Abstract: With the development of various researches in the field of computer vision, target tracking has become more and more popular and has been widely used in all walks of life. UAV-based arget tracking has also evolved. Compared with ordinary target tracking, UAV-based target tracking has many advantages, but there are also some challenges. In view of the limited data sets, low data quality, and lack of unified data labeling in UAV-based target tracking, this paper designs a new UAV-based target tracking model based on unsupervised learning. This model improves the backbone network and tracking method of UDT model. Combining the SiamFc network structure and the unsupervised target tracking idea of UDT, the backbone network of the model is improved to an AlexNet lightweight neural network, and target tracking is achieved through forward tracking and multi-frame backward verification methods. Comparative experiment results show that the designed model is better than the model before improvement and other classic tracking models.

Key words: UAV, target tracking, unsupervised learning, siamese network