ABSTRACT Accurate detection of hulling quality‐efficiency is critical for optimizing buckwheat processing and product quality. Current reliance on manual visual inspection is inefficient and impedes industrial progress. To address this issue, an online intelligent detection system is proposed based on an improved YOLOv8 algorithm. The model incorporates three key modifications: (1) a Dual Attention mechanism within the C2f module; (2) the WIoUv3 loss function replacing CIoU; and (3) an expansion to four detection heads. This refined framework enables real‐time detection of buckwheat hulling efficiency. The system consists of four core components: an image‐based detection module, a 6QB‐150 buckwheat huller, control actuators, and auxiliary subsystems. An image acquisition device captures real‐time samples, and the improved model accurately distinguishes three hulling statuses: hulled kernels, broken kernels, and unhulled kernels. Detection results are transmitted to a host computer, which interfaces with a PLC to regulate a three‐phase asynchronous motor, enabling adaptive control of grinding disc speed and equipment operation. Ablation experiments demonstrated substantial performance improvements over the baseline, with increases of 7.7% in precision, 8.1% in recall, 5.9% in mAP@50, and 4.2% in mAP@50–95. Comparative trials confirmed the model's superiority over YOLOv5, YOLOv6, YOLOv8n, and YOLOv10n, with negligible computational increase. In practical experiment, model‐detected rates for broken, hulled, and unhulled kernels deviated by merely +0.13%, −0.71%, and +0.53%, respectively, from manual counts, attesting to its reliability. The deployment of this system significantly enhances operational efficiency, offering an effective solution for automated quality control in buckwheat processing and contributing to agricultural modernization.
Fan et al. (Thu,) studied this question.