| 高光谱成像特征重构下的岩土体斜坡结构滑坡敏感区自动识别技术 |
 点此下载全文 |
| 引用本文：黄歆，蒙亮，韦耀阳.高光谱成像特征重构下的岩土体斜坡结构滑坡敏感区自动识别技术[J].计算技术与自动化,2025,(1):35-40 |
| 摘要点击次数: 86 |
| 全文下载次数: 69 |
|
|
| 中文摘要:研究了高光谱成像特征重构下的岩土体斜坡结构滑坡敏感区自动识别技术，以改善滑坡识别效果，为滑坡危害预防提供数据支持。提取岩土体斜坡结构高光谱图像像元的LBP特征并计算其特征值后，确定像元的空间邻域块，依据标签信息剔除高光谱图像背景，再根据光谱距离计算各空间邻接块像素点权值，获得中心像素点空间重构特征值，将其与获取的LBP特征值作融合处理，实现高光谱图像特征重构，将其与反映滑坡敏感性特征的影响指标一起输入构建的基于DETR的滑坡敏感区识别模型中，通过改进后的ResNet-26网络提取岩土体斜坡结构滑坡特征图，由位置编码单元对已切割的各正方形图像块作位置编码，通过Transformer编解码器对其作训练后，由前馈神经网络输出滑坡敏感区的定位识别结果。实验结果表明：该技术可实现岩土体斜坡结构滑坡敏感区识别，ω取值为6时，滑坡敏感区识别效果最优。 |
| 中文关键词:高光谱成像 特征重构 滑坡敏感区识别 LBP特征 空间邻域块 Transformer编码器 |
| |
| Automatic Identification Technology for Landslide Sensitive Areas of Rock Slope Structures under Hyperspectral Imaging Feature Reconstruction |
|
|
| Abstract:To study the automatic identification technology of landslide sensitive areas in rock slope structures under hyperspectral imaging feature reconstruction, in order to improve the effectiveness of landslide identification and provide data support for landslide hazard prevention. After extracting the LBP features of rock slope structure hyperspectral image pixels and calculating their eigenvalues, the spatial neighborhood blocks of the pixels are determined. The background of the hyperspectral image is removed based on label information, and the pixel weights of each spatial adjacent block are calculated based on spectral distance to obtain the spatial reconstruction feature values of the central pixel points. They are fused with the obtained LBP feature values to achieve hyperspectral image feature reconstruction, It is input into the landslide sensitive area recognition model based on DETR together with the impact indicators reflecting the landslide sensitivity characteristics. The landslide feature map of rock slope structure is extracted through the improved ResNet-26 network. The position coding unit encodes the positions of the cut square image blocks. After training them through the Transformer codec, the Feedforward neural network outputs the positioning recognition results of the landslide sensitive area. The experimental results show that this technology can achieve the identification of landslide sensitive areas in rock slope structures. When the value of ω is 6, the identification effect of landslide sensitive areas is the best. |
| keywords:hyperspectral imaging feature reconstruction identification of landslide sensitive areas LBP features spatial neighborhood blocks transformer encoder |
| 查看全文 查看/发表评论 下载pdf阅读器 |
次