Aprendizado profundo para classificação e quantificação de doenças e pragas em imagens de folhas de café

• Autor(a) principal: Esgario, Jose Guilherme Mota
• Data de Publicação: 2019
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da Universidade Federal do Espírito Santo (riUfes)
• Texto Completo: http://repositorio.ufes.br/handle/10/13831
• topic: Ciência da Computação; Redes neurais (Computação); Redes neurais convolucionais; Reconhecimento de padrões; Segmentação semântica; Estresse biótico; Severidade; Doenças e pragas em folhas de café; Convolutional neural networks; Pattern recognition; Semantic segmentation; Biotic stress; Severity; Coffee diseases and pests; subject.br-rjbn

Biotic stress consists of damage to plants through other living organisms. Efficient control of biotic agents such as pests and pathogens (viruses, fungi, bacteria, etc.) is closely related to the concept of agricultural sustainability. Agricultural sustainability promotes the development of new technologies that allows the reduction of environmental impacts, greater accessibility to farmers and, consequently, increase on productivity. The use of computer vision with deep learning methods allows the identification of the stress-causing agent. So, corrective measures can be applied as soon as possible to mitigate the problem. Among the most commonly used approaches, convolutional neural networks present good results and high success rates. The goal of this work is to design an effective and practical system capable of identifying and estimating the stress severity caused by biotic agents on coffee leaves. In this work two approaches based on convolutional neural networks were investigated. The first approach consists of a one-step system trained to classify the leaf dataset with labels of biotic stress and severity. The second approach is formed by a two-step system, the semantic segmentation step, trained with the segmentation masks dataset, and the classification step, trained with the symptom dataset. The approaches were compared by highlighting their strengths and weaknesses. In addition, the use of multi-task learning and data augmentation techniques have been investigated in order to improve robustness and accuracy. The experimental results obtained an overall accuracy of 93% for biotic stress classification and 85% for severity estimation using the leaf dataset. Furthermore, it was found that by classifying only the symptoms, the accuracy was superior to 97%. The calculated severity using the segmentation masks showed a determination coefficient of around 0.98, suggesting that the model can obtain severity values very close to the real ones. The obtained results indicate that the proposed system might be a suitable tool to assist both experts and farmers in the identification and quantification of biotic stresses in coffee plantations