Deep learning and cluster analysis based soil organic matter inversion model

In order to achieve scientific management of soil environment and effective ecological protection, it is important to accurately understand the quantitative characteristics of soil organic matter(SOM)and its spatial distribution pattern.To address the problem of low accuracy of soil organic matter content prediction by traditional prediction models, this paper combines spatial clustering into a deep learning method to build a remote sensing-based organic matter content prediction model.The results show that the inversion accuracy R2 (coefficient of determination) and RMSE (root mean square error) of the clustered deep neural network model reach 0.82 and 2.25, respectively, which are 36.67% and 52.23% higher than the accuracy of the deep neural network model, and the model overfitting problem will be alleviated and the computational efficiency will be improved after the introduction of the spatial clustering function. It can provide effective technical support for inversion of SOM using multispectral remote sensing data and fine monitoring and management of soil quality at regional scale.