Similarity Analysis of Simulated SAR Target Images

Target detection and recognition is the core application of radar image interpretation. It suffers from a shortage of training data in practical. A natural idea is to enrich the training data by the simulated samples. A great many generative models were presented in the preceding works. Therefore, how to evaluate the generated images is to be an urgent problem. Very few studies have focused on this challenging problem previously. To fill the blank, an objective evaluation method based on fuzzy comprehensive evaluation (FCE) is proposed in this paper, which evaluates simulated images in the perspective of statistics. This method combines Fisher separability criterion, FCE and image feature extraction. Multiple experiments are performed on real Synthetic Aperture Radar(SAR) dataset. The experimental results demonstrate the effectiveness of the proposed method.