一种基于深度学习的舰船目标融合识别算法

A Ship Target Fusion Recognition Algorithm Based on Deep Learning

低分辨率SAR(合成孔径雷达)卫星遥感图像具有大幅宽特点,是广域海上目标搜索的重要手段,但相较于高分辨率遥感图像,可用于目标识别的特征较少,导致识别性能有待提高。论文探索并设计了一种基于深度学习的低分辨率多极化SAR卫星遥感图像舰船目标融合识别的算法。首先,以多极化SAR图像为输入,利用改进的VGG16网络提取图像特征得到各极化通道的分类结果,同时自动学习各极化通道的权重,然后,对分类结果进行决策级加权融合;最后,利用公开的OpenSARShip数据集进行实验验证,结果表明,与单极化SAR图像舰船目标识别以及其他融合识别方法相比,该算法一定程度上提升了舰船目标识别效果。

Low-resolution SAR(Synthetic Aperture Radar)satellite remote sensing images have a Large width feature,which is an important means of wide-area maritime target search.However,compared with high-resolution remote sensing images,there are fewer features for target recognition,resulting in recognition performance to be improved.This paper explores and designs an al⁃gorithm for ship target fusion recognition of low-resolution multi-polarization SAR satellite remote sensing image based on deep learning.Firstly,using the multi-polarized SAR image as input,the improved VGG16 network is used to extract the image features to obtain the classification results of each polarization channel,and the weight of each polarization channel is automatically learned.Then,the classification result is subjected to decision-level weighted fusion,the experimental results are verified by the open Open⁃SARShip dataset.The results show that compared with the single-polarized SAR image ship target recognition and other fusion rec⁃ognition methods,the algorithm of this paper improves the ship target recognition effect to some extent.