Hyperspectral Estimation Of Soil Organic Matter Content Based On CNN-FCM

Soil organic matter provides nutrients for the growth of crops and plays an important role in the global carbon cycle and agricultural activities.Hyperspectral remote sensing technology provides technical support for real-time and efficient monitoring of soil organic matter content due to its multiple bands,high resolution,large amount of data,and rich information.Convolutional neural networks（CNN）have been used in soil organic matter estimation,but there is a problem that the model fitting accuracy and prediction accuracy are not synchronized.Therefore,this paper takes Zhangqiu District and Jiyang District of Jinan City,Shandong Province as the research area,based on the collected organic matter content and outdoor spectral data of 121 soil samples,and establishes a convolutional neural network and fuzzy identification based on soil organic matter content high.Spectral estimation model to improve estimation accuracy.The main research contents and conclusions are as follows:（1）The principal components of soil spectral data in Zhangqiu and Jiyang regions were determinedIn order to make full use of soil spectral information and data dimensionality reduction,principal component analysis（PCA）method was used,and SPSS software was used to perform PCA operation to select the principal components of the spectrum.The results show that the cumulative contribution rate of the 36 principal components reaches 99.99%,almost including all the original feature information.（2）A hyperspectral estimation model of soil organic matter content based on convolutional neural network was establishedBased on convolutional neural network theory,a hyperspectral estimation model of soil organic matter content based on convolutional neural network was established.Firstly,the principal components of the soil spectrum are converted into a four-dimensional spectrum information array,and then various parameters and network structure are adjusted through experimental simulation,and finally the optimal estimation model is obtained.The results show that when the model adopts a 3×3 convolution kernel,an average pooling layer,a fully connected layer,and the network iterates 600 times,the convolutional neural network model achieves the optimal prediction effect,of which 19 The coefficient of determination R~2of the test sample estimation results was 0.841,and the average relative error was 7.123%.（3）A hyperspectral estimation model of soil organic matter content based on CNN-FCM was establishedIn order to overcome the deficiencies of convolutional neural network（CNN）in prediction,a hyperspectral estimation model of soil organic matter content based on CNN-FCM was established based on convolutional neural network theory and fuzzy recognition theory.First,a four-dimensional spectral information matrix is constructed by using the principal components of the soil spectrum,and then the feature information of the spectral data is extracted by the CNN convolution layer,and the feature information output by the CNN convolution layer is used as the feature vector input by the CNN and FCM,respectively.The optimal estimation model is obtained by optimizing the model parameters and structure.Finally,the estimation results of CNN and FCM are calculated respectively,and a fusion model is established according to the internal relationship between the two results.The results show that when the CNN model structure is a3?3 convolution kernel,an average pooling layer with a scale of2?2,a fully connected and output layer,the number of fuzzy classifications in the FCM model is 10,and a linear function is used to establish In the internal relationship model of the two results,the model has the highest estimation accuracy,with R~2of 0.896 and MRE of 5.464%,which are all better than CNN,FCM,MLR,SVM and BP neural network models.The research shows that the CNN-FCM estimation model avoids the complex preprocessing operations in the traditional modeling method.While ensuring the high-precision fitting and operation rate of the model,it also ensures the high-precision estimation of the model,and effectively overcomes abnormality.The influence of sample points significantly improved the estimation accuracy compared with the traditional model,and provided a new idea for the estimation of soil organic matter content.