Working face status detection in coal mine based on YOLOv8-EST.

With the continuous development of deep learning, the application of deep neural network-based object detection in coal mines is expanding. Meanwhile, with the increasing requirements of recognition accuracy in production applications, most studies choose to expand the depth and parameters of the network to improve the recognition accuracy. However, due to the limited computational resources in the coal mining face, it is challenging to meet the computational demands of a large number of hardware resources. Therefore, this paper proposes a novel object detection algorithm, YOLOv8-EST, specifically designed for coal mining face. The algorithm introduces the Swin Transformer module in the YOLOv8 network to enhance the feature extraction capability. The MLP module contains the GELU activation function, and the present invention obtains GELUS by improving the GELU activation function, which is changed to exponential computation when X takes on a positive value, ensuring that the computational complexity is reduced on the basis of improving the model training performance. A deep learning network is used to generate relative position encoding, replacing simple linear encoding, improving the expression ability of the relative position labeling so that it can capture more complex spatial relationships. The introduction of the EMA Attention mechanism further enhances the network's ability to capture relevant features by focusing on important regions of the input image. These modifications aim to improve the detection performance without significantly increasing the computational load. The proposed YOLOv8-EST algorithm is validated on a CM dataset collected by self-acquisition in real coal mining face work, and shows significant improvement in both detection accuracy and efficiency. Comparative analysis with other existing algorithms demonstrates the efficiency and usefulness of the algorithm in coal mining scenarios. Experiments confirm that YOLOv8-EST significantly reduces the computational requirements and size of the model while maintaining high accuracy. This makes it an ideal solution for real-time, efficient, and accurate detection of the state of the coal mining face, which helps to improve production safety and efficiency.