Symmetry, Год журнала: 2025, Номер 17(5), С. 745 - 745

Опубликована: Май 13, 2025

The gangue mixed in raw coal has small feature differences from coal, order to solve the existing recognition, methods generally have slow detection speed and are difficult deploy at edge end of problem, a lightweight target algorithm is proposed enhance research for field mining. Firstly, EfficientViT module backbone network; secondly introduction DRBNCSPELAN4 module, which can better capture information different scales; finally, shared convolutional head Detect_LSCD reconstructed further reduce model size improve gangue. experimental results indicate that compared with original algorithm, mAP@50–95 improved by 1.2%, weight size, number parameters, floating point operations reduced 52.34%, 55.35%, 50.35%, respectively, inference accelerated 20.87% on Raspberry Pi 4B device. In sorting, not only high-precision, real-time performance, but also achieves significant model, making it more suitable deployment equipment meet requirements controlling robotic arm sorting

Язык: Английский