Squeeze-and-Excitation

A channel attention mechanism that compresses spatial information through global average pooling and learns channel recalibration weights, thereby enhancing the feature representation capability of convolutional neural networks.