Deep Learning-Based Method for Classification of Sugarcane Varieties

The classification of sugarcane varieties using products derived from remote sensing allows for the monitoring plants with different profiles without necessarily having physical contact study objects. However, differentiating between can be challenging due to similarity spectral characteristics each crop. Thus, this aimed classify four through deep neural networks, subsequently comparing results traditional machine learning techniques. In order provide more data as input models, along multi-band values pixels and vegetation indices, other information obtained sensor bands RGB combinations by reconciling so yield crop varieties. methodology created discriminate consisted a dense network, number hidden layers determined greedy layer-wise method multiples neurons in layer; additionally, 5-fold evaluation training was composed Sentinel-2 band data, combinations. Comparing acquired model hyperparameters selected Bayesian optimisation, except network manually defined parameters, it possible observe greater precision 99.55% SVM model, followed developed study, random forests, kNN. final prediction resulted 99.48% accuracy six-hidden-layers demonstrating potential networks classification. Among that contributed most classification, chlorophyll-sensitive bands, especially B6, B7, B11, some combinations, had impact on correct samples model. regions encompassing near-infrared shortwave infrared proved suitable discrimination