Spatial Feature Extraction using Pretrained Convolutional Neural network for Hyperspectral Image Classification

Hyperspectral images (HSIs) captured a detail range of electromagnetic spectrum from visible to near to infrared to each pixel. Due to variability of spectral data and lack of labeled data HSI classification is a challenging work. The convolutional neural network (CNN) have successfully used in object detection and classification. A model based on spatial features extracted using pretrained convolutional neural network presented in this paper. The features are extracted at fully connected layer. Our proposed model consists of principle component analysis (PCA) for dimension reduction, followed by pretrained CNN for the purpose of spatial features and SVM classifier. In experiments, we used pretrained CNN (AlexNet) and HSI data set (Indian Pine). Experimental result with HSI dataset demonstrate that classifier based on our proposed model provide very competitive performance of overall accuracy (97.76%), average accuracy (98.77%) and K-score (0.9732). In addition, results of the experiments are compared with state-of-the-art HSI classification methods.