Objectives The target information error of forced reconnaissance is large, and the course is variable. This leads to a decrease in target maneuver detection performance and affects the analysis of the target intention. Therefore, this paper proposes a maneuvering target detection method based on prior knowledge.

Methods The method introduces two prior knowledge through solidifying expert experience. The first prior knowledge is that there is a significant difference in the target heading before and after maneuvering, while the target heading is approximately the same during non-maneuvering periods. The second prior knowledge is that the difference in heading between before and after maneuvering has a local extremum. The position of the maneuver inflection point in the trajectory always maximizes the difference between adjacent sub-trajectories. Based on the definition of trajectory smoothness measurement, a calculation method for course maneuver evaluation factor based on principal component analysis（PCA） is proposed. By evaluating the heading maneuver factor, a preliminary screening of maneuver inflection points can be conducted. In order to find trajectory points that satisfy the second prior knowledge, a maneuver inflection point screening method based on maximum filtering is proposed. The trajectory point that satisfies both the first prior knowledge and the second prior knowledge is the maneuvering position.

Results The simulation results indicate that compared with mainstream interacting multiple model and information entropy-based algorithms, the proposed method provides more accurate detection results for target maneuvering detection with the least missed detections and the smallest distance error when compressing the trajectory.

Conclusions The results of this study prove the progressiveness of the proposed algorithm, which can effectively improve the accuracy and robustness of maritime target maneuver detection and provide strong support for maritime target behavior analysis and command decision-making.