Endoscopic Surgical Procedure Recognition Based on Convolutional Networks and Feature Focusing

Abstract

To address the issues of weak dynamism, feature redundancy, and noise interference in process recognition during endoscopic surgery, this paper proposes a deep learning framework integrating convolutional networks and feature focusing. This study designs a lightweight dual-channel feature extraction module, constructs a spatio-temporal attention-guided feature network, and combines dynamic threshold segmentation with multi-scale feature fusion. Experimental results show that, in terms of computational efficiency, the inference speed reaches 57fps, which is 40% higher than the traditional method; in regard to professional standardization, the average rule compliance rate remains at 89.3%. Under 50% image noise interference, the model accuracy remains at 88.7%, which can resist interference well in complex surgical scenarios. These data demonstrate the strong adaptability of the framework to complex surgical scenarios. This method can provide key technical support for intelligent surgical navigation and surgical quality assessment.