Surgical instrument segmentation method based on improved MobileNetV2 network

摘要

The surgical instrument robot can realize the delivery and classification management of surgical instruments through the visual image analysis. The segmentation of surgical instrument images is the core technology of instrument robots. In this study, we propose a surgical instrument segmentation framework based on improved MobilenetV2 network. We use the first 8 layers of MoblieNetV2 for feature extraction. In order to strengthen the effective feature learning of related tasks, the CBAM module of the attention mechanism and atrous spatial pyramid pooling module is embedded into the MobileN etV2 encoder part. Two low-level feature sources are added to the decoder part, which helps to retain important feature information and improves segmentation accuracy. The experimental results show that the MIoU, PA, fps value of the improved network on the surgical instrument dataset are 0.861,0.885,and 24.8. Compared with other semantic segmentation networks, the improved network segmentation performance is better, and it has certain industrial value for the realization of instrument robots.