|
Food production is one of the significant challenges for the world's population. Countries like Brazil, with a vast territorial dimension and good availability of resources, stand out in the production of grains, especially soy. Soy cultivation requires care and management to ensure phytosanitary production and reduce the risk of diseases such as Asian Soybean Rust (ASR) caused by the fungus Phakopsora pachyrhizi. In Brazil, soy cultivation occurs in the spring/summer (September/March), with greater solar energy and rainfall in the country. Brazil has established a fallow period to reduce the risk of ASR, which prohibits planting outside the agricultural calendar. However, there is the possibility of authorizing planting in the floodplains of the tropical plains of the Formoso River basin, Tocantins, Brazil. The government of the State of Tocantins created the State Program for the Control of ASR, authorizing the planting of soybeans during the dry season (April to September) through registration and monitoring of areas. However, other plantings, such as beans, with a shorter cycle and less water demand, also occur. This study aims to monitor the soybean crop development phases considering data collected in the field by the Agricultural Defense Agency (ADAPEC) and digital processing using deep-learning techniques of Sentinel-1 image time series. The phenological differences of cultivation farms enabled agricultural mapping and the fight against ASR. The digital processing steps of the Sentinel-1 time series dataset (10 m resolution) consisted of image preprocessing using Sentinel Application Platform (SNAP); time series filtering using Savitzky-Golay; evaluation of deep learning methods (Long Short-Term Memory - LSTM, Bidirectional LSTM - Bi-LSTM, Gated Recurrent Unit - GRU, and Bidirectional GRU - Bi-GRU); and accuracy analysis. However, the classification has some erroneous portions that can be improved by increasing the number of classes and samples in future works.
|
