Fig. 4
From: A multi-scale dual-stream fusion network for high-accuracy sEMG-based gesture classification
Structure of the Bi-directional Attention Fusion Module (BAFM). Two input feature maps are independently enhanced through \(1 \times 1\) convolutions, concatenated, and transformed via successive \(1 \times 1\) convolutions with Batch Normalization and ReLU activations. The final fusion combines the features using element-wise multiplications and integrates them through MaxPooling.
