ES-YOLOv11n: Improved YOLOv11n Method for Remote Sensing Image Object Detection

Object detection in remote-sensing imagery often suffers poor detection accuracy because of the small sizes of the objects and the complex backgrounds. In response to this problem, an improved model ES-YOLOv11n based on YOLOv11n is put forth in the present work. ES-YOLOv11n introduces the Expectation-Maximization Attention mechanism, which enables adapative focusing of the model on the target region in remote-sensing imagery and hence capture more detailed features of the object. In addition, in our new model, the SIoU loss function displaces the traditional loss function to optimize the regression capacity and positioning precision of the bounding box, thus increasing the detection accuracy of the model. Experiments on the RSOD dataset reveal that our model reaches an mAP of 89.0%, which is 2.5% higher than the value achieved by the YOLOv11n model.