Hyperspectral Image Classification Based on a Shuffled Group Convolutional Neural Network with Transfer Learning

Abstract

Convolutional neural networks (CNNs) have been widely applied in hyperspectral imagery (HSI) classification. However, their classification performance might be limited by the scarcity of labeled data to be used for training and validation. In this paper, we propose a novel lightweight shuffled group convolutional neural network (abbreviated as SG-CNN) to achieve efficient training with a limited training dataset in HSI classification. SG-CNN consists of SG conv units that employ conventional and atrous convolution in different groups, followed by channel shuffle operation and shortcut connection. In this way, SG-CNNs have less trainable parameters, whilst they can still be accurately and efficiently trained with fewer labeled samples. Transfer learning between different HSI datasets is also applied on the SG-CNN to further improve the classification accuracy. To evaluate the effectiveness of SG-CNNs for HSI classification, experiments have been conducted on three public HSI datasets pretrained on HSIs from different sensors. SG-CNNs with different levels of complexity were tested, and their classification results were compared with fine-tuned ShuffleNet2, ResNeXt, and their original counterparts. The experimental results demonstrate that SG-CNNs can achieve competitive classification performance when the amount of labeled data for training is poor, as well as efficiently providing satisfying classification results.

Hyperspectral Remote Sensing Image Classification Using Deep Convolutional Capsule Network

“…Zhang et al [31,32] and Mu et al [33] introduced feature fusion techniques to combine different scale information to enhance the robustness of deep networks against overfitting. Liu et al [34] performed transfer learning between different HSI datasets to improve the HSI classification for small-sample conditions. To learn more representative features from the original HSI, sparse representation [35], metric learning [36], and attention techniques [37] are also used for refining the learned spectral-spatial features.…”

Hyperspectral Image Classification with Localized Graph Convolutional Filtering

“…Yu et al (2020) proposed a novel method to exploit local spectral similarity and nonlocal spatial similarity by considering spatial consistency [26]. Liu et al (2020) proposed a novel lightweight shuffled graph convolutional network (GCN) to accelerate the training procedure through a limited number of training data [27]. Making a mark on the latest, the recent novelties regarding graph representation learning have attracted more and more attention from the community.…”

Residual Multi-Attention Classification Network for A Forest Dominated Tropical Landscape Using High-Resolution Remote Sensing Imagery

“…These gradually developed networks provide frameworks for remote sensing image classification [19]. Semantic segmentation networks, especially convolution neural networks, are widely used in land use and land cover (LULC) classifications [20][21][22][23], road extraction [24,25], building extraction [26,27], etc. At present, there are still many problems to be solved in the classification of tropical forest area and its surroundings based on deep neural networks.…”