UNet Architecture: Replacing Max Pooling with Convolution for Path Expansion

The UNet replaces the 2x2 max pooling in the downsampling section with a 2x2 convolution to expand the path, and reduces the number of feature channels by half through upsampling at each layer. The image size is expanded by 2x2 pixels after each layer's convolution. In the final step of upsampling, a 1x1 convolution kernel is used to map the image with 64 feature channels to two feature channels. These two feature channels respectively output the image target and background parts.