A novel approach for change detection of remotely sensed images using semi-supervised multiple classifier systemInformation Sciences

About

Authors
Moumita Roy, Susmita Ghosh, Ashish Ghosh
Year
2014
DOI
10.1016/j.ins.2014.01.037
Subject
Artificial Intelligence / Computer Science Applications / Information Systems and Management / Software / Theoretical Computer Science / Control and Systems Engineering

Similar

Text

Article history:

Received 17 July 2013

Received in revised form 23 December 2013

Accepted 25 January 2014

Available online 14 February 2014

In this article, a novel approach using ensemble of semi-supervised classifiers is proposed e formed, one for lassified in e they can acquisition

Various methodologies exist in the literature to carry out supervised change detection, e.g., post classification [1,8,18] multi-date classification [1], kernel based methods [9], etc. Having several advantages, applicability of supervised m in change detection is poor due to mandatory requirement of sufficient amount of ground truth information, collec which is expensive, hard and monotonous too. On the contrary, in unsupervised approach [10–17], there is no need of http://dx.doi.org/10.1016/j.ins.2014.01.037 0020-0255/ 2014 Elsevier Inc. All rights reserved. ⇑ Corresponding author. Tel.: +91 33 2575 3110/3100; fax: +91 33 2578 3357.

E-mail address: ash@isical.ac.in (A. Ghosh).

Information Sciences 269 (2014) 35–47

Contents lists available at ScienceDirect

Information SciencesChange detection can be viewed as an image segmentation problem, where two groups of pixels are to b the changed class and the other for the unchanged one. Process of change detection can be broadly c categories: supervised [7–9] and unsupervised [10–17]. Supervised techniques have certain advantages lik itly recognize the kind of changes occurred and are robust to different atmospheric and light conditions ofto two explicdates. , direct ethods tion of1. Introduction

Change detection is a process of detecting temporal effects of multi-temporal images [1,2]. This process is used for finding out changes in land covers over time by analyzing remotely sensed images of a geographical area captured at different time instants. Changes can occur due to natural hazards (e.g., disaster, earthquake), urban growth, deforestation, etc. [1–5].

Change detection is one of the most challenging tasks in the field of pattern recognition and machine learning [6].Keywords:

Change detection

Elliptical basis function neural network

Fuzzy k-nearest neighbor classifier

Multilayer perceptron

Ensemble classifier

Semi-supervised learningfor change detection in remotely sensed images. Unlike the other traditional methodologies for detection of changes in land-cover, the present work uses a multiple classifier system in semi-supervised (leaning) framework instead of using a single weak classifier.

Iterative learning of base classifiers is continued using the selected unlabeled patterns along with a few labeled patterns. Ensemble agreement is utilized for choosing the unlabeled patterns for the next training step. Finally, each of the unlabeled patterns is assigned to a specific class by fusing the outcome of base classifiers using some combination rule.

For the present investigation, multilayer perceptron (MLP), elliptical basis function neural network (EBFNN) and fuzzy k-nearest neighbor (k-nn) techniques are used as base classifiers. Experiments are carried out on multi-temporal and multi-spectral images and the results are compared with the change detection techniques using MLP, EBFNN, fuzzy knn, unsupervised modified self-organizing feature map and semi-supervised MLP. Results show that the proposed work has an edge over the other state-of-the-art techniques for change detection.  2014 Elsevier Inc. All rights reserved.A novel approach for change detection of remotely sensed images using semi-supervised multiple classifier system

Moumita Roy a, Susmita Ghosh a, Ashish Ghosh b,⇑ aDepartment of Computer Science and Engineering, Jadavpur University, Kolkata, India bCenter for Soft Computing Research, Indian Statistical Institute, Kolkata, India a r t i c l e i n f o a b s t r a c t journal homepage: www.elsevier .com/locate / ins additional information like ground truth. Due to depletion of labeled patterns, unsupervised techniques seem to be compulsory for change detection. Unsupervised change detection process can be of two types: context insensitive (spectral based) [1,12] and context sensitive (spatial based) [10,11,13–16,19].

In change detection, it may so happen that the category information of a few labeled patterns could be collected easily by experts [20]. However, if the number of these labeled patterns is small, then this information may not be sufficient for developing any supervised method. In such a scenario, knowledge of labeled patterns, though not much in amount, may be completely unutilized if unsupervised approach is carried out. Under this circumstance, semi-supervised approach [21,22] can be opted instead of unsupervised or supervised ones. Semi-supervision uses a small amount of labeled patterns with abundant unlabeled ones for learning, and integrates the merits of both supervised and unsupervised strategies to make full utilization of the collected patterns. Semi-supervision has been used successfully for improving the performance of clustering and classification [23–26] when sufficient amount of labeled data are not present.

Semi-supervised approaches were explored for the use of multiple classifier system (MCS) [27–30]. Many applications in real life domains, i.e., change detection, medical image analysis, face recognition suffer from the problem of unavailability of labeled information. Therefore, semi-supervised MCS are required and have been studied in past [27–30] (see Table 1). As to the knowledge of the authors, no such applications exists in change detection domain using semi-supervised MCS. This motivated us to explore the capacity of ensemble classifier embedded with semi-supervision framework to improve the performance of change detection process when a few labeled patterns are available.

In the proposed method, merits of both semi-supervised learning and ensemble learning are integrated in a single platform for detecting changes from remotely sensed images. The traditional algorithms [1,8,9,18,31] for change detection is mainly relaying on a single classifier in either supervised or semi-supervised framework. Unlike this, in the present work, a set of semi-supervised classifiers is used for change detection. In the present investigation, multilayer perceptron (MLP) [32], elliptical basis function neural network (EBFNN) [32–34] and fuzzy k-nearest neighbor techniques (k-nn) [35] are used 2. The proposed algorithm 36 M. Roy et al. / Information Sciences 269 (2014) 35–47In the present work, an ensemble of semi-supervised classifiers is proposed for change detection. The contribution of the present work is twofold: at first an algorithm is designed to integrate semi-supervised learning and ensemble learning in a single platform and then the proposed algorithm is used for the betterment of change detection process when a few labeled patterns are available. Unlike the other state-of-the-art techniques in the literature of semi-supervised multiple classifier system (i.e. co-training [29], tri-training [30], co-forest [44]), the proposed algorithm during the iterative learning process utilizes the agreement between all the networks in the ensemble for collecting the most confident labeled patterns.