Aplicação da transformada de Hilbert-Huang e da seleção paraconsistente de características no campo da entomologia agrícola: uma abordagem para identificação e estimativa de densidade de cigarras em lavouras cafeeiras

• Autor(a) principal: Souza, Uender Barbosa de
• Data de Publicação: 2023
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Repositório Institucional da UFG
• Texto Completo: http://repositorio.bc.ufg.br/tede/handle/tede/13175
• topic: Processamento de sinais; Transformada de Hilbert-Huang; Decomposição em modo empírico; Engenharia paraconsistente de características; Aprendizado de máquinas; Support vector machines; Manejo de cigarras; Signal processing; Hilbert-Huang transform; Empirical mode decomposition; Paraconsistent feature engineering; Machine learning; Support vector machines; Cicada management; ENGENHARIAS::ENGENHARIA ELETRICA

The sounds emitted by various insect species carry specific and reliable acoustic characteristics. For this reason, acoustic identification of insects has been widely investigated by the scientific community in the field of pattern recognition. In Brazil, the cicada species Quesada gigas is considered a pest in coffee crops due to its sap-feeding habits, which can cause losses to farmers during intense attacks. Given this scenario, and considering the fact that the most striking characteristic of cicadas is sound emission for reproductive purposes, this work proposes a system designed to assist in the management of pest insects in coffee crops. Specifically, this system aims to detect the presence of cicadas or estimate their density through acoustic signals. The approach innovatively combines the extraction and analysis/selection of sound features through the Hilbert-Huang Transform (HHT) and Paraconsistent Feature Engineering, two emerging methods of growing interest in the scientific community. A detailed study was conducted on the influence of eight stopping criteria for the Empirical Mode Decomposition, the first step of HHT, considering variations of parameters, temporality, signal sampling, and also the encoding of these signals in two formats. The first defines vectors by the energies of the Intrinsic Mode Functions (IMFs) in order of extraction, while the second distributes the energies of the IMFs according to the 25 frequency bands of the Bark Scale. Additionally, the dimensional variation of the vectors was analyzed. The experiments allowed determining low computational cost configurations, showcasing the efficacy of the proposed system, with models based on Support Vector Machines achieving accuracies above 98% in both identification and density estimation of cicadas. Thus, this document details the theoretical foundations, design, development of the system, discussions about possibilities for its practical application, and an initial analysis of a developed application for smartphones. This proposal has the potential to encourage the reuse of old electronic equipment, promoting sustainable and economically viable practices.