基于改进的引导图像滤波和深度去噪自编码器的微弱目标跟踪算法

微弱目标易被周围环境中强烈的噪声干扰,为解决现有目标跟踪算法由于低信噪比导致跟踪准确度低的问题,提出一种将引导图像滤波器和深度去噪自编码器集成到粒子滤波器框架中的跟踪算法。通过引导图像滤波(guided image filter, GIF)算法对目标图像进行滤波处理,保留有价值的模板信息并使不准确的背景模板模糊,有效增强目标图像;通过改进的深度学习算法对深度去噪自编码器训练和微调,更好地适应目标外观变化;构造粒子分类器框架根据粒子重要性权重定位目标。实验结果表明,该算法在微弱目标跟踪准确度和抗干扰能力上优于多种现有主流跟踪算法。

Guided Image Filter and Deep Denoising Autoencoder Based Tracking for Dim Target

The dim target is easily interfered by strong noise in the surrounding environment. To solve the problem that the existing target tracking algorithm has low tracking accuracy due to low SNR, a pilot image filter and deep denoising self-encoder are integrated into the particle. Tracking algorithm in the filter framework. The target image is filtered by the GIF algorithm, the valuable template information is preserved and the inaccurate background template is blurred, and the target image is effectively enhanced; The improved deep learning algorithm pretrains the deep denoising self-encoder to better adapt to the change of target appearance in tracking process; the particle importance weight is generated by the particle classifier response to accurately locate the target position. The experimental results show that the proposed algorithm outperforms many existing mainstream tracking algorithms in tracking accuracy and anti-interference ability of weak targets.