Unsupervised Hyperspectral Image Change Detection via Deep Learning Self-Generated Credible Labels

Abstract

Change detection (CD) aims to identify differences in scenes observed at different times. Hyperspectral image (HSI) is preferred for the understanding of land surface changes, since it can provide essential and unique features for CD. However, due to the high-dimensionality and limited data, the HSI-CD task is challenged. While model-driven CD methods are hard to achieve high accuracy due to the weak detection performance for fine changes, data-driven CD methods are hard to be generalized due to the limited data sets. The state-of-art method is to combine a single model-driven method with a data-driven convolutional neural network (CNN). Wherein the pseudo-labels can be generated automatically by the model-driven method and then fed to CNN for training. However, the final detection accuracy is limited by the model-driven method which produces pseudo-labels with one-sidedness and low accuracy. Therefore, the generation of credible pseudo-labels is anticipated and crucial for such a combination. Herein, a novel strategy, the combination of two complementary model-driven methods, structural similarity (SSIM) and change vector analysis (CVA), is proposed to generate credible labels for training a subsequent CNN. The results show that the final accuracy is higher than that of SSIM and CVA. The main contributions of this paper are threefold: 1) a new paradigm for generating credible labels is proposed. 2) SSIM is used for the first time for HSI-CD tasks. 3) an unsupervised end-to-end framework is presented for the HSI-CD. Experimental results demonstrate the effectiveness of the proposed framework.Index Terms-Change detection (CD), deep learning, heterogeneous images, hyperspectral image (HSI), structural similarity (SSIM).

A Mutual Teaching Framework with Momentum Correction for Unsupervised Hyperspectral Image Change Detection

“…The other is shown in Figure 1b. The results obtained by the traditional algorithms are assigned to all samples as pseudo-labels to train neural networks, which is more commonly used [17,[19][20][21]25,26,33,34]. These methods are easy to implement and closer to end-to-end patterns, avoiding the intermediate steps of difference image analysis.…”

“…Thus, it is difficult to obtain in large quantities. To solve the problem, existing unsupervised HSI-CD methods usually use pseudo-labels generated by traditional algorithms [17,[19][20][21]. One of the main challenges is that the training process of neural networks is susceptible to noise in pseudo-labels.…”

“…Li et al [20] added a noisy model with zero-mean Gaussian distribution to their loss function, yet the experimental effect was general. The authors in [21] adopted two unsupervised algorithms to jointly generate credible labels. However, the same problem still exists, where it is impossible to filter all noisy labels by only one-time sample selection.…”

ESatSR: Enhancing Super-Resolution for Satellite Remote Sensing Images with State Space Model and Spatial Context

“…In order to interact with the Image Feature Extraction Module, we flattened the spatial contextual information of three different scales into three 1D sequences and concatenated them together to obtain the spatial prior information I 1 sp . Since each convolution reduces the size of feature maps by half, we can obtain the spatial contextual information at the scales of 1 8 , 1 16 , and 1 32 . The shape of concatenated spatial prior information I 1 sp can be noted as (D, ( 1 8 2 + 1 16 2 + 1 32 2 )HW).…”

“…Due to its wide coverage and high accessibility, it has been widely used in disaster detection, early warning, resource exploration, land cover classification, and other fields. High-resolution satellite remote sensing images are widely applied across various scenarios such as change detection [1,2] and object detection [3,4]. Due to the far distance between spacecrafts and the ground, satellite remote sensing images span extensive geographic areas, but the spatial resolution is relatively low, resulting in numerous small targets appearing in the images.…”