Semi-supervised Learning with Multilayer Perceptron for Detecting Changes of Remote Sensing Images Swarnajyoti Patra, 1 Susmita Ghosh 1,⋆ , and Ashish Ghosh 2 1 Department of Computer Science and Engineering Jadavpur University, Kolkata 700032, India susmitaghoshju@gmail.com 2 Machine Intelligence Unit and Center for Soft Computing Research Indian Statistical Institute 203 B. T. Road, Kolkata 700108, India Abstract. A context-sensitive change-detection technique based on semi-superv-ised learning with multilayer perceptron is proposed. In order to take contextual information into account, input patterns are generated considering each pixel of the difference image along with its neighbors. A heuristic technique is suggested to identify a few initial labeled patterns without using ground truth information. The network is initially trained using these labeled data. The unlabeled patterns are iteratively processed by the already trained perceptron to obtain a soft class label. Experimen- tal results, carried out on two multispectral and multitemporal remote sensing images, confirm the effectiveness of the proposed approach. 1 Introduction In remote sensing applications, change detection is the process of identifying dif- ferences in the state of an object or phenomenon by analyzing a pair of images ac- quired on the same geographical area at different times [1]. Such a problem plays an important role in many different domains like studies on land-use/land-cover dynamics [2], burned area assessment [3], analysis of deforestation processes [4], identification of vegetation changes [5] etc. Since all these applications usually require an analysis of large areas, development of automatic change-detection techniques is of high relevance in order to reduce the effort required by manual image analysis. In the literature [2,3,4,5,6,7,8,9,10], several supervised and unsupervised tech- niques for detecting changes in remote-sensing images have been proposed. The supervised methods need “ground truth” information whereas the unsupervised approaches perform change detection without using any additional information, besides the raw images considered. Besides these two methods of learning, an- other situation may arise where only a few training patterns are available. The semi-supervised learning [11] comes into play in such a situation. In this article ⋆ Corresponding author. A. Ghosh, R.K. De, and S.K. Pal (Eds.): PReMI 2007, LNCS 4815, pp. 161–168, 2007. c  Springer-Verlag Berlin Heidelberg 2007