Abstract: Ultra-large-scale remote sensing imagery has become fundamental data sources for digital earth applications, high-resolution mapping, cultural heritage preservation, industrial inspection, and large-scale environmental monitoring. However, the enormous data volume of such images poses significant challenges to realtime visualization, interactive browsing, and efficient memory management. Conventional image visualization approaches either require loading the entire image into memory or rely on external pyramid files, resulting in excessive memory consumption, long initialization time, additional storage overhead, and unsatisfactory user interaction when processing ultra-large images. To address these issues, this paper proposes a efficient visualization method for ultra-large-scale remote sensing imagery based on a pyramid mapping mechanism and multithreaded interaction. First, an image pyramid is adaptively constructed according to the physical dimensions of the display viewport, enabling resolution-aware visualization across different zoom levels. The pyramid data are appended to the end of the original image Read-Only Memory (ROM) address space without modifying the original image structure, thereby preserving data integrity while supporting efficient hierarchical access. Next, the display viewport is partitioned into multiple independent sub-areas to support progressive rendering. The top-level pyramid image is used to initialize the display content, and a geometric mapping relationship between the pyramid images and the original high-resolution image is established. Based on the spatial intersection between pyramid tile images and the display viewport, only the required image blocks are dynamically scheduled and loaded, effectively reducing redundant disk access and improving response efficiency. Furthermore, a multithreaded framework for data loading and rendering is introduced to enhance interactive performance. The required pyramid tiles are asynchronously loaded from ROM into Random Access Memory (RAM), while the corresponding image blocks stored in RAM are progressively rendered to refresh each sub-area of the display viewport until the full-resolution visualization is completed. This strategy overlaps input/output operations with rendering processes and significantly improves the smoothness of zooming and panning interactions. Comparative experiments were conducted using three ultra-large remote sensing images and four widely used professional remote sensing software packages, namely Inpho, ENVI, ArcGIS, and QGIS. Experimental results demonstrate that the proposed method achieves the smoothest zooming and panning experience while maintaining relatively low memory consumption. Compared with conventional solutions, the storage overhead required for maintaining pyramid information is less than 1.2%, and no additional temporary pyramid files are generated. The proposed method also substantially reduces hardware requirements for interactive visualization of ultra-large images while providing faster view initialization, and more efficient human-computer interaction. These advantages make the proposed approach particularly suitable for rapid visualization and interactive analysis of gigapixel and even terabyte-scale remote sensing imagery on commodity computing platforms.