Automatic recognition of man-made objects in SAR images using support vector machines

Over the past two decades the remote sensing technology is applied in a large scale in environmental research and policy, i.e. water pollution monitoring and conservation of soil, etc. The methods for recognition of man-made objects in remote sensing images are providing capabilities for mapping and monitoring crucial objects or sites in environmental management, i.e. hazardous chemicals storerooms, oil depots, etc. However, the task of recognizing key man-made objects from large images is time consuming and complex. In the paper we aim at developing an automatic and fast image processing method for the recognizing man-made objects in synthetic aperture radar (SAR) images, which is a supervised learning approach based on support vector machines. Firstly, a sample image data set which contains several classes of interested man-made objects is manually extracted from SAR images. Then we train the image data set by least squares support vector machines. After cross-validation and an exhaustive grid search, a model that can predict target label of data instances in the testing set is obtained. Finally we can implement classification in random image set using above prediction model and recognize the man-made objects. This approach needs no a priori knowledge, and only a set of train examples for the learning step is needed.