A real-time deep learning method for detection and localization of non-emerged cabbage seedlings

Real-time detection and localization of non-emerged seedlings play a pivotal role in effective vegetable crop management. Traditionally, checking the non-emergence of crop seedlings relies on labor-intensive and time-consuming manual inspections. To address this issue, this paper introduces a precise, rapid, and real-time method for detecting and locating non-emerged seedlings using an unmanned aerial vehicle (UAV). The proposed Enhanced YOLOv8n detector improves upon the base model through a more efficient backbone and a bidirectional feature-fusion neck, achieving a mAP50–95 of 81.4%, an 8% improvement over the base model, while preserving real-time inference capability. Furthermore, integrating the Enhanced YOLOv8n, the YOLO tracking algorithm, and image processing algorithms enabled the development of a robust real-time detection and localization method for non-emerged seedlings. The recognition accuracy for non-emerged cabbage seedlings reached 96.9%. Notably, this method can distinguish the quantity of consecutive non-emerged seedlings, demanding only one-fifth of the manual effort, demonstrating substantial potential for practical applications. In summary, the UAV-deployed real-time detection and localization method exhibits excellent capability in detecting and locating non-emerged cabbage seedlings, thus significantly reducing the manpower needed for crop management.