Fig. 4

Schematic of the EMCA attention mechanism. \({\text{F}}_{\text{s}}\) denotes the feature map at the s-th scale, \({\text{Z}}_{\text{s}}\) is the channel descriptor obtained via GAP, \({\text{a}}_{\text{s}}\) is the channel attention weight produced by \({\text{Conv1D}}_{\text{k}}\) and Sigmoid, and \({\overline{F}}_{s}\) is the enhanced feature map.