基于卷积特征深度融合的海上目标跟踪算法

针对海上复杂环境下深度学习方法跟踪速度慢和尺度变化问题,以及现有跟踪算法仅使用单层深度特征或手动融合多层特征的问题,提出一种基于卷积神经网络特征深度融合的多尺度相关滤波海上目标跟踪算法。以VGG-NET-16深度模型为基础,加入多层特征融合结构,实现深度卷积融合网络,用于特征提取,通过相关滤波算法构建定位滤波器,确定目标的中心位置,通过多尺度采样构建尺度滤波器,实现对目标的判断。实验结果表明,该算法可对海上移动目标实现多尺度的有效跟踪。

Maritime target tracking algorithm based on convolutional features deep fusion

Aiming at the problems of low tracking speed and scale variation of deep learning methods in complex offshore environments,and only single-layer depth features or manual multi-layer feature fusion were used in existing tracking algorithms,a multi-scale correlation filter tracking algorithm based on convolutional neural network(CNN)feature depth fusion for sea targets was presented.A multi-layer feature fusion structure was added to implement a deep convolution fusion network for feature extraction based on the VGG-NET-16 depth model,location filters were constructed using correlation filter algorithm,the center location of the targets was determined.The scale filter was constructed by multi-scale sampling,and the target judgment was realized.Experimental results show that the proposed algorithm can achieve effective multi-scale tracking of moving targets on the sea.