Classificação da produtividade de soja e avaliação de grupos de maturação por meio de imagens multiespectrais
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidade Federal de Viçosa
Abstract
A participação da soja (Glycine max (L.) Merril) em diversos setores agrícolas e industriais, fazem da cultura um dos principais produtos do agronegócio brasileiro. Tal fato é reflexo do progresso genético sobre a expansão da oleaginosa pelo país nas últimas décadas. Tendo em vista a importância do desenvolvimento de cultivares mais precoces e produtivas este trabalho teve como objetivo avaliar o potencial do uso de imagens multiespectrais obtidas por VANTs para substituir ou simplificar a coleta de dados em campo, além de classificar grupos de maturação e produtividade em plantas de soja, visando auxiliar na seleção de genótipos superiores. O estudo avaliou dois experimentos conduzidos em delineamento de blocos casualizados com três repetições no município de Viçosa, MG. O experimento I contou com 84 populações e os 15 genitores avaliados em um total de 297 parcelas. Para o experimento II foram utilizadas 15 populações e seis genitores, totalizando 63 parcelas. Foram utilizadas 11 imagens, por experimento, obtidas por meio de missões de voo, na altura de 40 m, realizadas durante o ciclo da soja. O modelo de drone utilizado foram o DJI Matrice 100 (DJI Innovations, Shenzhen, China) equipado com a câmera multiespectral MicaSense RedEdge MX (MicaSense, Seattle, WA, EUA). Com base nas bandas espectrais das imagens processadas, foram calculados os índices de vegetação ARVI, CIG, CIRE, CVI, DVI, EVI, GNDVI, LCI, MCARI1, MSR, NDRE, NDVI, NGRDI, PRI, RDVI, RI, RVI, SAVI, SIPI, SR, TVI. Em seguida, cada conjunto de dados de índices por data de voo, foram aplicados na Rede Neural Feed Forward (FFNN). As análises foram realizadas por meio do software R para tratamento das imagens e para a construção dos modelos de predição. O modelo baseado nos IVs, obteve precisão e acurácia acima de 60% na classificação de produtividade. Os IVs, calculados a partir dos dados de refletância do dossel, não apresentaram diferenças significativas entre os grupos de maturação que pudessem contribuir para identificar cultivares mais precoces. No entanto, 113 dias após o plantio, entre os estádios R6-R7, os IVs ARVI, NDVI, TVI, SIPI, PRI, RI, RVI, LCI e NGRDI se mostraram mais aptos a identificar diferença entre os grupos de maturação. Palavras-chave: Índices de vegetação. Modelos de classificação. Redes neurais. Glycine max
The participation of soybean (Glycine max (L.) Merril) in several agricultural and industrial sectors makes the culture one of the main products of Brazilian agribusiness. This fact is a reflection of the genetic progress on the expansion of the oilseed across the country in recent decades. Considering the importance of developing earlier and more productive cultivars, this study aimed to evaluate the potential of using multispectral images obtained by UAVs to replace or simplify field data collection, in addition to classifying maturation and productivity groups in plants. of soy. The study evaluated two experiments carried out in a randomized block design with three replications in the city of Viçosa, MG. Experiment I had 84 populations and 15 parents evaluated in a total of 297 plots. For experiment II, 15 populations and six parents were used, totaling 63 plots. Eleven images were used, per experiment, obtained through flight missions, at a height of 40 m, carried out during the soybean cycle. The drone model used was the DJI Matrice 100 (DJI Innovations, Shenzhen, China) equipped with the MicaSense RedEdge MX multispectral camera (MicaSense, Seattle, WA, USA). Based on the spectral bands of the processed images, the vegetation indices ARVI, CIG, CIRE, CVI, DVI, EVI, GNDVI, LCI, MCARI1, MSR, NDRE, NDVI, NGRDI, PRI, RDVI, RI, RVI were calculated, SAVI, SIPI, SR, TVI. Then, each dataset of indices by date were applied to the Neural Feed Forward Network (FFNN). The analyzes were performed using the R software to process the images and phenotypic data, and to build the prediction models. The model based on IVs obtained precision and accuracy above 60% in the productivity classification. IVs, calculated from canopy reflectance data, did not show significant differences between maturation groups that could contribute to identifying earlier cultivars. However, 113 days after planting, between the R6-R7 stages, the IVs ARVI, NDVI, TVI, SIPI, PRI, RI, RVI, LCI and NGRDI were more able to identify differences between the maturation groups. Keywords: Vegetation Indexes. Classification models. Neural networks. Glycine max
The participation of soybean (Glycine max (L.) Merril) in several agricultural and industrial sectors makes the culture one of the main products of Brazilian agribusiness. This fact is a reflection of the genetic progress on the expansion of the oilseed across the country in recent decades. Considering the importance of developing earlier and more productive cultivars, this study aimed to evaluate the potential of using multispectral images obtained by UAVs to replace or simplify field data collection, in addition to classifying maturation and productivity groups in plants. of soy. The study evaluated two experiments carried out in a randomized block design with three replications in the city of Viçosa, MG. Experiment I had 84 populations and 15 parents evaluated in a total of 297 plots. For experiment II, 15 populations and six parents were used, totaling 63 plots. Eleven images were used, per experiment, obtained through flight missions, at a height of 40 m, carried out during the soybean cycle. The drone model used was the DJI Matrice 100 (DJI Innovations, Shenzhen, China) equipped with the MicaSense RedEdge MX multispectral camera (MicaSense, Seattle, WA, USA). Based on the spectral bands of the processed images, the vegetation indices ARVI, CIG, CIRE, CVI, DVI, EVI, GNDVI, LCI, MCARI1, MSR, NDRE, NDVI, NGRDI, PRI, RDVI, RI, RVI were calculated, SAVI, SIPI, SR, TVI. Then, each dataset of indices by date were applied to the Neural Feed Forward Network (FFNN). The analyzes were performed using the R software to process the images and phenotypic data, and to build the prediction models. The model based on IVs obtained precision and accuracy above 60% in the productivity classification. IVs, calculated from canopy reflectance data, did not show significant differences between maturation groups that could contribute to identifying earlier cultivars. However, 113 days after planting, between the R6-R7 stages, the IVs ARVI, NDVI, TVI, SIPI, PRI, RI, RVI, LCI and NGRDI were more able to identify differences between the maturation groups. Keywords: Vegetation Indexes. Classification models. Neural networks. Glycine max
Description
Citation
DIAS, Eleniz Aparecida. Classificação da produtividade de soja e avaliação de grupos de maturação por meio de imagens multiespectrais. 2022. 33 f. Dissertação (Mestrado em Genética e Melhoramento) - Universidade Federal de Viçosa, Viçosa. 2022.
