A Comparison Of Machine Learning Methods For Target Recognition Using Isar Imagery

The ability to accurately classify targets is critical to the performance of automated/assisted target recognition (ATR) algorithms. Supervised machine learning methods have been shown to be able to classify data in a variety of disciplines with a high level of accuracy. The performance of machine learning techniques in classifying ground targets in two-dimensional radar imagery were compared. Three machine learning models were compared to determine which model best classifies targets with the highest accuracy: decision tree, Bayes', and support vector machine. X-band signature data acquired in scale-model compact ranges were used. ISAR images were compared using several techniques including two-dimensional cross-correlation and pixel by pixel comparison of the image against a reference image. The highly controlled nature of the collected imagery was ideally suited for the inter-comparison of the machine learning models. The resulting data from the image comparisons were used as the feature space for testing the accuracy of the three types of classifiers. Classifier accuracy was determined using N-fold cross-validation.