Fig. 4

WheatLFANet global regression architecture. The output channel numbers C₁, C₂, and C₃ are 32, 64, and 128, respectively, with (Conv k × k) where k is the size of the convolutional kernel. Compared to the core architecture of TasselLFANet, the downsampling method is replaced by a normal Conv layer, and the feature mapping layer is a lighter CSPLayer