融合双路径编码的无人机影像灾后损毁地物识别OA

With the rapid development of UAV remote sensing technology,high-resolution image has been widely utilized for post-disaster damaged terrain object recognition.However,the traditional methods often struggle to capture rich spatial textures and semantic information effectively,which leads to blurred boundaries,loss of fine details,and omission of small-scale targets.To overcome these issues,we proposed a dual-path coding fusion approach for damaged terrain object recognition in UAV images.The proposed method incorporates a dual-path coding architecture based on convolutional neural network and Transformer,which could extract both local texture features and global contextual representations,addressing the shortcomings of single network architecture.Additionally,a refined multi-scale pyramid pooling module is introduced to integrate features at different scales,enhancing the model's ability to recognize terrain objects with varying shapes and sizes.Furthermore,a dynamic semantic prompting mechanism based on category prior knowledge is proposed to improve discrimination in low-contrast and complex scenes.Experimental results demonstrate that the proposed method outperforms mainstream models in terms of overall accuracy and other key evaluation metrics,significantly improving the precision of post-disaster damaged terrain object recognition.