Integration of CNN and OS-CFAR for Enhanced Radar Target Detection

Radar images are important in target detection in both military and civilian fields, but interference from clutter and noise significantly degrades detection accuracy. Traditional methods such as Constant False Alarm Rate (CFAR) algorithms have limited effectiveness in clutter suppression and perform poorly especially in complex environments. In this paper, we propose an innovative approach that combines traditional radar signal processing techniques with deep learning to enhance the performance of target detection in radar images. We design a specialized convolutional neural network (CNN) architecture for predicting target response images from planar position indicating (PPI) radar images and combine it with Ordered Statistic CFAR (OS-CFAR) algorithm to achieve robust clutter removal. The dataset consists of 626×626 pixel PPI radar images, generated by polar to Cartesian coordinate transformation, containing the input images and the corresponding ground truth response images. The main contributions include (1) proposing a CNN architecture for radar image target detection, (2) integrating the OS-CFAR algorithm with deep learning to effectively suppress clutter, and (3) verifying the superiority of the method through performance evaluation and visualization. Experimental results show that the method significantly improves the detection accuracy and robustness in complex clutter environments.