Radar Target Discrimination with F-CNN: Fewer Parameters and Higher Accuracy

In order to discriminate the real targets, the clutter and the dense multi-false targets, we propose a factorized convolutional neural network-based algorithm for radar targets discrimination. We establish the factorized convolutional neural network model with depthwise separable convolution. To reduce the parameters of the model, we establish the simplified factorized convolutional neural network by reducing the numbers of both convolutional filters and connection nodes of fully connected layers. The result of the measured data demonstrates that, as compared with the existing model, the simplified factorized convolutional neural network has higher discrimination rate for the real targets, the clutter and the dense multi-false targets, and its parameters are less than ten percent counterpart of a recent proposed model.