Intelligent recognition of civil aviation aircrafts based on improved YOLOv8 model using multisource satellite images

This paper presents an advanced method for the intelligent recognition of civil aviation aircraft based on an enhanced YOLOv8 model utilizing multi-source satellite imagery. The method optimized the Neckstructure of RepGFPN by using upsampling connections based on dynamic upsampling technology and enhanced multi-scale feature fusion technology to improve the accuracy of objects detection under the complex background conditions, as well as the multi-scale, multi rotation angles in remote sensing images. To improve the robustness of the model, this paper employs data augmentation techniques such as image resampling, flipping, small angle rotation, noise addition, cropping and filling to enhance the model accuracy. Experimental results show that the improved YOLOv8 model significantly improves the positioning and classification accuracy of aircraft targets, achieving an F1-score increase of 3% for optical images and 2% for SAR images. This method provides strong support for intelligent monitoring and management of civil aviation aircraft.