Efficient online detection device and method for cottonseed breakage based on Light-YOLO.

Abstract

High-quality cottonseed is essential for successful cotton production. The integrity of cottonseed hulls plays a pivotal role in fostering the germination and growth of cotton plants. Consequently, it is crucial to eliminate broken cottonseeds before the cotton planting process. Regrettably, there is a lack of rapid and cost-effective methods for detecting broken cottonseed at this critical stage. To address this issue, this study developed a dual-camera system for acquiring front and back images of multiple cottonseeds. Based on this system, we designed the hardware, software, and control systems required for the online detection of cottonseed breakage. Moreover, to enhance the performance of cottonseed breakage detection, we improved the backbone and YOLO head of YOLOV8m by incorporating MobileOne-block and GhostConv, resulting in Light-YOLO. Light-YOLO achieved detection metrics of 93.8% precision, 97.2% recall, 98.9% mAP50, and 96.1% accuracy for detecting cottonseed breakage, with a compact model size of 41.3 MB. In comparison, YOLOV8m reported metrics of 93.7% precision, 95.0% recall, 99.0% mAP50, and 95.2% accuracy, with a larger model size of 49.6 MB. To further validate the performance of the online detection device and Light-YOLO, this study conducted an online validation experiment, which resulted in a detection accuracy of 86.7% for cottonseed breakage information. The results demonstrate that Light-YOLO exhibits superior detection performance and faster speed compared to YOLOV8m, confirming the feasibility of the online detection technology proposed in this study. This technology provides an effective method for sorting broken cottonseeds.

Plant phenotyping relevance

綿実の破損状態という植物器官の形態・品質特性を対象に、二眼カメラによる取得装置、オンライン検出システム、Light-YOLO解析手法を開発し、性能検証まで行っているため、植物フェノタイピング手法が中心である。

Code and data availability

The supplied blocks describe a cottonseed breakage dataset (416 images, 6240 seed instances) and the Light-YOLO model, but contain no public deposit, availability URL, or code release for the dataset, images, or trained model. A 'Data availability statement' section is listed in the outline but its text is not included

No evidence-backed public reproduction asset is currently recorded.