Pointer-Type Instrument Recognition Based on Improved U2-net
Jingwei Li, Hongxing Zhang, Hailang Jia, Dawei Shi, Xinyue Wang, Dengyin Zhang · 2023
The reading and recognition technology of pointer-type instruments have laid the foundation for the realization of intelligent industry. The existing image segmentation method has achieved good results in the reading and recognition stage, but the problem that the low recognition accuracy of pointer and scale in the actual complicated scene is inevitable. As a result, an image segmentation method based on the modified U2-Net is proposed to improve the recognition accuracy of strip scale and pointer. Firstly, this method presents a SPRSU (Strip Pooling Residual U-block) module to add the attention of strip objects during the process of pointer-type recognition. The recognition accuracy of strip-shaped objects in the segmentation process is considerably improved by adding a strip pooling layer to the residual joint. Secondly, the focal loss function is used to solve the problem of uneven distribution of image samples in the process of instrument segmentation. The experimental results suggests that the proposed method increases pointer and scale recognition accuracy in complicated scenarios.