An urgent research challenge is the development of automated methods for monitoring maize diseases using computer vision and untanned aerial vehicles (UAVs). Traditional visual inspection by agronomists is labor-intensive and inefficient for detecting early infection stages, leading to significant yield losses. The aim of this study is to develop and validate a method for detecting maize leaf spot from UAV-acquired RGB images using a convolutional neural network (CNN) ResNet-50. The study object was maize fields in the Republic of Bashkortostan. Imaging was performed with an industrial UAV DJI Matrice 300 equipped with a 20 MP RGB camera. A dataset of ~14,000 images was collected, including 6000 healthy and 8000 diseased leaves. ResNet-50 with fine-tuning was used for binary classification. Model performance was evaluated using Accuracy, Precision, Recall, and F1-score. The model achieved an overall Accuracy of ≈92 % and F1-score of 0.91, reliably distinguishing healthy and infected leaves under field conditions. Based on the infection index I, Variable Rate Application (VRA) maps were generated, prescribing fungicide application rates of 120, 180, and 250 L/ha across field zones. Unlike most studies limited to classification, the proposed approach is integrated into the agro-engineering framework of precision agriculture, CNN outputs are converted into ISOXML/Shape prescription maps compatible with ISOBUS/ RTK sprayers. The practical significance lies in reducing pesticide costs and chemical load on agroecosystems while maintaining crop protection efficiency. Future work will focus on extending detection to multiple diseases and incorporating multispectral data.
S.G. Mudarisov (Wed,) studied this question.