Traditional target tracking algorithms rely on a single template update method, wihich fails to handle severe deformation and occlusion, thus limiting their adaptability in challenging scenarios. Therefore, we propose an object tracking method based on spatial inhibition and trajectory association optimization. First, the improved ResNet-50 network is used in the feature extraction part of the algorithm. Spatial inhibition attention (SIA) is introduced. By using an energy function, the weight allocation of neurons exhibiting spatial inhibition is optimized to enhance attention to target features. Then, a selective query recall (SQR) strategy is added. The training focuses on the later stages of Transformer training to avoid a lack of training focus in the decoder and prevent cascading errors in the multi-layer decoding sequence. Furthermore, we construct a long-short-trajectory (LST) framework, and introduce a degree of trajectory correlation (DTC). Based on the traditional template update algorithm, a DTC prediction head is proposed. The global trajectory association is modeled through the minimum cost flow (MCF) algorithm, and the continuous quadratic programming constraints are used for back-propagation to update the trajectory reference template. Finally, the proposed method is experimentally evaluated on public datasets such as GOT-10k, LaSOT, TrackingNet, NfS30, UAV123, and OTB100, and the results verify its effectiveness.