基于改进Yolov3 算法的舰船目标检测识别系统

针对复杂战场环境下对海目标检测识别的需求，设计了一种基于改进Yolov3 算法的海面舰船目标实时检测识别系统。使用微调分类网络、增加训练尺度、聚类目标边框维度、二级特征分类等方法对Yolov3 检测识别网络模型进行了优化，在提高识别精度的同时有效降低了漏检率和虚警率。实验结果表明，优化后的网络模型在自建的舰船图像数据库中将检测识别平均准确率提高到了79.3%，对真实海上航拍视频中舰船目标识别的平均准确率达到了81% 以上。

Improved Yolov3 Based Ship Target Detection and Recognition System

For the needs of sea target detection and recognition in complex battle field environment, an improvedYolov3-algorithm-based real-time detection and recognition system is proposed for marine ship targets. Several methodsare utilized to optimize the detection network model of the system, including network fine-tuning, training scalesincreasing, target box clustering and two-phase feature classification. These optimizations can effectively reduce themissed detection rate and false alarm rate. Experimental results show that the designed system raises the accuracy ofdetection and recognition to 79.3% on ship image database built, and achieves the mean accuracy on real aerial videosof marine ship scenes over 81%.