Simplifying Crop Health Assessment Problem Detection with Drones and Deep Learning

A crop health assessment problem detection with drones and Deep Learning is an agricultural early detection system that employs multispectral imaging alongside deep learning methodologies to assess the color of rice crops by examining leaf pigmentation based on the Leaf Color Chart (LLC) parameter. This approach addresses challenges in the agricultural field by minimizing the labor-intensive aspects of manual evaluations and mitigating the drawbacks of conventional or automated systems that are susceptible to variations in lighting. The system uses a DJI Mavic 3 multispectral drone equipped with RGB and near-infrared sensors, in conjunction with a YOLOv8 model for classification of leaf colors (L2-L5). The method involves collecting images with a timestamp and GPS data that is captured by drone sensors. These data will then be processed through detection servers correlating to leaf color conditions with the time, date, and Location as a context. Based on the test result, the system can achieve a detection accuracy of 98.07% (mAP50) at optimal altitudes of 20-25 meters. While maximum detection accuracy is attained at lower altitudes of 10-15 meters, higher altitudes offer broader area coverage. Field trials conducted in Gresik and Jogja rice paddies test the system's effectiveness with a hybrid RGB-NDVI dataset, where NDVI serves as a lighting-invariant verification tool. The locally processed data produces georeferenced 2D health maps via a web platform, which enables farmers to pinpoint problematic areas across extensive fields while preserving the context of each measurement.