Alteration Detection of Multispectral/Hyperspectral Images Using Dual-Path Partial Recurrent Networks

Jinlong Li,Li Feng,Xiaochen Yuan

doi:10.3390/rs13234802

Abstract

Numerous alteration detection methods are designed based on image transformation algorithms and divergence of bi-temporal images. In the process of feature transformation, pseudo variant information caused by complex external factors will be highlighted. As a result, the error of divergence between the two images will be further enhanced. In this paper, we propose to fuse the variability of Deep Neural Networks’ (DNNs) structure flexibly with various detection algorithms for bi-temporal multispectral/hyperspectral imagery alteration detection. Specifically, the novel Dual-path Partial Recurrent Networks (D-PRNs) was proposed to project more accurate and effective deep features. The Unsupervised Slow Feature Analysis (USFA), Iteratively Reweighted Multivariate Alteration Detection (IRMAD), and Principal Component Analysis (PCA) were then utilized, respectively, with the proposed D-PRNs, to generate two groups of transformed features corresponding to the bi-temporal remote sensing images. We next employed the Chi-square distance to compute the divergence between two groups of transformed features and, thus, obtain the Alteration Intensity Map. Finally, threshold algorithms K-means and Otsu were, respectively, applied to transform the Alteration Intensity Map into Binary Alteration Map. Experiments were conducted on two bi-temporal remote sensing image datasets, and the testing results proved that the proposed alteration detection model using D-PRNs outperformed the state-of-the-art alteration detection model.

Highlights

Imagery alteration detection generally refers to the techniques of divergence recognition in the same geographical location observed over time in order to detect the pixel-level alterations caused by various natural and human factors, such as the alterations of river channels, geological disasters, artificial buildings, vegetation cover, and so on
9(R3,R4), the black and white pixels indicate the areas detected with the proposed ADM-Dual-path Partial Recurrent Networks (D-PRNs) and the state-of-the-art models Deep Slow Feature Analysis (DSFA)-64-2, DSFA-128- as invariand variant, respectively
The three post-processing methods, Iteratively Reweighted Multivariate Alteration Detection (IRMAD), in white represent the variant areaup forfor the the entire image, while those in FigureD9(R4) repreand Principal Component Analysis (PCA), were employed to evaluate and make deficiency of the proposed sent the variant area for the known areas

Summary

Introduction

Imagery alteration detection generally refers to the techniques of divergence recognition in the same geographical location observed over time in order to detect the pixel-level alterations caused by various natural and human factors, such as the alterations of river channels, geological disasters, artificial buildings, vegetation cover, and so on. The availability of open and shared bi-temporal multispectral/hyperspectral imagery datasets, as well as Synthetic Aperture Radar (SAR) imagery datasets, facilitates researchers in testing the superiority of proposals for alteration detection. Many imagery datasets are radiometric corrected [1,2,3,4], which offers a foundation for the following works such as image recognition [5,6,7], analysis, and classification [8,9]. In the field of alteration detection, the widely used transformation algorithms have been proposed to extract and map the original image data into a new space. The transformation algorithms are mainly the classical Multivariate Alteration Detection (MAD) [10], which serves other transformation algorithms, Iteratively Reweighted Multivariate Alteration Detection (IRMAD) [11] and the Principal Component Analysis (PCA) [12], as well as the Independent Component Analysis

Results

Discussion

Conclusion