Abstract

Background: The precision agriculture sector benefits greatly from advanced semantic segmentation techniques for land cover mapping and crop monitoring. Traditional methods often struggle with accuracy and efficiency due to the complexity of agricultural environments. Problem: Existing segmentation methods lack the ability to handle the diversity and scale of agricultural images effectively, leading to suboptimal classification and segmentation results. Method: This study introduces a three-stage semantic segmentation process leveraging deep learning and intelligent agents. The process begins with feature extraction using Chaotic Evolutionary Agents and parallel coding, followed by feature fusion and enhancement to create comprehensive feature maps. In the segmentation stage, a dual approach is adopted: region-based classification with U-Net for region candidates and pixel-based classification for fine-grained results. The final stage involves post-processing with boundary optimization to refine segmentation outputs. Results: The proposed method shows a significant improvement in segmentation accuracy and computational efficiency compared to existing methods. The method achieves an average accuracy of 92.5% and a reduction in processing time by 30% compared to traditional algorithms.