Abstract

Abstract: Changedetection (CD) in multitemporal images is an important application of remote sensing.Recent technological evolution provided very high spatial resolution (VHR)multitemporal optical satellite images showing high spatial correlation amongpixels and requiring an effective modeling of spatial context to accurately capturechange information. Here, we propose a novel unsupervised context-sensitive frameworkdeep change vector analysis (DCVA)for CD in multitemporal VHR images thatexploit convolutional neural network (CNN) features. To have an unsupervised system,DCVA starts from a suboptimal pretrained multilayered CNN for obtaining deepfeatures that can model spatial relationship among neighboring pixels and thuscomplex objects. Deep change vectors are analyzed based on their magnitude toidentify changed pixels. Then, deep change vectors corresponding to identify changedpixels are binarized to obtain a compressed binary deep change vectors thatpreserve information about the direction (kind) of change. Changed pixels areanalyzed for multiple CD based on the binary features, thus implicitly usingthe spatial information.

Keywords: Changedetection (CD), Deep Change Vector Analysis (DCVA), Deep Features,Multi-temporal Images, Remote Sensing, Very High-Resolution Images.

------------------------------------------------------