A Method for Detecting Small Slow Targets on Sea Surface Based on FC-MCNN

Abstract

Detecting small slow targets on sea surface has always been a hot topic in radar target detection. Classical target detection methods based on single input mode are particularly prone to losing the phase information of the radar echo signal, which affects the detection performance of the radar. To solve the problem of phase information loss, a False alarm Control method of Multi-input Convolutional Neural Network (FC-MCNN) is proposed in this article. First, several classic Convolutional Neural Network (CNN) models are compared and analyzed. Then, a Multi-input Convolutional Neural Network (MCNN) is designed to extract more useful information from radar echo signal. Simultaneously, in order to improve the generalization performance of the model and control the false alarm probability, a controllable false alarm Support Vector Machine (SVM) is used to modify the final output module. In addition, the datasets are also preprocessed before being input into the proposed model. Finally, the performance of the proposed method is verified by the IPIX measured data sets. The results show that the proposed method has a higher detection probability and better computational performance than the traditional CNN detection methods.

Keywords

• Target detection
• constant false alarm rate (CFAR)
• signal processing
• convolutional neural network (CNN)
• support vector machine (SVM)