UAV and remote sensing image stitching has significant value in agricultural monitoring, environmental surveillance, and urban inspection. However, traditional stitching methods often suffer from visual artifacts caused by imperfect seam-line selection, as well as inefficiency and limited precision in complex scenarios. This paper proposes a knowledge-driven seamline optimization framework that integrates superpixel segmentation, edge saliency analysis, and shortest-path algorithms. First, SLIC superpixel segmentation partitions images into homogeneous regions, reducing computational redundancy through semantic-aware knowledge modeling and improving computational efficiency through adaptive superpixel seed distribution, priority queue acceleration, and parallelization. Second, edge saliency analysis extracts high-frequency structural features as prior knowledge to avoid seamlines crossing prominent edges. Finally, a multi-feature fusion mechanism combining color consistency, edge significance, and spatial continuity guides the Dijkstra shortest-path algorithm to optimize seamline placement for natural transitions, enabling knowledge-based integration of heterogeneous visual cues. Furthermore, the proposed method incorporates a parallelized shortest-path search and dynamically adjusted superpixel constraints to further refine seamline placement. Experimental results on multiple high-resolution remote sensing and UAV datasets demonstrate that the method not only enhances the perceptual quality of stitched images but also effectively reduces visible artifacts.