Two-Phase Object-Based Deep Learning for Multi-Temporal SAR Image Change Detection

Xinzheng Zhang,Guo Liu,Xichuan Zhou,Xin Jian,Ce Zhang,Peter M Atkinson,Yongming Li,Xiaoheng Tan

doi:10.3390/rs12030548

Abstract

Change detection is one of the fundamental applications of synthetic aperture radar (SAR) images. However, speckle noise presented in SAR images has a negative effect on change detection, leading to frequent false alarms in the mapping products. In this research, a novel two-phase object-based deep learning approach is proposed for multi-temporal SAR image change detection. Compared with traditional methods, the proposed approach brings two main innovations. One is to classify all pixels into three categories rather than two categories: unchanged pixels, changed pixels caused by strong speckle (false changes), and changed pixels formed by real terrain variation (real changes). The other is to group neighbouring pixels into superpixel objects such as to exploit local spatial context. Two phases are designed in the methodology: (1) Generate objects based on the simple linear iterative clustering (SLIC) algorithm, and discriminate these objects into changed and unchanged classes using fuzzy c-means (FCM) clustering and a deep PCANet. The prediction of this Phase is the set of changed and unchanged superpixels. (2) Deep learning on the pixel sets over the changed superpixels only, obtained in the first phase, to discriminate real changes from false changes. SLIC is employed again to achieve new superpixels in the second phase. Low rank and sparse decomposition are applied to these new superpixels to suppress speckle noise significantly. A further clustering step is applied to these new superpixels via FCM. A new PCANet is then trained to classify two kinds of changed superpixels to achieve the final change maps. Numerical experiments demonstrate that, compared with benchmark methods, the proposed approach can distinguish real changes from false changes effectively with significantly reduced false alarm rates, and achieve up to 99.71% change detection accuracy using multi-temporal SAR imagery.

Highlights

With its cloud penetrating capability, synthetic aperture radar (SAR) images have drawn a large amount of attention, for example, in environmental surveillance, urban planning and military applications over the past decades
This indicates that change detection in areas with strong scattering is less affected by speckle noise because of the high signal-to-noise ratio (SNR)
The object-based change detection approach can effectively exploit the spatial context of neighbourhood pixels, which is conducive to increasing the ability to identify unchanged class (UC) and changed class (CC)

Summary

Introduction

With its cloud penetrating capability, synthetic aperture radar (SAR) images have drawn a large amount of attention, for example, in environmental surveillance, urban planning and military applications over the past decades. Depending on the availability of a difference image (DI), change detection approaches can be divided into two categories. One is post-classification comparison which is undertaken to identify changed and unchanged regions directly from two images that were classified independently before the analysis. In this approach, the change detection result is not influenced by radiation normalization and geometric correction. The other approach is post-comparison analysis, in which change detection is achieved by generating a DI from two multi-temporal images, and obtaining the final change map from it. The classification errors in this case do not accumulate, but the way that the DI is generated may influence the validity of the change detection results [1]

Objectives

Results

Discussion

Conclusion