Automatic Classification for Ground Targets under Complex Background Based on Bag of Words Model

In order to improve the performance of the automatic classifier for ground targets under complex background, a new classifier which was based on bag of words model and support vector machine (SVM) was build. Firstly, the keypoints of a target image were extracted into vectors by using Scale-Invariant Feature Transform algorithm, and vectors which were extracted from all trained images were merged into large vectors. Secondly, the large vectors were clustered into small vectors which included K vectors by using K -means clustering algorithm. To analyze how different values of K affected the classifying results, K was set to several different numbers. Thirdly, vectors which were extracted from all trained images were classified, and K -dimensional numerical vectors were obtained. Finally, the classifier was build with SVM algorithm, and the K -dimensional vectors were as input features of the classifier. The training images were randomly chosen from 90% of images dataset which included missile launch vehicles and tanks, and the rest 10% were as the testing images. The trained classifier is tested, and the f1-score value is 0.79. The results show that the new classifier can perform well.