Abstract
To improve the matching accuracy of cross view object recognition between unmanned aerial vehicle aerial images and satellite images, an object recognition matching algorithm based on attention mechanism and feature fusion of convolutional neural network and Transformer (AFF-CNN-HTransformer) is proposed. Firstly, to address the issue of large-scale differences and unclear image features between drone aerial images and satellite images, a preprocessing algorithm is proposed that uses satellite images as a reference and registers the direction and scale of aerial images based on the attitude angle information of the aerial camera. Secondly, in response to the low accuracy and efficiency of existing image matching methods in matching drone aerial images with satellite images, a two-step matching process is designed. The target matching area of the satellite image is reduced through a rough matching process, and fine matching is performed through an AFF-CNN-HTransformer feature fusion recognition process. The findings from the experiments indicate that, when juxtaposed against currently available image object recognition and matching techniques, the algorithm presented in this study exhibits superior accuracy in the matching process, while simultaneously reducing Match recognition time.