Using VGG16 to Military Target Classification on MSTAR Dataset

Abstract

Synthetic aperture radar has the characteristics of all-weather, all-weather, long range, high resolution, etc., and has played an important role in the fields of battlefield reconnaissance, detection and guidance. Target recognition technology based on SAR images, especially ground military target recognition technology, has received widespread attention. The MSTAR dataset is composed of SAR images of ground stationary targets provided by the Defense Advanced Research Projects Agency (DARPA) and the Air Force Research Laboratory (AFRL), including civilian and military targets. The convolutional neural network consists of a series of convolutional layers, pooling layers and fully connected layers. It can obtain effective feature representation from big data and recognize it through automatic learning, eliminating the complicated feature extraction algorithm and feature matching process. Now it has been widely used in the field of target interpretation. Experiments show that using the existing neural network VGG16 to classify military targets on the MSTAR data set can obtain good classification accuracy.