Traditional manual mapping methods face crucial challenges in ensuring favorable observation conditions in difficult mapping areas in the Antarctic. This study proposes a large-scale block adjustment (BA) method without ground control points (GCP) using high-resolution optical satellite remote sensing images. This method mainly includes constructing an adjustment model based on virtual control points (VCPs), implementing a multilevel gross error detection and elimination strategy, and providing a robust and fast solution of large-scale BA equations accelerated by a GPU device. To validate the effectiveness of the proposed method, an experimental analysis was conducted using 39 images obtained from the ZiYuan-3 satellite images in the Antarctic. The experimental results indicate that sparse matrix storage technology can effectively reduce memory requirements, and the GPU parallel computing method can address efficiency problem in solving large-scale adjustment parameters. The accuracy of the proposed method reaches the subpixel level after the BA procedure, fully meeting the requirements of cartographic mosaicing.