Uso de redes neurais convolucionais para a segmentação de mancha anelar do mamoeiro com uso de imagens por UAV

Abstract

Precision agriculture is a technology-based approach that uses and analyzes data to improve agricultural practices. It enables farmers to make the most informed decisions, maximize resource efficiency and promote more sustainable agricultural production. This work aims to integrate precision agriculture with Machine Learning (ML) techniques for the detection of the ringspot disease in papaya plantations, from a small sampling Ąeld of papaya cultivation in the interior of Bahia, in order to analyze the behavior of the segmentation networks regarding the disease. The images were captured using an unmanned aerial vehicle on papaya plantations. The BF techniques used were segmentation neural networks: UNET, PSPNET, LINKNET and FCN combined with the VGG16 deep architecture, for the training, validation and testing process, the dataset formed by the captured images was used. The trained networks were validities on the results of accuracy and F1-Score, in which they obtained results above 79%, but these indications are not affirmative for a conclusive validation for the detection of the disease, with variations of false positives in the result of the images.