Agronomy, Год журнала: 2024, Номер 14(11), С. 2558 - 2558

Опубликована: Окт. 31, 2024

The quality inspection of potato seed tubers is pivotal for their effective segregation and a critical step in the cultivation process potatoes. Given dearth research on intelligent tuber-cutting machinery China, particularly concerning identification bud eyes defect detection, this study has developed multi-target recognition approach utilizing deep learning techniques. By refining YOLOv5s algorithm, novel, lightweight model termed DCS-YOLOv5s been introduced simultaneous tuber buds defects. This initiates with data augmentation images obtained via image acquisition system, employing strategies such as translation, noise injection, luminance modulation, cropping, mirroring, Cutout technique to amplify dataset fortify model’s resilience. Subsequently, original undergoes series enhancements, including substitution conventional convolutional modules backbone network depth-wise separable convolution DP_Conv module curtail parameter count computational load; replacement C3 module’s Bottleneck GhostBottleneck render more compact; integration SimAM attention mechanism augment proficiency capturing features defects, culminating model. findings indicate that outperforms detection precision velocity, exhibiting superior efficacy compactness. metrics, Precision, Recall, mean Average Precision at Intersection over Union thresholds 0.5 (mAP1) 0.75 (mAP2), have improved 95.8%, 93.2%, 97.1%, 66.2%, respectively, signifying increments 4.2%, 5.7%, 5.4%, 9.8%. velocity also augmented by 12.07%, achieving rate 65 FPS. target model, attaining compactness, substantially heightened precision, presenting beneficial reference dynamic sample context potato-cutting machinery.

Язык: Английский