Remote Sensing Image Classification Exploiting Multiple Kernel Learning

We propose a strategy for land use classification, which exploits multiple kernel learning (MKL) to automatically determine a suitable combination of a set of features without requiring any heuristic knowledge about the classification task. We present a novel procedure that allows MKL to achieve good performance in the case of small training sets. Experimental results on publicly available data sets demonstrate the feasibility of the proposed approach.

I. Introduction

The automatic classification of land use is of great importance for several applications in agriculture, forestry, weather forecasting, and urban planning. Land use classification consists in assigning semantic labels (such as urban, agricultural, residential, etc.) to aerial or satellite images. The problem is closely related to that of land cover classification and differs from it mainly in the set of labels considered (forest, desert, open water, etc.). In the past, these problems have been often addressed by exploiting spectral analysis techniques to independently assign the labels to each pixel in the image [1]. Recently, several researchers have experimented with image-based techniques consisting, instead, in extracting image features and in classifying them according to models obtained by supervised learning. Multiple features should be considered because the elements in the scene may appear at different scales and orientations and, due to variable weather and time of the day, also under different lighting conditions.