We propose the use of a denoising autoencoder to improve radar resolution and target detection probability in noise-contaminated range- Doppler diagrams. Conventionally, target detection has been performed using constant false alarm rate (CFAR) algorithms, which are not ideal in high-clutter environments. Consequently, various deep learning algorithms have been suggested to improve target detection. In this paper, a denoising autoencoder based on a convolutional neural network is proposed to enhance radar resolution and target detection probability beyond what is possible with CFAR and existing deep learning algorithms. Departing from the conventional autoencoder training approach, we use noise-contaminated radar images as input data and noiseless radar images as output data in the training process. The denoising autoencoder removes the noise from the range-Doppler diagram and increases the strength of the target signals while preserving accurate location targeting, thereby improving the target detection probability. Range resolution and Doppler resolution are also enhanced. We compare the performance of the denoising autoencoder with that of CFAR algorithms using the receiver operating characteristic curve, demonstrating that the autoencoder outperforms the algorithms.
Kim et al. (Tue,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: