Online detection method for rice processing quality based on DSS-YOLOv8: Application to the hulling stage
Year: 2026
Type: article
Abstract: The rice huller, which removes husks from paddy grains using counter-rotating rubber rollers, is a critical component in rice processing. However, roller wear inevitably enlarges the roller gap, reducing the hulling rate and increasing kernel breakage. Current quality control relies on periodic manual inspection and parameter adjustment, a process that is labor-intensive, time-consuming, and lacks real-time responsiveness. To address these limitations, this study developed an integrated online detection system that combines automated feeding, impurity removal, image acquisition, and real-time quality indicator calculation. A customized detection model, DSS-YOLOv8, was designed by enhancing the YOLOv8 architecture with Dense blocks, Shuffle Attention modules, and an optimized detection head. These improvements strengthen feature representation and enhance discrimination among visually similar grain categories in the huller discharge. Furthermore, a position-prediction-based deduplication algorithm was proposed to eliminate duplicate counts across consecutive frames, ensuring reliable grain-level statistics under continuous flow conditions. Experimental results demonstrated that the system achieved a mean average precision (mAP) of 99.0% and a mean F1-score of 97.1%, with a high processing speed of 88.5 frames per second (FPS). The system accurately calculated hulling and breakage rates with maximum absolute errors below 0.81% while stably processing 200 g of discharge per minute. These findings demonstrate the potential of combining deep learning with online monitoring to provide an accurate, efficient, and practical solution for intelligent quality control in rice hulling. • SDS-YOLOv8 improves detection accuracy for rice processing quality by enhancing feature representation. • A position-prediction-based deduplication algorithm eliminates duplicate counts across frames. • Achieved 99.0% mAP, with detection latency reduced to 11.3 ms per frame. • Real-time detection of hulling and breakage rates with errors below 0.81%. • Online platform stably processes 200 g of rice discharge per minute.
Language: English
Topic: Smart Agriculture and AI
Subfield: Plant Science
Domain: Life Sciences
Open Access status: gold
Estimated APC paid: $2,530
Award: G7264201747
OpenAlex ID: W7119469181