Weed Classification with Drone Image Using DINOV2

Increasing interest in efficiency and intelligence of agriculture is driven by the growth of global population and food demand. Weed infestation is a critical factor restraining the soybean growth. Traditional weed management solutions are often inefficient as well as harmful to the environment. In this study, soybean fields weed recognition system based on DINOv2 is designed and implemented. The combination of drones captured images through all stages in Soybean lifecycle enabled precise classification and detection between soybean and weeds. The paper uses unsupervised learning and applies data augmentation methods including multi-scale cropping, color jitter, random occlusion for better generalization. Experimental results show that DINOv2 can achieve remarkable results on feature extraction and precision, while YOLOv5 is more competitive in terms of real-time efficiency with a lightweight design. This work explores how deep learning can be applied in agricultural applications and offers optimization strategies for precision agriculture. The findings contribute to reducing the application of pesticide, boosting soybean yield and contributing to sustainable agriculture.