Intelligent Detection of Remote Sensing Image Objects with Attention Mechanism
Yunqiu Wang · 2023
This work aims to overcome the limitations of existing remote sensing images with a wide variety of target objects, which makes it difficult to directly apply salient target detection methods for natural scene images. First, a complete salient target detection scheme for remote sensing images is proposed, which can achieve efficient and accurate detection and provide technical support for the practical application of salient target detection for remote sensing images. In order to address the challenge of diverse target object types in remote sensing images, an Attention Pyramid Decoder Network (APDNet) model is proposed. This model embeds a global attention mechanism to capture feature correlations among different target object types, effectively alleviating the interference of target object type variations on detection performance. Next, the model adopts a pyramid decoder structure to integrate multi-level encoding information at different scales, achieving multi-level feature information aggregation and better feature decoding. Experimental results show that the APDNet model can effectively detect salient objects with diverse types, verifying the stability of the proposed method. The relevant research on intelligent detection of remote sensing image targets can promote the development of computer vision technology in the field of remote sensing and further promote the application of remote sensing data in various scenes. This work has crucial research significance and practical value.