Método para classificação de padrões da Lagarta do cartucho (Spodoptera frugiperda) na cultura do milho baseado em processamento de imagens digitais e aprendizado de máquina

• Ano de defesa: 2021
• Autor(a) principal: Bertolla, Alex Bisetto
• Orientador(a): Cruvinel, Paulo Estevão
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de São Carlos; Câmpus São Carlos
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciência da Computação - PPGCC
• Palavras-chave em Português: Processamento de imagens digitais; Reconhecimento de padrões; Classificação de padrões de pragas; Aprendizado de máquina
• Palavras-chave em Inglês: Digital image processing; Pattern recognition; Pests classification; Machine learning
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO
• Link de acesso: https://repositorio.ufscar.br/handle/20.500.14289/15648

The detection, identification, and control of the Fall Armyworm (Spodoptera frugperda) pest into the maize culture (Zea mays) are greatly dependent on the human factor. Currently, such control occurs mainly through the use of capture traps. This makes the diagnosis of infestations of this pest inefficient and can cause significant damage to production, as well as in general some additional use of pesticides. The objective of this research is to use image and signal processing techniques to establish a method for recognizing the Fall Armyworm (Spodoptera frugperda) pattern in maize culture, allowing its early, reliable and supervised recognition, which improves the state of art of controller procedments in order to obtain an automatized process. Image aquisition, image enhancement, segmentation, features extraction, the use of Principal Components Analisys (PCA) and superviosioned classification techniques were considered for the method development. For image aquisition, it has been used an online image data base. For image enhancement, Gaussiam and Non-Local Means filters were experimented for noise reduction. Mean Square Error (MSE) and Peak Signal-to-Noise Ratio (PSNR) were measured for filters avaliation. For semgmentation process, thresholding with seed pixels were experimented on HSV and CIE L*a*b* color spaces. In order to automatize, Otsu technique was applied in the process of image segmentation. For feature extraction, Histogram of Orientation Gradient (HOG) and invariants moments of Hu were experimentd, in order to obtain texture and geometric information, respectively, as well as, for feature vector dimensionality reduction Princial Components Analisys (PCA) were experimented. For pattern classification of Fall Armyworm (Spodoptera frugperda) a set of classifiers based on Support Verctor Machine (SVM) was established. The developed method has shown to be suitable for Fall Armyworm (Spodoptera frugperda) pattern classification, wich has contributed to the porpouse of decision making for pest identification and control on maize culture. The method has also contributed to the evolution of digital image processing techniques and analisys tools.