Target Detection in Sea Clutter Based on Transfer Learning

Abstract

Weak target detection in sea clutter has been a popular research topic in the field of radar target detection. Lack of real data, underused historical data, interference of various factors, and high cost of collecting real data are the common problems in handling sea clutter data. This study aims to address the problems by introducing transfer learning as a new method. Transfer learning utilizes knowledge from previously collected data and several new samples to significantly improve the target detection results. The proposed method combines TrAdaBoost and support vector machine. To detect the targets, we perform the method by using real sea clutter dataset from 1993 (source data) and 1998 (target data). In addition, three types of target datasets are used to test the accuracy of the method. The accuracy rates are higher than 70%. The results show that targets in sea clutter can be effectively observed and detected with the proposed method. The performance of the proposed method is better than that of the target detection method that uses the task dataset only.