Automated drone-assisted detection system for rice leaf pathologies a deep learning approach

Abstract

Early and accurate detection of plant diseases is vital for maintaining agricultural productivity. This study investigates an automated disease identification system specifically designed for the IR64 rice cultivar. By combining drone-captured aerial imagery (UAV) taken during the plant's vegetative stage with public datasets, we established a comprehensive training dataset. The study evaluates and compares four convolutional neural network (CNN) architectures, InceptionV3, ResNet50, EfficientNetV2S, and MobileNetV2, assessing their predictive accuracy and real-world computational efficiency. Our 10-fold cross-validation results indicate varying levels of inference speed and accuracy among the models. InceptionV3 and MobileNetV2 displayed the highest stability and minimal misclassification rates across multiple disease types. In contrast, the performance of ResNet50 and EfficientNetV2S fluctuated significantly depending on the detected pathogen. In conclusion, coupling UAV imagery with fine-tuned deep learning models provides a fast, scalable solution for continuous crop monitoring and precision agriculture.