Graph-augmented object detection for robust cotton boll maturity assessment

This study proposes an enhanced framework for segmenting matured cotton bolls by integrating YOLOv8-Seg with Graph Neural Networks (GNNs) to improve detection performance under real-field challenges such as occlusion, clutter, and variable lighting. The proposed model enhances conventional YOLOv8-Seg by introducing a graph-based reasoning module that captures spatial dependencies and co-occurrence patterns among detected bolls. By converting YOLOv8 outputs into graph-structured representations, the GNN refines feature embeddings to achieve more context-aware segmentation. Experimental evaluation indicates that the proposed framework maintains a precision of 1.00 while increasing recall from 0.93 to 0.95 and F1-score from 0.82 to 0.97 compared to the baseline YOLOv8-Seg. These results suggest its improved robustness and reduced false negatives in complex agricultural scenes. The integration of GNN-based spatial reasoning establishes a scalable pathway toward intelligent cotton maturity assessment, yield estimation, and precision harvesting automation.