Estimativa de produtividade da cultura do milho a partir de imagens multiespectrais

• Autor(a) principal: Abreu Júnior, Carlos Alberto Matias de
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFU
• Texto Completo: https://repositorio.ufu.br/handle/123456789/36220; http://doi.org/10.14393/ufu.di.2022.264
• topic: Cultura do milho; Corn crops; Produtividade; Yield; Imagens multiespectrais; Multispectral images; Algoritmos de aprendizado de máquina; Machine learning algorithms; Modelos espectrais de produtividade; Spectral yield models; Agronomia; Agronomy; CNPQ::CIENCIAS EXATAS E DA TERRA::GEOCIENCIAS::GEOFISICA::SENSORIAMENTO REMOTO; Agronomia; Produtividade agrícola; Dados geoespaciais

Estimated yield is an important agronomic parameter for the domestic and international market in the demand, transport capacity and storage of agricultural products. In this respect, continuous yield monitoring requires the creation of methods to enhance mapping of this variable in the field. As such, this study aimed to determine the ideal phenological stage to estimate yield and generate models for this purpose on a local scale, that is, using data from different study areas to train and validate the model. To that end, we used multispectral images from different orbital sensors combined with vegetation indices from the reflectance bands of each satellite and models generated by the following machine learning algorithms: support vector machine (SVM), neural network (NN), random forest (RF) and random trees (RT). In Chapter 1, Planet satellite images were used to determine the ideal phenological stage and in Chapter 2, yield estimate models were generated via a Sentinel 2 satellite image. The root mean square error percentage (RMSE%) and mean absolute percentage error (MAPE) were used as evaluation metrics to assess the accuracy and trend of the yield estimates. Based on analysis to determine the ideal phenological stage, it was concluded that the spectral model based on the image of phenological stage R2 obtained the best RMSE% (9.17%) and second best MAPE (7.07%). The local-scale estimation models generated made it possible to estimate yield with an RMSE% and MAPE of 20.97 and 19.19%, respectively.