In order to solve the problem of a complex target background which is hard to identify, as well as a complex target size in remote sensing image target detection task, a remote sensing image detection optimization model based on YOLOv5 has thus been proposed. Firstly, a mosaic enhancement is carried out on the input data to increase the diversity of samples; meanwhile, an adaptive anchor frame calculation is used to obtain the optimal initial value of the anchor frame. Then the feature layer extracted through the backbone network is to be fused, thus obtaining the optimal feature layer, to be followed by an optimization of the positioning loss, with CIoU loss used as the positioning loss function and Focal loss as the classification loss. Finally, adaptive image scaling is used for a reduction in information redundancy and speeding up of model detection. The optimized model can capture image features effectively and achieve a fast and accurate target location. Training tests are carried out on 10 types of geospatial object detection data sets (NWPU-VHR 10) as well as RSOD data sets. Comparative experiments show that, the mAP of the optimized model reaches 0.989 6, which is 2.31% higher than the model prior to the optimization. Compared with the optimization of other models using the same data sets, the mAP is 8.19% higher, verifying the fact that the optimized model can effectively improve the detection accuracy of remote sensing images.