Publication

Abstract : Target classification is an important application in modern radar systems. Recognition of targets in radar data is a key task for surveillance and monitoring the areas such as military. Automatic classification systems is becoming one of the challenges in modern era. The feature extractor is often hand-designed with domain knowledge in machine learning and can have a significant impact on classification accuracy. Convolutional neural network (CNN) have recently achieved best results through the automatic learning of feature hierarchies from massive training data in many applications such as computer vision and speech recognition tasks. We build a four layer convolutional neural network for the target classification task. The Dataset used for radar classification is MSTAR (Moving and Stationary Target Acquisition and Recognition) which consists of 8 class targets. The data collected from different angles like 15°, 17°, 30° and 45° from each class is considered. This paper shows that there is increase in the performance of target classification by building the convolutional neural network (CNN). We achieved 95% accuracy in the target classification without any trivial preprocessing and feature extraction mechanism. The Support vector machine (SVM) in the machine learning algorithm was able to achieve 86.9% classification performance with bag of words features. Furthermore our CNN result with any preprocessing steps shows that it can be used in real time applications like civilian and military purposes.