GLF-Net: A Target Detection Method Based on Global and Local Multiscale Feature Fusion of Remote Sensing Aircraft Images
Lei Yu, Haicheng Hu, Zhi Zhong, Haoyu Wu, Qiuyue Deng
Abstract
Due to the influence of weather and shooting height, remote sensing aircraft image has the characteristics of small target samples, fuzzy target samples, and complex background information. Because of these characteristics, most target detection methods applied to original images have the defect of low precision and recall due to insufficient feature extraction. To solve these problems, based on the target features of remote sensing aircraft images, a remote sensing aircraft detection method based on global and local multiscale feature fusion is proposed. The method applies the encoder-decoder architecture, which extracts the local features of the target through the network encoder part, and extracts the global features of the target through the decoder part. The fused features are input into the classifier for evaluation, and a group with high scores is selected as the final detection output. The experimental results show that the proposed method has higher precision, recall, and F1 score. Compared with other detection methods, it has smaller memory usage and fewer parameters.