Pruning Multi-Scale Multi-Branch Network for Small-Sample Hyperspectral Image Classification

In recent years, the use of deep learning models has developed rapidly in field hyperspectral image (HSI) classification. However, most network cannot make full rich spatial-spectral features images, being disadvantaged by their complex and low classification accuracy for small-sample data. To address these problems, we present a lightweight multi-scale multi-branch hybrid convolutional The contains two new modules, pruning block (PMSMBB) 3D-PMSMBB, each which part part. Each branch kernel different scales. training phase, can extract feature information through perceptual fields using asymmetric convolution feature, effectively improve model. model lighter, is introduced master module, remove insignificant parameters without affecting part, achieving light weight testing are jointly transformed into one convolution, adding any extra to network. study method was tested on three datasets: Indian Pines (IP), Pavia University (PU), Salinas (SA). Compared with other advanced models, this (PMSMBN) had significant advantages HSI For instance, SA dataset multiple crops, only 1% samples were selected training, proposed achieved an overall 99.70%.