Influence of Different Optimizers on Classification Results in Deep Learning

Abstract

In satellite remote sensing technology, hyperspectral images not only have the same spatial information as traditional RGB images and hyperspectral images, but also have rich spectral information. Deep learning can connect the training data and label data through nonlinear mapping to extract different levels of information features from hyperspectral data. Based on the remote sensing data sets of Indian pine, saline field and Pavia University, this paper uses the Hybridsn model to compare the classification performance of different optimizers. The experimental results show that the adaptive time estimation optimizer makes the model show good classification performance in the classification process.