Design of Lightweight Vision System for Agricultural Drones: Real-time Identification of Crop Diseases under Complex Lighting Conditions

Abstract

Global food security is facing multiple pressures such as shrinking arable land and extreme weather, driving agriculture towards high precision and intelligence. Unmanned aerial vehicles (UAVs), as an aerial monitoring platform, provide important technical support for crop disease monitoring with their mobility and imaging capabilities. However, the complex lighting conditions in the field have led to a decline in image quality, and the limited computing power of the embedded platform has severely restricted the accuracy and real-time performance of disease identification. For this reason, this paper systematically reviews the latest progress in lightweight visual model design and light enhancement methods. In terms of model lightweighting, channel pruning, quantization techniques and knowledge distillation strike an effective balance between accuracy and efficiency; In terms of illumination enhancement, from traditional Retinex algorithms to physically guided generative adversarial networks, there is a significant improvement in image quality and model robustness. Future research could focus on areas such as polarization imaging and federated learning to further enhance the generalization ability and reliability of the system in complex environments.