Given the advantages of high measurement accuracy and low equipment cost, image target based detection methods have broad prospects in applications such as space rendezvous, docking, and maintenance. Thanks to its powerful feature extraction and expression capabilities, convolutional neural networks have achieved significantly better performance than traditional methods in image target detection applications. However, existing methods based on convolutional neural networks have problems such as inaccurate detection of small targets, insufficient ability under complex background interference, and large fluctuations in detection rates under high reflectivity, occlusion, and motion blur. Focusing on the application requirements of target detection and recognition in complex space environments, a novel end-to-end image target detection method is proposed based on YOLO v8n to address these issues. Firstly, a multi-scale information extraction and fusion method was designed. Furthermore, based on the extracted multi-scale features, an attention mechanism based on shape information was designed to accelerate the dynamic adjustment of each output layer iteration and improve the backbone network structure for feature extraction. This method can effectively detect targets in space environments such as small targets, complex backgrounds, and occlusions, achieving better detection results than existing similar methods. Experiments show that the improved model has slightly increased the number of parameters and computational complexity compared to the YOLOv8n model, with mAP @ 0.5 increasing by 2.9% and mAP @ 0.5–0.95 increasing by 3.4%, resulting in a significant improvement in detection accuracy.