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

[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.