Remote sensing image object detection has a wide range of applications in intelligent transport, such as dynamic monitoring of road network operation status, intelligent law enforcement on roads, and intelligent monitoring of road disasters. Due to the characteristics of small and dense targets, large scale changes, and arbitrary direction distribution in remote sensing images, general object detectors have poor detection performance when directly applied to remote sensing images. To address these challenges,this paper proposes a remote sensing image object detection algorithm based on improved Retinanet. First, the model introduces Improved Downsampling Module (IDM) on the base feature extraction network ResNet50, which performs multiple down-sampling processing on features, and then dynamically selects the spatial receptive field using the convolution kernel selection mechanism to model the multi-scale semantic information of the scene. Finally, the classification and regression results of the target object are obtained. Experimental results show that the proposed method improves the mAP by 3.2% on the large-scale remote sensing image object detection dataset DOTA compared to the original Retinanet network, enabling more accurate localization and identification of remote sensing targets.