|
| 基于高效局部注意机制的遥感影像固体废物分割算法研究 |
| Efficient Local Attention Mechanism Based Solid Waste Segmentation Algorithm for Remote Sensing Images |
| |
| DOI: |
| 中文关键词: 固体废物 高效局部注意机制 几何纹理特征 遥感影像 语义分割 |
| 英文关键词: Solid wastes Efficient local attention mechanism Geometric texture features Remote sensing image Semantic segmentation |
| 基金项目:国家自然科学青年基金资助项目(42401461);河北省自然科学基金资助项目(F2023402009); 河北省高等学校科学技术研究基金资助项目(QN2022034) |
|
| 摘要点击次数: 871 |
| 全文下载次数: 117 |
| 中文摘要: |
| 针对固体废物遥感检测中多类别语义分割的难题,提出基于高效局部注意机制、融合U型网络结构的深度学习分割模型(ELAU-Net),该模型利用深度卷积提取固体废物的几何纹理特征,并在多尺度特征图上施加注意力机制实现全局语义与局部细节融合。基于谷歌遥感卫星影像的试验表明,ELAU-Net对建筑垃圾、石头和废土堆的检测精确率分别为95%、90%和91%,显著优于DeepLabV3+、HR-Net及PSP-Net等对照模型。 |
| 英文摘要: |
| To address the challenges of multi class semantic segmentation in solid waste remote sensing detection, this paper proposed a deep learning segmentation model based on an efficient local attention mechanism and a fused U shaped network structure(ELAU Net). This model utilized depthwise convolution to extract geometric texture features of solid waste, and applied an attention mechanism on multi scale feature maps to achieve the fusion of global semantics and local details. Experimental results on Google remote sensing satellite imagery demonstrated that ELAU Net achieved significantly higher detection accuracy for construction waste (95%), stones (90%) and waste soil pile(91%) compared to control models such as DeepLabV3+, HR Net and PSP Net. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |