Efficient Rice-Grain Segmentation, Alignment, and Classification Using YOLOv8 and Hybrid Attention Mechanisms

This study proposes an integrated and efficient framework for automated classification of rice grains, addressing key challenges in segmentation, orientation alignment, and category identification. The methodology incorporates a GraphCut-based segmentation strategy to isolate individual grains, followed by a dominant-axis rotation mechanism that standardizes their alignment. A lightweight yet powerful YOLOv8 model, finetuned for this task, is employed to categorize grains into five quality-based classes (T1-T5). To improve class distinction, both morphological and texture features are considered during training. Experimental results show strong performance, achieving a mean Average Precision (mAP) of 92.7%, an IoU of 88.4%, and an inference time of 15.2 ms per image. When benchmarked against conventional techniques such as SVM, KNN, Mask R-CNN, and Watershed-based segmentation, the proposed approach consistently outperforms them. The unified pipeline is not only accurate but also computationally efficient, making it well-suited for deployment in real-time grain quality monitoring systems within agricultural and industrial environments.