An Improved CFAR_CNN Algorithm for Ship Target Detection

Abstract

The remote sensing monitoring of ships has important significance in military and civilian fields. The target of remote sensing ship detection is to locate the position of the ship in the remote sensing image and extract its features. For SAR (Synthetic Aperture Radar) remote sensing data, traditional ship target detection algorithms have been difficult to meet the needs of speed and accuracy. With the development of artificial intelligence technology, target detection technologies represented by deep learning algorithms are great progress has been made in the ship inspection of images, and it is also a research hotspot in this field. The constant false alarm rate (CFAR) algorithm is prone to false alarms for areas such as land, islands, and azimuth blur. The CNN (Convolutional Neural Network) model has a strong classification ability and can classify targets and non-targets. In this paper, we propose an improved CFAR_CNN model. CFAR is used to detect the candidate ship samples, and the trained CNN model classifies the candidate ship samples, retains the ship targets, and removes the non-vessel targets. The combination of these two algorithms is the CFAR_CNN algorithm proposed in this article. The CFAR-CNN algorithm for SAR remote sensing images is mainly divided into two stages. The CFAR algorithm is responsible for the first stage of ship detection, and the CNN algorithm is used to remove false alarms generated by the CFAR algorithm. Through the training and testing of the Gaofen-3 satellite data, the CFAR_CNN algorithm has obvious potential compared with the CFAR algorithm alone.