Modifications of the Multi-Layer Perceptron for Hyperspectral Image Classification

Recently, many convolutional neural network (CNN)-based methods have been proposed to tackle the classification task of hyperspectral images (HSI). In fact, CNN has become de-facto standard for HSI classification. It seems that traditional networks such as multi-layer perceptron (MLP) are not competitive However, in this study, we try prove MLP can achieve good performance if it is properly designed and improved. The Modified-MLP contains two special parts: spectral–spatial feature mapping information mixing. Specifically, mapping, each input sample divided into a sequence 3D patches with fixed length then linear layer used map features. For mixing, all features within single feed solely architecture model across following Furthermore, obtain abundant different scales, Multiscale-MLP aggregate neighboring multiscale shapes acquiring information. addition, Soft-MLP further enhance by applying soft split operation, which flexibly capture global relations at positions sample. Finally, label smoothing introduced mitigate overfitting problem (Soft-MLP-L), greatly improves MLP-based method. Modified-MLP, Multiscale-MLP, Soft-MLP, Soft-MLP-L tested on three widely datasets. leads highest OA, outperforms 5.76%, 2.55%, 2.5% Salinas, Pavia, Indian Pines datasets, respectively. obtained results reveal models provide compared state-of-the-art methods, shows still