AMS-MLP: adaptive multi-scale MLP network with multi-scale context relation decoder for pepper leaf segmentation.

Area of Science:

• Computer Vision
• Agricultural Technology
• Machine Learning

Background:

• Pepper leaf segmentation is crucial for disease monitoring and ensuring healthy growth.
• Existing Transformer-based methods face challenges with computational inefficiency, high parameter counts, and poor edge information utilization.

Purpose of the Study:

• To develop an efficient and accurate framework for pepper leaf segmentation.
• To address the limitations of current Transformer-based segmentation techniques.

Main Methods:

• Introduced the Adaptive Multi-Scale MLP (AMS-MLP) framework, incorporating a Multi-Path Aggregation Module (MPAM) and a Multi-Scale Context Relation Mask Module (MCRD).
• The AMS-MLP features an encoder, an Adaptive Multi-Scale MLP (AM-MLP) module with global and local branches and adaptive attention, and a decoder.
• MPAM fuses five-scale features for boundary extraction, while MCRD refines boundary features.

Main Results:

• Achieved high performance on three pepper leaf datasets with varying backgrounds.
• Reported mean Intersection over Union (mIoU) scores of 97.39%, 96.91%, and 97.91%.
• Obtained F1 scores of 98.29%, 97.86%, and 98.51% across the datasets.

Conclusions:

• The proposed AMS-MLP framework significantly enhances the accuracy and efficiency of pepper leaf image segmentation.
• Demonstrated superior performance compared to U-Net and other state-of-the-art models.
• Offers a promising solution for automated agricultural monitoring and disease detection.