Aiming at the problems of numerous small targets in remote sensing images, large number of target detector parameters and low detection efficiency, an improved lightweight remote sensing image small target detection model of YOLOv7-tiny was proposed. First, to address the complex fragmentation operations of the cross-stage local spatial pyramidal pooling network in the original model, a lightweight spatial pyramidal pooling structure was proposed to reduce the redundant convolution operator operations. Second, to address the problems of redundant modular connectivity of the neck network and the easy loss of spatial information of small targets in deep features, a single-scale prediction head method guided by deep semantic information was proposed to reduce the neck network and head network to reduce the computational cost of the neck network and head network. Experiments were carried out on remote sensing image datasets, and the results show that the improved model reduces the number of parameters by 49.6%, computational complexity by 28.5%, and inference speed by 73.1% compared with the original model, and outperforms other mainstream lightweight target detectors at this stage.