Abstract: X-band radar data are associated with severe background and clutter interference, target deformation, as well as obscure target and mutual interference. These issues often result in the target being missed or wrongly detected. In this study, we propose a multiple extended target optimization tracking algorithm based on a multi-kernel correlation filtering (MKCF). The proposed algorithm comprises both threshold and watershed methods for global and local detection of X-band radar targets, respectively. Then, Kalman filtering (KF) is introduced for adaptive identification and tracking of survival targets, newborn targets and disappearing targets. The KF includes multi-frame velocity information and the distance measure fused cross-parallel ratio. In addition, we propose a template optimization method for MKCF that can convert the template creation time into an adaptive weight and optimize the template fusion with respect to great likelihood. This template optimization method can effectively adapt to multiple extended target tracking with complex deformation; it can also adapt to the re-identification of missing targets. Our experimental results show that the proposed algorithm has high tracking robustness for X-band radar multi-target tracking.