the demands of rapid production, significantly limiting coal companies' capacity expansion and economic benefits. Secondly, the accuracy of manual sorting is significantly affected by subjective factors, such as worker experience, concentration, and visual fatigue. This can easily lead to misidentification, with coal mistakenly discarded as gangue, resulting in a waste of coal resources. Alternatively, gangue can be incorporated into coal products, reducing coal quality and affecting subsequent combustion efficiency. It can also cause wear and corrosion of combustion equipment, as well as environmental pollution, increasing operating costs and environmental pressures.
[0004] With the rapid development of information technology, computer vision technology has achieved remarkable achievements in recent years. Deep learning, as one of the core technologies in the field of computer vision, provides a new solution for object detection tasks.
[0005] As an outstanding representative of deep learning target detection algorithms, the YOLO algorithm is known for its unique single-stage detection architecture. The YOLO algorithm can divide the input image into multiple grids, each of which is responsible for predicting information such as the category, location, and confidence level of the target. The target detection process for the entire image can be completed with a single forward propagation. Compared with traditional multi-stage target detection algorithms, it has the significant advantages of fast detection speed and strong real-time performance. At the same time, the YOLO algorithm's detection accuracy has been gradually improved during its continuous development and evolution. By adopting technical means such as deeper network structures, more advanced feature extraction modules, and optimized loss functions, it can accurately identify various targets in complex image scenes.
[0006] Applying the YOLO algorithm to gangue identification is expected to improve traditional gangue identification and sorting operations, achieving automated, intelligent, and efficient gangue identification. However, current YOLO models have a limited application scope for gangue sorting. Most models only achieve high accuracy in laboratory settings, lacking focus on field sorting and identification, and thus lack comprehensive online detection. Furthermore, gangue is typically small and irregularly shaped, making the YOLO model less capable of detecting small targets. Furthermore, gangue often accumulates and occludes each other, making it difficult for YOLO to accurately identify and locate targets in these conditions. Summary of the Invention
[0007] In order to solve the problems that the existing detection method based on the YOLO model has weak detection ability for small targets in complex industrial scenes and is difficult to accurately identify and locate when gangue occludes each other, the present invention proposes a method and device for rapid detection of gangue based on the ADS-YOLOv11n model.
[0008] The technical solution adopted by the present invention to solve the above problems is:
[0009] The present invention provides a method for rapid detection of coal gangue based on the ADS-YOLOv11n model, comprising the following steps:
[0010] Step 1: Collect coal gangue sorting images and perform data annotation and data enhancement processing to obtain a data set for model training;
[0011] Step 2: Improve the Yolov11n model, specifically as follows:
[0012] Step 2.1: Use AKConv convolution module to replace the ordinary convolution module in the backbone network, and reconstruct the C3K2 module with AKConv;
[0013] Step 2.2: Use the DySample dynamic upsampling module in the neck network to replace the nearest neighbor interpolation upsampling.
[0014] Step 2.3 introduces the SimAM attention mechanism;
[0015] Step 3: Build a coal gangue detection system.
[0016] Furthermore, the gangue detection system described in step 3 includes:
[0017] Image acquisition module, used to obtain real-time operation video of the gangue sorting conveyor belt;
[0018] The gangue detection module uses the ADS-YOLOv11 model to detect gangue in the collected video data.
[0019] The result output module is used to convert the position information of coal gangue into position information in the world coordinate system and output it.
[0020] Furthermore, in step 1, the data enhancement processing includes flipping, brightness adjustment, and rotation operations on the annotated image.
[0021] Furthermore, the training parameters of the ADS-YOLOv11n model include setting the input image size to 640×640, the optimizer to Adam, the number of model iterations to 100, the batch size to 16, and the learning rate to 0.01.
[0022] Furthermore, the performance of the ADS-YOLOv11n model is evaluated by using the category average accuracy, precision, recall, and F1 score as evaluation metrics.
[0023] The invention provides a rapid detection device for coal gangue based on the ADS-YOLOv11n model, comprising:
[0024] The image color collection module includes a sealed recognition box, which is equipped with an industrial camera and evenly distributed LED light sources for real-time acquisition of video data on the coal gangue conveyor belt;
[0025] The gangue detection module performs real-time detection of video streams based on the ADS-YOLOv11n model and outputs the data to the gangue grabbing system.
[0026] The result output module is used to convert the detection results into coordinate information in the world coordinate system and transmit it to the coal gangue sorting equipment.
[0027] The beneficial effects of the present invention are:
[0028] 1. This paper optimizes the Yolov11 model structure, uses the AKConv module in the backbone network to improve feature extraction capabilities, introduces DySample dynamic upsampling in the neck network to enhance small target detection, and adds the SimAM attention mechanism to optimize recognition accuracy in occlusion scenarios.
[0029] 2. Based on the optimized Yolov11 model, this invention successfully implements automated detection and sorting of coal gangue, ultimately achieving intelligent detection of coal gangue. During the entire coal gangue detection process, no human intervention is required, largely replacing manual sorting work, significantly reducing manual workload and labor costs by 60%.
[0030] 3. The present invention can reduce the risks and costs caused by manual errors or missed inspections, improve overall competitiveness and benefits, and ensure the accuracy of coal gangue detection, with a sorting accuracy rate exceeding 95%. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 This is a flow chart of the method for rapid detection of coal gangue based on the ADS-YOLOv11n model of the present invention;
[0032] Figure 2 It is a structural diagram of the identification box of the present invention;
[0033] Figure 3 、 Figure 4 This is the ADS-YOLOv11n model structure diagram of the present invention. DETAILED DESCRIPTION
[0034] The present invention provides a method, system and device for detecting coal gangue based on the ADS-YOLO11n model. The main purpose of the method is to achieve rapid and accurate identification of coal gangue. The flow chart of the coal gangue detection method and system provided by the present invention is as follows: Figure 1 shown.
[0035] This system mainly consists of three main modules: image acquisition module, gangue detection module and result output module. The following will explain each part in detail:
[0036] Image acquisition module: A sealed recognition box is installed on the gangue conveyor belt. The recognition box is equipped with an industrial camera and four evenly distributed LED light sources. The sealed recognition box can isolate the dust at the gangue selection site to a certain extent, while ensuring a certain light intensity environment inside the box. The industrial camera completes the real-time acquisition of the gangue video stream data set and transmits it to the PC host computer via USB3.0. The structure of the recognition box is as follows: Figure 2 shown.
[0037] Gangue detection module: After the PC receives the video stream, it is transmitted to the packaged ADS-YOLOv11n recognition program for recognition. The ADS-YOLOv11n model structure diagram is as follows: Figure 3 、 Figure 4 As shown in the figure, the present invention replaces the ordinary convolution module with the AKConv convolution module in the backbone network of the Yolov11n model, and reconstructs the C3K2 module with AKConv; ② replaces the nearest neighbor interpolation upsampling with the DySample dynamic upsampling module in the neck network; ③ introduces the SimAM attention mechanism.
[0038] In order to obtain the optimal weight of the gangue detection model, the present invention trains the ADS-YOLOv11n model. The process of training the ADS-YOLOv11n model mainly includes the following steps:
[0039] a. Collect images of coal gangue sorting in real coal mine scenes. For subsequent processing, the collected images are uniformly processed. Then, the collected dataset is annotated with the category and location information of coal and gangue in the images using labelimg annotation software. The annotated results are then made into a label file for subsequent training.
[0040] b. The present invention divides the annotated image data into a training set, a validation set, and a test set in a ratio of 8:1:1. At the same time, in order to enhance data diversity, improve the generalization performance of the model, and avoid overfitting of model training, the present invention uses data enhancement methods such as flipping, changing image brightness, and rotating to perform data enhancement processing on the divided data.
[0041] c. Input the training set into the ADS-YOLOv11n model for training. Set the following parameters to complete the training process: set the input image size to 640×640, and use Adam as the optimizer. Set the number of model iterations (epochs) to 100, the batch size to 16, and the learning rate to 0.01. After training, save the model weights that yield the best results.
[0042] d. Use the model weight with the best result to detect the images in the test set to verify the effect of the ADS-YOLOv11n model. At the same time, in order to better evaluate the effect of the model before and after improvement, this paper uses the mean average accuracy (mAP), precision, recall and F1 score as evaluation indicators.
[0043] mAP refers to the average of the average accuracy of all categories. The closer the result is to 1, the higher the accuracy of the model detection. The calculation formula for this indicator is:
[0044]
[0045] Where: mAP is the average of the average precision of all categories; m is the number of categories in the test sample; P is the precision of the model; R is the recall rate of the model.
[0046] Precision refers to the ratio of the sample that the target detection model judges to be positive to the sample that is actually positive. The closer the accuracy is to 1, the better the model effect.
[0047] Recall refers to the ratio of correct predictions made by the model among all actual positive samples. The closer the recall result is to 1, the better the model is.
[0048] The F1 score is the harmonic mean of precision and recall, which can better reflect the performance of the model. Its maximum value is 1 and its minimum value is 0. The closer the index result is to 1, the better the model detection accuracy. The above calculation formula is as follows:
[0049]
[0050]
[0051]
[0052] Where: P is the precision of the model; R is the recall of the model; F1 is the F1 score of the model; TP is the sample predicted by the model to be positive, but it is actually a positive sample; FP is the sample predicted by the model to be positive, but it is actually a negative sample; FN is the sample predicted by the model to be negative, but it is actually a positive sample.
[0053] Result output module: This module mainly detects the real-time video through the gangue detection module, uses a rectangular frame to mark the gangue and its location, obtains the required data, and outputs the detection results to other equipment to use other equipment to realize the sorting of gangue.
[0054] The above-described specific implementation method effectively implements a gangue target detection experiment based on the ADS-YOLOv11n model. Based on the optimized Yolov11 model, this invention successfully implements automated gangue detection and sorting, ultimately achieving intelligent gangue detection. Throughout the gangue detection process, no human intervention is required, and the accuracy of gangue detection is guaranteed.
[0055] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as a preferred embodiment as above, it is not intended to limit the present invention. Any technician familiar with the present profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical content disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent replacement and improvement of the above embodiments made according to the technical essence of the present invention, within the spirit and principles of the present invention, without departing from the content of the technical solution of the present invention, shall still fall within the scope of protection of the technical solution of the present invention.
Claims
1. A rapid detection method for coal gangue based on the ADS-YOLOv11n model, characterized in that: The method comprises the following steps: Step 1: Collect coal gangue sorting images and perform data annotation and data enhancement processing to obtain a data set for model training; Step 2: Improve the Yolov11n model, specifically as follows: Step 2.1: Use AKConv convolution module to replace the ordinary convolution module in the backbone network, and reconstruct the C3K2 module with AKConv; Step 2.2: Use the DySample dynamic upsampling module in the neck network to replace the nearest neighbor interpolation upsampling. Step 2.3 introduces the SimAM attention mechanism; Step 3: Build a coal gangue detection system.
2. The method for rapid detection of coal gangue based on the ADS-YOLOv11n model according to claim 1, characterized in that: The gangue detection system described in step 3 includes: Image acquisition module, used to obtain real-time operation video of the gangue sorting conveyor belt; The gangue detection module uses the ADS-YOLOv11 model to detect gangue in the collected video data. The result output module is used to convert the position information of coal gangue into position information in the world coordinate system and output it.
3. The method for rapid detection of coal gangue based on the ADS-YOLOv11n model according to claim 1, characterized in that: In step 1, the data enhancement process includes flipping, brightness adjustment, and rotation operations on the annotated image.
4. The method for rapid detection of coal gangue based on the ADS-YOLOv11n model according to claim 1, characterized in that: The training parameters of the ADS-YOLOv11n model include the input image size set to 640×640, the optimizer as Adam, the number of model iterations as 100, the batch size as 16, and the learning rate as 0.
01.
5. The method for rapid detection of coal gangue based on the ADS-YOLOv11n model according to claim 1, characterized in that: The performance of the ADS-YOLOv11n model is evaluated using the category average accuracy, precision, recall, and F1 score.
6. A rapid detection device for coal gangue based on the ADS-YOLOv11n model, characterized in that: The detection device comprises: The image color collection module includes a sealed recognition box, which is equipped with an industrial camera and evenly distributed LED light sources for real-time acquisition of video data on the coal gangue conveyor belt; The gangue detection module performs real-time detection of video streams based on the ADS-YOLOv11n model and outputs the data to the gangue grabbing system. The result output module converts the detection results into coordinate information in the world coordinate system and transmits it to the coal gangue sorting equipment.