面向无人艇的航海雷达与光电吊舱协同环境感知方法

为了提升无人艇对典型水面小目标的感知能力,本文对面向无人艇的航海雷达与光电吊舱协同环境感知方法进行了研究。首先,本文采用了高斯滤波和形态学滤波进行航海雷达图像处理;其次,本文研究了深度学习目标检测算法,提出了基于多尺度卷积融合结构和空间注意力加强的改进型单个深层神经网络(SSD)目标检测算法模型,提升对弱纹理小目标特征保持性,并在VOC2007数据集和典型水面场景下进行了验证。此外,本文提出了航海雷达与光电吊舱联合目标感知流程,对航海雷达感知目标分布图和光电吊舱感知目标分布图进行融合处理,得到最终的目标分布图,包括目标类别、方位、距离等信息。实验表明,本文的改进算法达到75.3%的平均准确率(mAP),同时保持了64.4 FPS的实时检测速度,航海雷达与光电吊舱能协同有效地对海面目标进行探测。

Cooperative environment perception method for nautical radar andphotoelectric pod facing to unmanned surface vehicles

In order to improve the perception ability of Unmanned Surface Vehicles ( USV) to typical small water surface targets,cooperative environment perception method for nautical radar and photoelectric pod facing to Unmanned Surface Vehicles is studied inthis paper. Firstly, Gaussian filtering and morphological filtering are used to process the image of the nautical radar; Secondly, the deeplearning target detection algorithm is studied and an improved single shot multibox detector ( SSD) target detection algorithm model isproposed based on multi-scale convolution fusion structure and spatial attention enhancement. The improvement was designed to improvethe feature retention of small target with weak texture. The proposed method was verified in the VOC207 data set and typical watersurface scene. In addition, the paper proposes the flow chart of joint target perception with nautical radar and photoelectric pod. Theperception target distribution maps of nautical radar and photoelectric pod are fused to get the final target distribution map, whichincludes the information such as categories, orientations, distances, etc. of the targets. The experiment shows that the improvedalgorithm reaches the mean average precision (mAP) of 75. 3% and maintains the real-time detection speed of 64. 4 FPS. Besides, thenautical radar and photoelectric pod can detect the sea target cooperatively and effectively.