The paper addresses the critical issue of water stress in viticulture, which is vital for improving grape yield and quality. The use of advanced deep learning methods and UAVs for data collection significantly enhances the accuracy and efficiency of water stress monitoring. The development of a Django-based web platform for interactive prediction and reporting to users is a substantial contribution to the practical applicability of the research. The proposed algorithm, based on the U-Net neural network, segments images to detect water stress using aerial RGB and thermal imagery. The model was successfully trained on the Agriculture-Vision dataset, showing promising results in segmenting agricultural patterns. Due to unfavourable weather conditions, data collection was limited, which may affect the completeness and reliability of the results. The use of transfer learning requires further refinement of the model to optimize parameters and adapt to new data.

--

UAV,RGB,Thermal,Deep learning,Convolutional neural networks,Water stress detection

Esta publicação é um output do(s) seguinte(s) projeto(s):

• Rede de Excelência de Viticultura Inteligente e Sustentável Assistida por Drones