Target tracking methods based on UAVs (unmanned aerial vehicles) face huge challenges, such as target blockage, viewing angle changes, scale changes and background noise, that can easily lead to low target tracking accuracy and poor robustness. In addition, the traditional neural network online and offline tracking methods easily lead to poor tracking real-time performance because of the limitations of drone target tracking computing. To address the above issues, the adaptive target tracking algorithm for UAVs based on a pseudo twin network (ATTPTN) is proposed. In the ATTPTN methods, the asymmetric pseudo twin network framework consists of the target tracking branch and the template library branch, and target tracking and template updating and selection tasks are performed. Because the fitting ability of the initial target model is gradually weakened, the template library uses a scoring strategy to selectively store the template features intermittently and shares the template input tracking branch with the target feature model for subsequent tracking. The target tracking branch adaptively completes the target tracking task based on an improved kernel density probability estimation method. The experimental simulation results show that compared with the KCF (Kernelized Correlation Filters), DSST (Discriminative Scale Space Tracker), the proposed ATTPTN method routing protocol is superior in terms of accuracy, robustness and expected average overlap.