Fig. 4: The illustration of Intra-layer Encoding Stream.

This figure presents the core structure of the Intra-layer Encoding Stream. First, each convolutional layer’s feature map is independently processed using an intra-layer binary encoding strategy based on local mean values, aiming to capture fine-grained local texture patterns. Second, this process yields four independent intra expert perspectives sign and magnitude mapping tensors. Last, we concatenate the intra-layer sign and magnitude encoding feature vectors to obtain the final feature vector.