A Local-Global Multi-Scale MLP Network for Plant Leaf Segmentation Based on Context Masking

The health status of plant leaves directly reflects whether the plant has healthy growth. By accurately segmenting the leaf area in plant leaf images, real-time monitoring of leaf diseases can be achieved. Although existing Transformer-based segmentation methods can capture long-range dependencies between pixels, they are computationally intensive and do not fully utilize edge information. To address these issues, this paper proposes a local-global Multi-scale MLP network framework, called LGM-MLP. The LGM-MLP architecture comprises an encoder network, an local-global multi-scale MLP module, and a decoder network. The LGM-MLP module divides the input features into two branches: the global multi-scale MLP branch captures long-range dependencies in the image, while the local multi-scale MLP branch focuses on extracting local feature maps. The local-global attention mechanism dynamically adjusts the weights of global and local features. In the decoder network, the CRD module is incorporated into the convolutional layer to enhance the extraction of boundary features. Comparative analysis with U-Net and state-of-the-art models reveals that our proposed method significantly improves the accuracy and efficiency of plant leaf image segmentation.