Contour matching for multisource terrain data based on morphological classification and unsupervised learning

Contour matching is a key link for multi-source terrain data integration, fusion, and updating, crucial for improving terrain data quality and terrain-aided navigation accuracy. However, existing studies mostly adopt a single similarity metric model and matching strategy, which is difficult to effectively deal with the significant differences in spatial location, geometry, and scale expression of heterogeneous data, resulting in the inability to accurately obtain the similarity score and matching relationship. Therefore, an unsupervised contour matching method based on morphological classification is proposed in this study. First, contours are classified into three categories (open, closed, and truncated) according to their morphological features, and differentiated similarity metrics and matching strategies are designed for the different categories to minimize interference from heterogeneous data. Subsequently, an unsupervised learning mechanism is introduced to determine the contour matching relationship based on the similarity score distribution, avoiding the need to set similarity thresholds. When applied to two experimental areas, the proposed method significantly outperforms the comparison method in several evaluation metrics, with F1 values improving from 1.26% to 67.96%. Notably, the proposed method shows higher matching accuracy and robustness for contour matching problems with significant morphological differences and scale interference.