面向复杂动态表情识别的感知增强网络OA

To address the difficulty of detecting low-intensity expressions in dynamic facial expression recognition tasks,this paper proposes a plug-and-play Dynamic Expression Aware Enhancement Network(DEAEN).Unlike the traditional 3D facial expression recog-nition model M3DFEL,DEAEN adds a global convolution layer and a multi-scale adaptive attention mechanism while keeping the overall structure unchanged.This effectively improves the feature extraction performance,allowing the model to recognize complex expressions more accurately.In addition,an intensity analysis module is designed to adaptively allocate recognition weights based on the strength of the expression,thereby enhancing the detection of subtle expressions.Experimental results on the DEFW dataset show that the improved model outperforms current mainstream methods in terms of weighted average recall(WAR)and unweighted average recall(UAR),espe-cially when recognizing complex expressions.Ablation studies further confirm that the combination of these modules plays a key role in enhancing the performance of the traditional 3D facial expression recognition model,thus verifying the effectiveness of the proposed meth-od in identifying low-intensity expressions.