A Depthwise Separable Fully Convolutional ResNet With ConvCRF for Semisupervised Hyperspectral Image Classification

Abstract

Hyperspectral images (HSIs) classification relies on the accurate and efficient extraction of discriminative features, detail preservation, and efficient learning with limited training samples. This paper, therefore, presents an advanced neural network architecture combined with convolutional conditional random fields (ConvCRF) and region growing (RGW) approaches to address these key issues. First, a depthwise separable fully convolutional residual network (DFRes) is proposed for efficient feature learning, where a fully convolutional operation ensures a larger field of view, and residual learning and depthwise separable convolution can mitigate the problem of vanishing gradient and overfitting. Second, because the collection of ground-truth labels is usually difficult, the proposed architecture integrates the RGW method to effectively overcome the problem of limited training samples. Third, ConvCRF is used to preserve the image details for fine-grained predictions. Finally, the above key components are coherently integrated into the new semisupervised framework, i.e., DFRes with Conditional random fields and Region growing (DFRes-CR). Experimental results on three hyperspectral datasets demonstrate that the proposed approach outperforms other state-of-the-art methods.

From Video to Hyperspectral: Hyperspectral Image-Level Feature Extraction with Transfer Learning

“…Therefore, using the global image as input can effectively avoid the flaws of the patchlevel method. This alternative method, which can be called the image-level method [23][24][25][26][27][28][29], is gaining popularity. The practice proves that the image-level method can improve classification accuracy and efficiency.…”

MADANet: A Lightweight Hyperspectral Image Classification Network with Multiscale Feature Aggregation and a Dual Attention Mechanism

“…Ever since AlexNet [11] secured victory in the 2012 ImageNet challenge, convolutional neural networks (CNNs) have garnered considerable attention. In subsequent years, the field of image classification witnessed the emergence of several classic CNN models, notably GoogleNet, VGGNet, ResNet, and DenseNet [12][13][14][15], and many researchers have also conducted studies on hyperspectral image classification based on these models [16][17][18][19][20][21][22][23][24]. However, to enhance accuracy, most of these models incorporate an extensive number of hidden layers and training parameters, which limits their suitability for deployment in resource-constrained mobile and embedded applications, particularly on satellite and airborne platforms [25].…”