FEL-YoloV8: A New Algorithm for Accurate Monitoring Soybean Seedling Emergence Rates and Growth Uniformity

Effective monitoring of soybean emergence rates and growth uniformity is crucial for soybean breeding evaluation and field management. Although Unmanned Aerial Vehicles (UAVs) have improved image acquisition efficiency, detecting soybean seedlings during the vegetative emergence (VE) stage remains challenging due to their small size, low contrast, and insufficient information. Existing studies often neglect emergence rate and growth uniformity quantification. This study proposes a fully automated method for monitoring soybean emergence rate and growth uniformity, applicable to both ground-based and UAV platforms. The FEL-YoloV8 model was constructed by enhancing the feature extraction module, improving the feature fusion module, and incorporating model lightweight modules. Based on the detection results from the FEL-YoloV8 model, the missing seedling locations, counts, and growth uniformity of soybeans were estimated. The study shows that the feature enhancement and fusion modules improved the model’s performance by 2.10%. Under comparable computational complexity, the performance of the FEL-YoloV8 model (AP = 0.979) surpasses current advanced models (e.g., Faster R-CNN, RT-DETR, YoloV8n, s, m, l, x, YoloV9, YoloV10, L-FFCA-Yolo, and TPH-YoloV5). The detection results from the FEL-YoloV8 model enabled the estimation of missing seedling locations and quantities in soybeans. The proposed method enables fully automated monitoring of soybean emergence rates and growth uniformity. This approach lays the foundation for accurate multi-platform evaluation of soybean emergence rates and growth uniformity, guiding soybean breeding evaluation and field management.