Coal sorting method, system, medium and equipment

Abstract

The invention discloses a coal sorting method and system, a medium and equipment, and relates to the technical field of target detection. The method comprises the following steps: acquiring real-time mining area coal data; performing target detection on the real-time mining area coal data by using the improved coal target detection model to obtain a coal detection result; according to a preset confidence interval, retaining or eliminating coal data meeting a preset confidence condition in the coal detection result; marking and storing the reserved coal data; carrying out online training on the stored coal data through incremental learning, and carrying out parameter updating on the improved coal target detection model; performing coal target detection on newly obtained real-time mining area coal data by using the improved coal target detection model after parameter updating to obtain a real-time coal target detection result; and the coal is sorted according to a real-time coal target detection result.

Description

Technical Field

[0001] The present application relates to the field of target detection technology, and in particular to a coal sorting method, system, medium and equipment. Background Art

[0002] The background of coal identification technology involves various links such as coal mining, transportation, sorting, storage, processing and use. With the development of industrial automation and intelligence, coal identification technology has played an important role in improving coal utilization efficiency and quality control.

[0003] Conventional methods of coal sorting involve manual or mechanical sorting, typically by screening or crushing to classify coal types. This method is highly inefficient. Computer vision technology, which uses cameras to capture coal images and deep learning to automatically identify and classify coal, is being used. However, when inspecting coal from different mining areas or batches, existing deep learning object detection models can produce discrepancies in the recognition of data from different mining areas or batches, leading to misclassification and reduced coal object recognition accuracy.

[0004] Therefore, there is an urgent need for a coal sorting method to improve and ensure the accuracy of coal detection in different mining environments or different batches. Summary of the Invention

[0005] Based on this, it is necessary to provide a coal sorting method, system, medium and equipment to address the above technical problems.

[0006] The present invention adopts the following technical solutions: The present invention provides a coal sorting method, comprising: Acquire real-time coal data from a mining area; perform target detection on the real-time coal data from the mining area using a coal target detection model to obtain coal detection results; the coal detection results include bounding box coordinates, confidence levels, and category labels; According to the preset first confidence interval, second confidence interval and third confidence interval, the coal data in the coal detection results that meet the preset confidence conditions are retained or eliminated, including: if the confidence is in the first confidence interval, the corresponding coal data is eliminated; if the confidence of the coal detection result is in the second confidence interval, the corresponding coal data is weighted and marked, and the weighted marked coal data is retained and stored; if the confidence of the coal detection result is in the third confidence interval, the corresponding coal data is retained and stored; if the confidence of the coal detection result is in the second confidence interval, the corresponding coal data is weighted and marked, and the weighted marked coal data is retained and stored. The coal data corresponding to the confidence level of the second confidence interval are labeled multiple times to obtain multiple labeling results of the coal data, including: bounding box coordinates, confidence levels and category labels; the bounding box coordinates after multiple labeling are weighted to obtain weighted bounding box coordinates; the confidence levels of the same category labels in the multiple labeling results are added to obtain a category score, and the category label corresponding to the highest category score is the target category of the coal data corresponding to the confidence level of the second confidence interval, and the weighted bounding box coordinates and target category are retained; wherein the maximum value of the first confidence interval is less than the minimum value of the second confidence interval; and the maximum value of the second confidence interval is less than the minimum value of the third confidence interval; The stored coal data is trained online through incremental learning to obtain a coal target detection model with updated parameters; The coal target detection model with updated parameters is used to perform target detection on real-time mining area coal data to obtain the latest coal detection results, and coal is sorted in real time based on the latest coal detection results.

[0007] Preferably, the coal target detection model is obtained by introducing a context guidance module into the YOLOv8 target detection model.

[0008] Preferably, the expression of the weighted bounding box coordinates is: ; Where, For the The confidence level of the annotation, , They are the four coordinate values ​​of the bounding box.

[0009] Preferably, the context guidance module comprises: a convolutional layer, a local feature extractor, a surrounding context extractor, a joint feature extractor and a global context extractor; In the local feature extractor, local features of the coal data processed by the convolution layer are extracted; In the surrounding context extractor, spatial pyramid pooling is used to aggregate the context information of coal data at different scales; In the joint feature extractor, the local features and the context information are spliced, and the spliced ​​features are integrated through the batch normalized linear unit and the parameterized linear unit to obtain the key features of the coal data; In the global context extractor, context information is extracted from the key features of the coal data through global average pooling, and the context information is deeply processed through two fully connected layers to obtain the final coal image feature map.

[0010] Preferably, the formula for extracting context information from the key features of the coal data by global average pooling and deeply processing the context information by two fully connected layers is: ; Where, is the final coal image feature map, The features extracted by the feature extractor, .

[0011] The present invention also provides a coal sorting system, comprising: Data acquisition module, used to obtain real-time coal data of mining areas; The target detection module is used to perform target detection on real-time mining area coal data using a coal target detection model to obtain coal detection results; based on a preset first confidence interval, a second confidence interval, and a third confidence interval, retain or eliminate coal data in the coal detection results that meet preset confidence conditions, including: if the confidence level is within the first confidence interval, then eliminate the corresponding coal data; if the confidence level of the coal detection result is within the second confidence interval, then weighted label the corresponding coal data, and retain and store the weighted labeled coal data; An online training module is used to perform online training on the stored coal data through incremental learning to obtain a coal target detection model with updated parameters; The coal sorting module is used to perform target detection on real-time mining area coal data through the coal target detection model with updated parameters, obtain the latest coal detection results, and sort the coal in real time based on the latest coal detection results.

[0012] The present invention also provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned coal sorting method is implemented.

[0013] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned coal sorting method when executing the program.

[0014] At least one of the above technical solutions adopted by the present invention can achieve the following beneficial effects: In a coal sorting method provided by the present invention, according to a preset confidence interval, a coal target detection model that introduces a context-guided module is used to perform target detection on the acquired coal data, while capturing and fusing the local features of the coal data in the mining area, the surrounding context information, and the global context information, thereby effectively improving the feature capture and learning capabilities of the YOLOv8 model, achieving high-precision coal detection, and screening and weighted labeling of the coal data based on the coal detection results to obtain high-quality coal data; utilizing the incremental learning and online training methods of high-quality coal data, a model operation mode of training while detecting is realized.

[0015] In summary, based on the detection results obtained by using the improved coal target detection model, the present invention uses high-quality coal data corresponding to the detection results to train and update the coal target detection model in real time, which can adapt to new mining area coal data in real time, ensure the prediction accuracy of the coal target monitoring model, and thereby effectively ensure the accuracy of coal target detection in different mining areas, different batches and different environmental changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A schematic flow chart of a coal sorting method provided by the present invention; Figure 2 A schematic diagram of a multi-path convolution and fusion neural network architecture for a coal sorting method provided by the present invention; Figure 3 A schematic diagram of a multi-stage feedback convolutional loop network for a coal sorting method provided by the present invention; Figure 4 A schematic diagram of a multi-stage feedback convolutional loop network for a coal sorting method provided by the present invention; Figure 5 A schematic diagram of weighted labeling of coal data in a coal sorting method provided by the present invention; Figure 6 A schematic diagram of mechanical equipment for a coal sorting method provided by the present invention; Figure 7 A schematic diagram of the detection results of a coal sorting method provided by the present invention; Figure 8 A schematic structural diagram of a coal sorting system provided by the present invention; Figure 9A schematic diagram of a computer device for implementing a coal sorting method provided by the present invention. DETAILED DESCRIPTION

[0017] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments of the present invention and corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in the specification, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0018] The following describes in detail the technical solutions provided by various embodiments of the present application in conjunction with the accompanying drawings.

[0019] Figure 1 The present invention is a flow chart of a coal sorting method, which specifically includes the following steps: S101: Acquire real-time coal data of a mining area; perform target detection on the real-time coal data of the mining area using an improved coal target detection model to obtain coal detection results; retain or eliminate coal data that meets a preset confidence condition in the coal detection results according to a preset confidence interval; mark and store the retained coal data.

[0020] Optionally, coal-related images (such as mining areas, conveyor belts, coal piles, and other scenes) are acquired through a camera, and the raw data is cleaned (by removing blur and duplicate images), normalized (by unifying the resolution), and enhanced (by rotating, flipping, adding noise, etc.) to improve the generalization ability of the model.

[0021] Optionally, the context-guided module includes: a convolutional layer, a local feature extractor, a surrounding context extractor, a joint feature extractor and a global context extractor.

[0022] Optionally, the local feature extractor is used to extract local features of the coal data processed by the convolution layer; and the surrounding context extractor aggregates context information of different scales of the coal data through spatial pyramid pooling.

[0023] Optionally, the joint feature extractor includes a batch normalized linear unit and a parameterized linear unit, which is used to integrate the local features of the coal data and the features after splicing the context information of the coal data at different scales through its batch normalized linear unit and parameterized linear unit to obtain the key features of the coal data.

[0024] Optionally, the global context extractor extracts context information from key features of the coal data through global average pooling, and deeply processes the context information through two fully connected layers to obtain the final coal image feature map.

[0025] Specifically, see Figure 2 , is a schematic diagram of the neural network architecture of multi-path convolution and fusion, which are the local feature extractor F1, the surrounding context extractor F2, the joint feature extractor F3 and the global context extractor F4. First, the input image is processed by the standard convolution layer, and then collaboratively processed by the local feature extractor F1 and the surrounding context extractor F2. The local feature extractor F1 focuses on the local features of the coal, while the surrounding context extractor F2 aggregates the context information of different scales of the image through spatial pyramid pooling. This ensures that the model can not only understand the information of each pixel, but also understand the relationship between these areas in the overall context. Subsequently, the key coal information captured by the local feature extractor F1 and the surrounding context extractor F2 is integrated through the joint feature extractor F3 composed of batch normalization and parameterized rectified linear units. Finally, the global context extractor F4 extracts context information from the entire image through global average pooling, and deeply processes this coal information through two fully connected layers to enhance the coal features learned by the joint feature extractor F3, thereby obtaining the features of the final output feature map. The formula is: ; Where, is the final coal image feature map, is the feature extracted by the feature extractor Fi, .

[0026] Specifically, see Figure 3 , a schematic diagram of a multi-stage feedback convolutional recurrent network. The context-guided module captures contextual features at all stages, fully leveraging information at both the semantic and spatial levels. This is crucial for accurately classifying every pixel in an image. This paper improves YOLOv8 with the proposed context-guided module to produce a better backbone feature extraction network. It simultaneously captures local features, surrounding context, and global context, fusing this information to improve semantic segmentation accuracy and enhance the feature capture and learning capabilities of the backbone network.

[0027] S102: retaining or eliminating coal data that meets preset confidence conditions in the coal detection results according to the preset first confidence interval, second confidence interval, and third confidence interval; Optionally, the maximum value of the first confidence interval is smaller than the minimum value of the second confidence interval; the maximum value of the second confidence interval is smaller than the minimum value of the third confidence interval; Optionally, the preset confidence intervals include a first confidence interval, a second confidence interval, and a third confidence interval; the retention or elimination of coal data that meets the preset confidence conditions in the coal detection results specifically includes: if the confidence of the coal detection result is in the first confidence interval, the corresponding coal data is eliminated; if the confidence of the coal detection result is in the second confidence interval, the corresponding data is weighted and marked and retained; if the review is passed, the coal data is retained, otherwise, the coal data is eliminated; if the confidence of the coal detection result is in the third confidence interval, the corresponding data is retained.

[0028] Specifically, the preprocessed data is input into the improved YOLO-v8 model, which analyzes the data, detects coal blocks, attempts to detect the coal targets within them, and outputs the detection results, including the target's location, category, and confidence level. The confidence scores are then divided into three confidence intervals: The first confidence interval is ≤50%, which is considered a low-confidence prediction and classified as a negative sample. These samples are considered to be incorrect detection results. The second confidence interval is 50% < confidence <90%, which is a medium-confidence prediction. The coal data corresponding to this confidence level is weighted and retained. The third confidence interval is ≥90%, which is a high-confidence prediction, marked as a correct prediction, and directly stored and trained.

[0029] Optionally, if the confidence level of the coal detection result is in the second confidence interval, the corresponding coal data is weightedly labeled, specifically including: labeling the coal data corresponding to the confidence level in the second confidence interval multiple times to obtain multiple labeling results of the coal data, including: bounding box coordinates, confidence levels, and category labels; weighting the bounding box coordinates after multiple labeling to obtain weighted bounding box coordinates; bounding box coordinates The weighted coordinates are expressed as: ; Where, For the The confidence level of the annotation, , They are the four coordinate values ​​of the bounding box.

[0030] The confidence levels of the same category labels in multiple annotation results are added together to obtain the category score. The category label corresponding to the one with the highest category score is the target category of the coal data corresponding to the confidence level of the second confidence interval, and the weighted bounding box coordinates and target category are retained.

[0031] Specifically, the coal data corresponding to the confidence level is weightedly labeled, including: for coal data with confidence levels in the second confidence interval, the data with confidence levels between 50% and 90% are weightedly labeled according to their bounding box positions and categories, and the weighted labeled data are stored in the incremental learning training dataset for incremental learning and online training. Different weights are assigned to the data based on the confidence level of the data. The higher the confidence level, the greater the weight assigned in the weighted labeling. For example, for data with a detection result confidence level of 0.8 and data with a confidence level of 0.6, 0.8 will have a higher weight in weighted labeling. Because a higher confidence level means that the model is more confident about the detection result, it is given greater influence during labeling, see Figure 5 , is a schematic diagram of weighted annotation of coal data, coal represents the coal target, and the numbers represent the weights after weighted annotation.

[0032] S103: Perform online training on the stored coal data through incremental learning, and update parameters of the improved coal target detection model.

[0033] Specifically, the model is incrementally learned and trained online using stored high-confidence annotated data. Incremental learning allows the model to continuously learn new data based on existing knowledge, thereby continuously optimizing the model's parameters and performance.

[0034] Specifically, traditional target detection models are usually trained offline and have difficulty adapting to dynamically changing real-world scenarios. However, this system uses online learning to update the model in real time using new, high-confidence data. This approach allows the model to continuously learn new features and patterns, improving the accuracy and adaptability of coal detection. Online learning allows the model to continuously absorb new knowledge during operation without the need for large-scale offline retraining. This not only saves time and computing resources, but also enables the model to adapt more quickly to environmental changes and changes in data distribution. For example, in a coal production environment, the appearance and quality of coal may change as the mining area changes. The online learning mechanism allows the model to capture these changes in a timely manner and maintain high detection accuracy. By screening high-quality data through confidence, combined with incremental learning and model updates, continuous optimization and closed-loop operation of the model are achieved.

[0035] S104: using the improved coal target detection model after parameter update to perform coal target detection on the newly acquired real-time mining area coal data to obtain real-time coal target detection results; and sorting the coal in real time according to the real-time coal target detection results.

[0036] Specifically, see Figure 4 , the updated model can be put into use again to detect the newly collected data, thus forming a closed-loop continuous optimization process to achieve continuous optimization and improvement of the model.

[0037] Specifically, the model undergoes online incremental training based on newly acquired annotated data (including positive and negative samples), dynamically updating model parameters to improve its adaptability to complex scenarios. The updated model's performance is then tested using validation sets or real-time data. Key metrics include: Precision (the proportion of correctly predicted results in detection results); Recall (the proportion of correctly detected objects in actual data); and Mean Average Precision (mAP), a comprehensive measure of detection accuracy. Evaluation results branch: Performance improvement confirms the effectiveness of the model optimization and proceeds to the model update phase; Performance degradation reverts to the old model and retrains. The new model, which meets performance requirements, is deployed in the actual application environment, completing the iterative closed-loop and continuously improving detection results.

[0038] In addition, this manual also provides mechanical equipment for coal testing, see Figure 6 ,include: Screw conveyor: The coal blocks are placed at the tail end of the conveyor, and then the screw conveyor pushes the coal along the closed tubular trough through the rotating spiral blades. After the coal blocks are transported to the top of the machine, they fall freely onto the horizontal conveyor belt, and the conveyor belt carries the coal blocks to continue moving forward.

[0039] Horizontal conveyor belt: The coal is on the belt conveyor, and the conveyor belt is driven by an electric motor, which uses friction to drive the coal to move on the belt, thereby realizing horizontal transportation of coal.

[0040] CCD industrial camera: A CCD industrial camera is installed above the conveyor belt to detect coal lumps. The CCD camera uses a global shutter for synchronized exposure and triggers capture with a high-brightness light source, ensuring motion blur-free images and high-precision image acquisition. The low-noise characteristics of the CCD are then utilized for image preprocessing, including image denoising, contrast enhancement, and geometric correction, providing clear and stable input data for subsequent analysis.

[0041] Spray dust suppression device: High-pressure water mist nozzles convert water into fine water mist particles with a particle size of 20-50 microns. These nozzles utilize their large surface area and surface tension to actively absorb dust from the air. When the water mist collides with dust particles, it causes the dust to increase in weight and fall out of suspension through wetting, agglomeration, and gravity settling, ultimately settling onto the conveyor belt. The device uses a dust concentration sensor to monitor the particulate matter content in the environment in real time. An intelligent linkage control system dynamically adjusts the spray frequency and water volume, achieving a "spray when dust is present, stop when dust is absent" approach. This device not only efficiently captures dust in the PM2.5 to PM10 range (capture efficiency exceeding 90%), but also precisely controls the surface moisture of the coal (typically, moisture content increases by ≤1.5%), preventing coal clumping or conveyor belt slippage caused by over-humidification.

[0042] High-performance computers (HPCs) primarily analyze coal block detection results, perform inference on the input image, and output detection results, including bounding box coordinates, category labels, and confidence scores. Data analysis and training are performed based on confidence levels. Low-confidence predictions (confidence ≤ 50%) are used as negative samples. Coal data corresponding to medium-confidence predictions (50% < confidence < 90%) are weighted and labeled. High-confidence predictions (confidence ≥ 90%) are marked as correct predictions, directly stored, and trained to form a new data model. The performance of the new model is then evaluated. If the new model's performance improves, confirming the effectiveness of the optimization, the model is updated. If performance degrades, the model is reverted to the old version and retrained.

[0043] Test result diagram: This is the final test phenomenon, see Figure 7 , coal means coal, which is a coal sorting method provided in this manual of mechanical equipment through the detection result picture when the conveyor belt is moving.

[0044] In summary, by training while detecting, the model can adapt to new data distributions in real time, which is crucial for handling changing environments (such as different lighting conditions, new target categories, etc.). At the same time, the model can continuously learn new patterns during use, improving detection accuracy and robustness, especially when facing previously unseen or rare objects. In traditional machine learning processes, a large amount of labeled data is required to train the model. Training while detecting can reduce the dependence on pre-labeled data because the model can learn directly from real-time data. Training while detecting can make more efficient use of computing resources because it does not need to frequently train the entire model from scratch, but instead fine-tunes the existing model.

[0045] The above is a coal sorting method provided in one or more embodiments of this specification. Based on the same idea, this specification also provides a corresponding coal sorting system, as shown in 8.

[0046] Figure 8 A schematic diagram of a coal sorting system provided for this specification includes: Data acquisition module 801, used to obtain real-time mining area coal data; The target detection module 802 is configured to perform target detection on real-time mining area coal data using a coal target detection model to obtain coal detection results; retain or eliminate coal data that meets preset confidence conditions in the coal detection results based on a preset first confidence interval, a second confidence interval, and a third confidence interval, including: if the confidence level is within the first confidence interval, eliminating the corresponding coal data; if the confidence level of the coal detection result is within the second confidence interval, weighted labeling the corresponding coal data, and retaining and storing the weighted labeled coal data; An online training module 803 is used to perform online training on the stored coal data through incremental learning to obtain a coal target detection model with updated parameters; The coal sorting module 804 is used to perform target detection on the real-time mining area coal data through the coal target detection model with updated parameters, obtain the latest coal detection results, and sort the coal in real time according to the latest coal detection results.

[0047] The specific definition of a coal sorting system can be found in the definition of a coal sorting method above and will not be repeated here. Each module in the above-mentioned coal sorting system can be implemented in whole or in part through software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a memory in the computer device in software form, so that the processor can call and execute the corresponding operations of each of the above modules.

[0048] This specification also provides a computer-readable storage medium, which stores a computer program that can be used to execute the above Figure 1 A coal sorting method is provided.

[0049] This manual also provides Figure 9 The structural diagram of the computer equipment shown in FIG. Figure 9 As shown in the figure, at the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, it may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 A coal sorting method is provided.

[0050] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-described method embodiments. Any reference to memory, storage, database, or other media used in the embodiments provided herein may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0051] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present invention.

Claims

1. A coal sorting method, characterized in that: include: Obtain real-time coal data from mining areas; Use the coal target detection model to perform target detection on real-time coal data in the mining area to obtain coal detection results; The coal detection results include: bounding box coordinates, confidence and category labels; According to the preset first confidence interval, second confidence interval and third confidence interval, the coal data in the coal detection results that meet the preset confidence conditions are retained or eliminated, including: if the confidence is in the first confidence interval, the corresponding coal data is eliminated; if the confidence of the coal detection result is in the second confidence interval, the corresponding coal data is weighted and marked, and the weighted marked coal data is retained and stored; if the confidence of the coal detection result is in the third confidence interval, the corresponding coal data is retained and stored; if the confidence of the coal detection result is in the second confidence interval, the corresponding coal data is weighted and marked, and the weighted marked coal data is retained and stored. The coal data corresponding to the confidence level of the second confidence interval are labeled multiple times to obtain multiple labeling results of the coal data, including: bounding box coordinates, confidence levels and category labels; the bounding box coordinates after multiple labeling are weighted to obtain weighted bounding box coordinates; the confidence levels of the same category labels in the multiple labeling results are added to obtain a category score, and the category label corresponding to the highest category score is the target category of the coal data corresponding to the confidence level of the second confidence interval, and the weighted bounding box coordinates and target category are retained; wherein the maximum value of the first confidence interval is less than the minimum value of the second confidence interval; and the maximum value of the second confidence interval is less than the minimum value of the third confidence interval; The stored coal data is trained online through incremental learning to obtain a coal target detection model with updated parameters; The coal target detection model with updated parameters is used to perform target detection on real-time mining area coal data to obtain the latest coal detection results, and coal is sorted in real time based on the latest coal detection results.

2. A coal sorting method according to claim 1, characterized in that: The coal target detection model is obtained by introducing a context guidance module into the YOLOv8 target detection model.

3. A coal sorting method according to claim 1, characterized in that: The expression of the weighted bounding box coordinates is: ； Where, For the The confidence level of the annotation, , They are the four coordinate values ​​of the bounding box.

4. A coal sorting method according to claim 2, characterized in that: The context guidance module includes: a convolutional layer, a local feature extractor, a surrounding context extractor, a joint feature extractor and a global context extractor; In the local feature extractor, local features of the coal data processed by the convolution layer are extracted; In the surrounding context extractor, spatial pyramid pooling is used to aggregate the context information of coal data at different scales; In the joint feature extractor, the local features and the context information are spliced, and the spliced ​​features are integrated through the batch normalized linear unit and the parameterized linear unit to obtain the key features of the coal data; In the global context extractor, context information is extracted from the key features of the coal data through global average pooling, and the context information is deeply processed through two fully connected layers to obtain the final coal image feature map.

5. A coal sorting method according to claim 4, characterized in that: The formula for extracting context information from the key features of the coal data through global average pooling and deeply processing the context information through two fully connected layers is: ； Where, is the final coal image feature map, The features extracted by the feature extractor, .

6. A coal sorting system, characterized in that: include: Data acquisition module, used to obtain real-time coal data of mining areas; The target detection module is used to perform target detection on real-time coal data in the mining area using the coal target detection model to obtain coal detection results; According to the preset first confidence interval, second confidence interval, and third confidence interval, retain or eliminate coal data in the coal detection results that meet the preset confidence conditions, including: if the confidence level is within the first confidence interval, eliminate the corresponding coal data; if the confidence level of the coal detection result is within the second confidence interval, perform weighted labeling on the corresponding coal data, and retain and store the weighted labeled coal data; An online training module is used to perform online training on the stored coal data through incremental learning to obtain a coal target detection model with updated parameters; The coal sorting module is used to perform target detection on real-time mining area coal data through the coal target detection model with updated parameters, obtain the latest coal detection results, and sort the coal in real time based on the latest coal detection results.

7. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

8. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method according to any one of claims 1 to 5 is implemented.

Images