Abstract: A contextual visual-based synthetic-aperture radar (SAR) target detection method and apparatus, and a storage medium, belonging to the field of target detection is described. The method includes: obtaining an SAR image; and inputting the SAR image into a target detection model, and positioning and recognizing a target in the SAR image by using the target detection model, to obtain a detection result. In the present disclosure, a two-way multi-scale connection operation is enhanced through top-down and bottom-up attention, to guide learning of dynamic attention matrices and enhance feature interaction under different resolutions. The model can extract the multi-scale target feature information with higher accuracy, for bounding box regression and classification, to suppress interfering background information, thereby enhancing the visual expressiveness.