Lightweight coal and gangue detection algorithm based on LTC-Yolov8n

Accurate recognition of coal and gangue is an important step in realizing intelligent sorting of coal and gangue, to address the traditional target detection algorithm model of large computational volume, real-time is not high, a model based on the LTC-Yolov8n coal gangue detection is proposed. Firstly, a lightweight three-channel LTC block is constructed in the Yolov8n backbone to replace the Conv block, which enhances the feature extraction capability while reducing the computational effort. Second, the top-down characteristic fusion path at the neck is cut and a fusion module is introduced to increase the ability of the different scales of feature information to interact with each other. The multi-scale target detection head is constructed to realize feature enhancement. Finally, Focal-SIoU is used to replace the loss function, thus speeding up convergence during training and making further improvements to the precision of the model. Comparing with the yolov8n experimental results, our accuracy and recall improve from 96.4% and 95.5% to 98.1% and 97%, respectively. The floating-point computation is reduced by 34.5%, and the number of frames is improved to 147 frames per second so that the study can have certain theoretical values and technical references for coal and gangue detection.