Spectral-Spatial Feature Extraction and Classification by ANN Supervised With Center Loss in Hyperspectral Imagery

Abstract

In this paper, we propose a spectral-spatial feature extraction and classification framework based on artificial neuron network (ANN) in the context of hyperspectral imagery. With limited labeled samples, only spectral information is exploited for training and spatial context is integrated posteriorly at the testing stage. Taking advantage of recent advances in face recognition, a joint supervision symbol that combines softmax loss and center loss is adopted to train the proposed network, by which intra-class features are gathered while inter-class variations are enlarged. Based on the learned architecture, the extracted spectrum-based features are classified by a center classifier. Moreover, to fuse the spectral and spatial information, an adaptive spectral-spatial center classifier is developed, where multiscale neighborhoods are considered simultaneously, and the final label is determined using an adaptive voting strategy. Finally, experimental results on three well-known datasets validate the effectiveness of the proposed methods compared with the state-of-the-art approaches.

A CNN-Based Spatial Feature Fusion Algorithm for Hyperspectral Imagery Classification

“…Recently, the neural network-based models have been utilized in hyperspectral images classification, achieving remarkable improvements over the traditional methods in terms of classification performance. Earlier works include the stacked autoencoder (SAE) [24,25], the deep belief network (DBN) [26] and etc.. More recently, many researches are dedicated to the varieties of CNN and RNN-based models, as studied in [27][28][29][30][31][32][33][34][35][36][37].…”

“…Training with pixel patches is a natural idea to take advantage of both spectral and spatial information, and is adopted by most of the aforementioned neural network-based studies [27][28][29][30][31][32][33][34][35]37]. Representative works include the so-called 3D-CNN in [28], where pixel patches are directly fed to the deep model, and the integrated spectral-spatial features can be extracted from the hyperspectral data.…”

“…Moreover, to circumvent the obstacle raised by training with pixel patches, frameworks are built by concatenating a spectrum-based model and a post spatial information integration scheme. Following this idea, the pixel-pair feature algorithm (PPF) [38] and the ANNC with adaptive spatial-spectral center classifier (ANNC-ASSCC) [35] have been proposed in the literature. Together with the aforementioned MASR and 3D-CNN, the four methods are chosen to be compared with the proposed CSFF algorithm in the experiments, as to be reported in Section 3.…”

“…In [35], an ANN-based feature extraction and classification framework is proposed. The so-called ANNC-ASSCC algorithm consists of two successive steps:…”

“…In this paper, we choose the ANNC network in [35] as the spectral feature extraction model, and design a CNN-based Spacial Feature Fusion (CSFF) algorithm to posteriorly integrate the spatial information. The main characters of the proposed classification framework are listed as below.…”