FSCformernet: A Fourier-Transformer UNet for Efficient Semantic Segmentation of Plant Leaf

Plant leaf semantic segmentation technology is of great significance in monitoring and analyzing plant growth conditions, disease occurrence, and physiological responses. However, the currently existing semantic segmentation methods of plant leaves do not pay enough attention to the global information, spatial information and channel information of the image, resulting in unclear segmentation of leaf outline, especially in complex environments and small targets, especially in complex environments and small targets. To solve this problem, this paper proposes a U-shaped encoder-decoder symmetric network FSCformernet. To modeling global information, we use the Fourier transform to convert image information from spatial domain to frequency domain, and apply the shifted window attention mechanism. To enhance spatial information and channel information, we use the Fourier spatial interaction module and the Fourier channel deduction module. In addition, we also use center sharpening filtering based on Fourier transform in the U-shaped network to obtain a clearer segmentation effect. A large number of image segmentation experiments show that our model achieves impressive segmentation results with high efficiency in the field of leaf image segmentation.