Abstract

To implement the intelligence and automation of coal mines, coal recognition plays a crucial role. In order to further improve the accuracy and speed of intelligent coal recognition, this paper proposes a semantic segmentation model based on an improved PSPNET network. (1) The lightweight MobilenetV2 module is used as the backbone feature extraction network. Compared to traditional networks, it has fewer parameters while achieving higher recognition accuracy and speed.(2) The Convolutional Block Attention Module (CBAM) is introduced into the Pyramid Pooling Module (PPM) to enhance the network's ability to extract detailed features and effectively fuse spatial and channel information, thus improving the segmentation accuracy of the model.(3) Data augmentation and image feature enhancement methods are employed to overcome sample distribution differences, enhance model generalization, and adapt to coal-rock recognition tasks in different application scenarios. The proposed approach is tested on a self-made coal segmentation dataset and compared with the unimproved PSPNET, Hernet, U-net, and DeeplabV3+ models in terms of Mean Intersection over Union (Miou), recognition accuracy, edge detail recognition, model size, and parameter count. Experimental results demonstrate that compared to other models, the improved PSPNET network not only has lower computational complexity and parameter count but also exhibits stronger coal detail feature extraction capability, higher segmentation accuracy, and better processing efficiency.Finally, the improved PSPNET model was trained and tested on a coal rock image segmentation dataset with image feature enhancement.The accuracy, MIU and MPA of the improved PSPNET network reached 65.04, 73.15 and 74.27 respectively.It can be seen that the improved network has superior feature extraction ability and computational efficiency to achieve coal surface image recognition. This verifies the feasibility and effectiveness of the proposed method in the actual coal rock image recognition task.

This work is licensed under a Creative Commons Attribution 4.0 License.