Sensoriamento remoto suborbital de alta resolução na detecção da cochonilha rosada da cana-de-açúcar saccharicoccus sacchari (Cockerell, 1895) (hemiptera: pseudococcidae)

• Ano de defesa: 2024
• Autor(a) principal: Silva, Allef De Souza
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Fitotecnia e Ciências Ambientais; Programa de Pós-Graduação em Agronomia; UFPB
• Palavras-chave em Português: amostragem; entomologia; agricultura de precisão; aprendizagem de máquina; inteligência artificial; CNPQ::CIENCIAS AGRARIAS
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/35487

Sugarcane is widely cultivated worldwide, with Brazil being the largest global producer and exporter. As such, this crop plays a fundamental role in the country's economy, generating millions of reais annually through sugar and ethanol production. Like any other crop, sugarcane is susceptible to pest attacks throughout its vegetative cycle, which can hinder its development to the extent that, if the producer does not make timely and appropriate decisions, significant economic losses in productivity may occur. Therefore, precise and efficient monitoring is essential and can be conducted using Remotely Piloted Aircraft (RPAs). In this context, the objective of this study was to detect occurrence patterns of S. sacchari in sugarcane crops using the NDVI and NDRE vegetation indices. The chosen study site was a 1.5-hectare sugarcane cultivation area located on the Mandaú farm in the municipality of Alagoa Grande-PB. A survey of S. sacchari population density in sugarcane at the stalk development stage was conducted. Random sampling points were selected using a zigzag walk pattern within the observational unit, with a sampling area of 10 m². At each point, 20 plants were selected 10 from the row on the left and 10 from the row on the right. Images were captured using a DJI Mavic 3 multispectral RPA. In parallel with correlation matrices and networks, a supervised factorial machine learning model was implemented to analyze the relationship between infestation levels—low (BNI) and high (ANI)—and the NDVI and NDRE vegetation indices. Gaussian univariate analyses (regression and correlation) and multivariate analyses using unsupervised machine learning showed consistency in the relationship between vegetation indices and infestation levels. According to the spatial analysis of vector disposition in the biplot of the factorial model, mediated by the cosine of angle θ, negative relationships were observed (cosine of angle θ > 130º = r > 0.60; R² > 0.70; P < 0.05) between the variables BNI vs. NDRE and NDVI vs. ANI, as well as weak relationships between NDVI and BNI and between NDRE and ANI (cosine of angle θ < 130º = r < 0.60; R² < 0.70; P < 0.05). Therefore, the results of this study provide evidence that suborbital remote sensing using the NDVI and NDRE vegetation indices offers complementary information for the indirect identification of injuries caused by S. sacchari in sugarcane crops. NDRE was found to be more strongly associated with low infestations, while NDVI was related to high infestations.