Radar Target Recognition and Classification Based on Preprocessing and Improved DenseNet

Radar, as the main means of detecting and monitoring ground targets, is widely employed in the security field. In order to address the issue that existing practical applications of radar target categorization technology rely too heavily on manual feature extraction, this paper proposes a classification model based on radar raw data and improved DenseNet network. The processing of raw radar data is based on Fourier transform and constant false alarm detection to build a new data set. Secondly, attention mechanism is introduced on the basis of DenseNet, and then PReLU activation function is used to improve the network. The preprocessed radar data is used as the input of the network and their categories (car, people, cyclist or clutter) are predicted. The results of experimental show that the proposed improved model achieves an accuracy of 96.64% on the relevant dataset, which can improve the accuracy compared to the original DenseNet and other models.