水土保持措施识别与提取方法的研究进展

Research Progress on Identification and Extraction Methods of Soil and Water Conservation Measures

[目的]水土保持措施类型及其配置模式复杂繁多,准确识别与精细化提取水土保持措施详细配置信息是获取水土保持措施因子值的基础。[方法]水土保持措施信息获取方式主要有传统的野外调查、卫星遥感影像和无人机近景摄影等,其识别与提取方法主要包括目视解译、传统的机器学习、面向对象分类方法及深度学习模型。通过梳理国内外水土保持措施识别与提取方法的研究成果,总结存在的不足并提出研究展望。[结果]在语义分割中未来的特征融合与多模态学习、弱监督与半监督学习、集成学习和元学习等均可被运用到水土保持措施提取中。[结论]当前对于水土保持耕作措施识别与提取的成果鲜见报道,而农业实践中耕作措施较常见,后续应加强耕作措施识别提取的研究;人工智能结合大数据技术是未来水土保持措施信息高效精准识别与提取的发展方向,需要进一步研究采用半监督、弱监督学习方法,结合多模态学习、小样本标签等方法,获取高质量的标记样本数据,进行水土保持点、线状工程措施的提取;将多模态学习、实例分割方法等深度学习算法与面向对象分类方法相结合应用到水土保持植物措施的识别提取中,提高不同水土保持植物措施的分类提取精度,从而完善各类水土保持措施的信息提取方法,为准确获取水土保持措施因子值及核算水土保持碳汇能力提供支撑。

[Objective]The types of soil and water conservation measures and their configuration modes are complicated.Accurate identification and fine extraction of detailed configuration information of soil and water conservation measures are the basis for obtaining the factor values of soil and water conservation measures.[Methods]The information acquisition methods of soil and water conservation measures mainly include traditional field surveys,satellite remote sensing images,and UAV close-range photography.The identification and extraction methods mainly include visual interpretation,traditional machine learning,object-oriented classification methods,and deep learning models.By combing the research results of identification and extraction methods of soil and water conservation measures at home and abroad,the existing shortcomings are summarized and the research prospects are put forward.[Results]In semantic segmentation,future feature fusion and multimodal learning,weak supervision and semi-supervised learning,integrated learning and meta-learning can be applied to the extraction of soil and water conservation measures.[Conclusion]At present,there are few reports on the results of identification and extraction of soil and water conservation tillage measures.However,tillage measures are common in agricultural practice,and the research on identification and extraction of tillage measures should be strengthened in the future.Artificial intelligence combined with big data technology is the development direction of efficient and accurate identification and extraction of soil and water conservation measures in the future.It is necessary to further study the use of semi-supervised and weakly supervised learning methods,combined with multi-modal learning,small sample labels and other methods to obtain high-quality labeled sample data for soil and water conservation.Extraction of point and linear engineering measures;the combination of deep learning algorithms such as multimodal learning and instance segmentation methods with object-oriented classification methods is applied to the identification and extraction of soil and water conservation plant measures to improve the classification and extraction accuracy of different soil and water conservation plant measures.So as to improve the information extraction method of various soil and water conservation measures,and provide support for accurately obtaining the factor value of soil and water conservation measures and calculating the carbon sink capacity of soil and water conservation.