基于形态自适应网络的无人机目标跟踪方法

针对无人机影像目标跟踪过程中常出现的目标方向变化、目标遮挡变化、样本多样性不足等问题，提出了一种基于形态自适应网络的无人机航空影像目标跟踪算法。首先使用基于数据驱动的方法对数据集进行扩增，添加了遮挡样本和多旋转角度样本，提高样本多样性；提出的形态自适应网络模型通过旋转不变约束改进深度置信网络，提取强表征能力的深度特征，使得模型能够自动适应目标形态变化，利用深度特征变换算法获取待检测目标的预定位区域，采用基于Q学习算法的搜索机制对目标进行自适应精准定位，使用深度森林分类器提取跟踪目标的类别信息，得到高精度的目标跟踪结果。在多个数据集上进行了对比实验，实验结果表明该算法能够达到较高的跟踪精度，可以适应目标旋转、目标遮挡等形态变化情况，具有较好的准确性和鲁棒性。

Adaptive morphological network based UAV target tracking algorithm

To solve the problems such as the change of target direction, change of target occlusion and lack of sample diversity in the process of target tracking based on UAV images, this paper proposes a UAV aerial image target tracking algorithm based on the adaptive morphological network. First, the data-driven method is used to expand datasets, and multi rotation angle samples and occlusion samples are added to improve the diversity of samples. The proposed adaptive morphological network improves the deep belief network by rotating invariant constraints to extract deep features with strong representativeness, which enables the model to automatically adapt to the changes of target morphology. The deep feature transformation algorithm is used to obtain the pre-location area of the target to be detected. The target is located adaptively and accurately by the search agent based on the Q-learning algorithm. The category information of the tracking target is extracted by using the deep forest classifier, and the target tracking results with high precision are obtained. Comparative experiments are then carried out on several datasets. The experimental results show that the algorithm can achieve high tracking accuracy, adapt to the change of target angle and occlusion, and has good accuracy and robustness.