| 摘要: |
| 【目的】 探讨利用深度学习技术提高高光谱图像分类准确性的方法，解决高光谱数据处理的信息冗余和融合过程的信息损失等问题，为土地利用管理等提供强有力技术支撑。【方法】 文章以象山港为研究区域，基于珠海一号高分辨率遥感影像，通过PCA法降低影像数据的维度，提取最具代表性的特征，同时去除噪声和冗余信息；在数据降尺度的过程中采用ILGIF技术实现最小化信息损失来优化图像融合，保留关键空间和光谱信息，以提供更高质量的数据支持。在此基础上，采用深度学习U-Net模型，对土地利用类型进行精确分类。【结果】 采用深度学习中的语义分割方法能有效提高精度水平，U-Net模型的整体分类精度为90%，研究区内水体、林地、建成区的分类精度良好，依次为96.93%、88.34%和71.69%。【结论】 后续将进一步挖掘多模态（光学与微波）数据资源的潜力，融合多尺度的空间与光谱分辨率数据集的优势，探索并推动人工智能、大数据等技术集成在遥感图像分类领域的创新应用，实现对地表覆盖快速且精确地监测，为相关部门提供数据支撑。 |
| 关键词: 深度学习 土地利用分类 珠海一号 U-Net 卷积神经网络（CNN） |
| DOI：10.12105/j.issn.1672-0423.20240502 |
| 分类号: |
| 基金项目:上海市地质调查研究院科研创新项目“基于国产卫星遥感数据的耕地非粮化信息提取研究”（2024（D）-035（科））；上海市规划和自然资源局科研项目“耕地保护督察和用途管制专项监督检查数据分析”（310000000240207161167-00069437） |
|
| Research on land use classification based on deep learning and data enhancement |
|
Luan Shuli1,2, Tang Hang1, Liu Xi1, Zhang Nuoxin1,2, Zhang Haiqin1,2
|
|
1.Shanghai Institute of Natural Resources Survey and Utilization Research，Shanghai 200072，China;2.Shanghai Institute of Geological Survey Technology，Shanghai 200436，China
|
| Abstract: |
| [Purpose] This paper aimes to explore the use of deep learning technology to improve the accuracy of hyperspectral image classification，to solve the challenges of high dimensionality and information loss in hyperspectral data processing，and to provide strong technical support for land use management.[Method] Taking Xiangshan Port as the research area，based on the high-resolution remote sensing image of Zhuhai-1，the dimension of the image data was reduced by the PCA method，and the most representative features were extracted，while the noise and redundant information were removed. In the process of data downscaling，the ILGIF technology was used to optimize image fusion by minimizing information loss，retain key spatial and spectral information，and provide higher quality data support. On this basis，the U-Net model in deep learning was used to accurately classify land use types.[Result] The semantic segmentation method in deep learning could effectively improve the accuracy level，and the overall classification accuracy of the U-Net model was 90%，and the classification accuracy of water，woodland and built-up area in the study area was good，which was 96.93%，88.34% and 71.69%，respectively.[Conclusion] In the future，we will further explore the potential of multimodal（optical and microwave）data resources，integrate the advantages of multi-scale spatial and spectral resolution datasets，explore and promote the innovative application of artificial intelligence，big data and other technologies in the field of remote sensing image classification，to achieve a rapid and accurate monitoring of land surface coverage，and provide data support for the relevant departments. |
| Key words: deep learning models land use classification Zhuhai-1 U-Net CNN |
