CN-121978026-A - Soil organic carbon content deep learning inversion method based on intermediate variables

CN121978026ACN 121978026 ACN121978026 ACN 121978026ACN-121978026-A

Abstract

The application is suitable for the technical field of soil detection, and provides a soil organic carbon content deep learning inversion method based on intermediate variables. Based on the enhanced spectrum data, the pH value and the nitrogen content in the soil sample are predicted through an auxiliary inversion model, and the pH value predicted value and the nitrogen content predicted value are obtained. And predicting the organic carbon content in the soil sample by a joint inversion model based on the enhanced spectrum data, the pH value predicted value and the nitrogen content predicted value to obtain the organic carbon content predicted value. According to the method, soil parameters including pH value and nitrogen content which can be predicted with high precision are taken as intermediate variables to be introduced into the problem of predicting the organic carbon content, the intermediate variables are predicted by establishing an auxiliary inversion model, and a joint inversion model is established by the intermediate variables, so that the inversion precision of the organic carbon content of the soil is improved.

Inventors

Lai Yuyun
XIONG ZHENG
LIU ZIYAO
HUANG MENG

Assignees

中安数通科技(深圳)有限公司
深圳市现代农业装备研究院

Dates

Publication Date: 20260505
Application Date: 20251209

Claims (10)

1. The soil organic carbon content deep learning inversion method based on the intermediate variable is characterized by comprising the following steps of: acquiring enhanced spectrum data of a soil sample; Based on the enhanced spectrum data, predicting the pH value and the nitrogen content in the soil sample through an auxiliary inversion model to obtain a pH value predicted value and a nitrogen content predicted value, wherein the auxiliary inversion model is a model which is obtained by training in advance based on a convolutional neural network and is used for predicting the pH value and the nitrogen content in the soil sample; And predicting the organic carbon content in the soil sample through a joint inversion model based on the enhanced spectrum data, the pH value predicted value and the nitrogen content predicted value to obtain an organic carbon content predicted value, wherein the joint inversion model is a model which is trained in advance based on the convolutional neural network and is used for predicting the organic carbon content in the soil sample.
2. The intermediate variable-based soil organic carbon content deep learning inversion method of claim 1, wherein the obtaining enhanced spectral data of the soil sample comprises: collecting original spectrum data of the soil sample through a spectrometer; carrying out fractional order differential processing on the original spectrum data to obtain processed spectrum data; and carrying out standardization processing on the processed spectrum data to obtain the enhanced spectrum data.
3. The intermediate variable-based soil organic carbon content deep learning inversion method of claim 2, wherein the performing fractional order differential processing on the raw spectrum data to obtain processed spectrum data comprises: Performing inverse transformation processing on the hyperspectral reflectance curve function corresponding to the original spectrum data to obtain a processed hyperspectral reflectance curve function; The processed hyperspectral reflectivity curve function is carried out in a preset interval And performing step-difference transformation processing to obtain the processed spectrum data.
4. The intermediate variable-based soil organic carbon content deep learning inversion method of claim 1, wherein the predicting the organic carbon content in the soil sample by a joint inversion model based on the enhanced spectral data, the ph predicted value and the nitrogen content predicted value to obtain an organic carbon content predicted value comprises: Respectively carrying out standardized treatment on the enhanced spectrum data, the pH value predicted value and the nitrogen content predicted value to obtain the enhanced spectrum data after treatment, the pH value predicted value after treatment and the nitrogen content predicted value after treatment; Carrying out fusion treatment on the treated enhanced spectrum data, the treated pH value predicted value and the treated nitrogen content predicted value to obtain a fusion feature vector; And predicting the content of organic carbon in the soil sample through the joint inversion model based on the fusion feature vector to obtain the predicted value of the content of the organic carbon.
5. The intermediate variable-based soil organic carbon content deep learning inversion method according to claim 4, wherein, Taking the enhanced spectrum data as an independent variable X, the organic carbon content as a dependent variable Y, and the pH value predicted value and the nitrogen content predicted value as intermediate variables M; determining the intermediate variable M from the enhanced spectral data and the auxiliary inversion model by the following formula: ; determining a dependent variable Y from the intermediate variable M, the independent variable X, and the joint inversion model by the following formula: ; Wherein, the Representing the said auxiliary inversion model, Representing the joint-inversion model in question, Representing random errors.
6. The intermediate variable based soil organic carbon content deep learning inversion method of claim 5, further comprising: estimating the performance of the joint inversion model by adopting a fitting goodness, a root mean square error and an average absolute error; and under the condition that the performance of the joint inversion model does not meet the preset error condition, performing performance optimization on the joint inversion model by adjusting the learning rate in an optimizer until the performance of the joint inversion model meets the preset error condition and has no rising space.
7. Soil organic carbon content deep learning inversion device based on intermediate variable, characterized by comprising: The data acquisition module is used for acquiring the enhanced spectrum data of the soil sample; The first prediction module is used for predicting the pH value and the nitrogen content in the soil sample through an auxiliary inversion model based on the enhanced spectrum data to obtain a pH value predicted value and a nitrogen content predicted value, wherein the auxiliary inversion model is a prediction model which is trained in advance to obtain the pH value and the nitrogen content in the soil sample; The second prediction module is used for predicting the organic carbon content in the soil sample through a joint inversion model based on the enhanced spectrum data, the pH value predicted value and the nitrogen content predicted value to obtain an organic carbon content predicted value, wherein the joint inversion model is a predictive model which is trained in advance to obtain the organic carbon content in the soil sample.
8. A terminal device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the method according to any one of claims 1 to 6 when executing the computer program.
9. A computer readable storage medium, characterized in that the computer readable storage medium stores a computer program which, when executed by a processor, implements the method according to any of claims 1 to 6.
10. A computer program product comprising a computer program which, when run, causes the method of any one of claims 1 to 6 to be performed.

Description

Soil organic carbon content deep learning inversion method based on intermediate variables Technical Field The application belongs to the technical field of soil detection, and particularly relates to a soil organic carbon content deep learning inversion method based on intermediate variables. Background The soil is indistinguishable from human life and natural ecology, so accurate prediction of soil parameters is crucial for soil treatment. The organic carbon content of the soil is an important influencing factor of the soil fertility level, and determines whether the soil can provide sufficient nutrients for crops, so that reliable estimation of the organic carbon can effectively guide aspects such as fertilization decision of farmers and the like. Although the traditional laboratory organic carbon detection method has accurate results, the problems of low detection precision, high cost and the like exist, and the treatment requirement of the agricultural big data era can not be met. Disclosure of Invention The embodiment of the application provides a soil organic carbon content deep learning inversion method based on intermediate variables, which can solve the problem of low prediction precision of the traditional soil organic carbon content prediction method. In a first aspect, an embodiment of the present application provides a soil organic carbon content deep learning inversion method based on intermediate variables, where the method includes: acquiring enhanced spectrum data of a soil sample; Based on the enhanced spectrum data, predicting the pH value and the nitrogen content in the soil sample through an auxiliary inversion model to obtain a pH value predicted value and a nitrogen content predicted value, wherein the auxiliary inversion model is a model which is obtained by training in advance based on a convolutional neural network and is used for predicting the pH value and the nitrogen content in the soil sample; And predicting the organic carbon content in the soil sample through a joint inversion model based on the enhanced spectrum data, the pH value predicted value and the nitrogen content predicted value to obtain an organic carbon content predicted value, wherein the joint inversion model is a model which is trained in advance based on the convolutional neural network and is used for predicting the organic carbon content in the soil sample. In a possible implementation manner of the first aspect, the acquiring the enhanced spectrum data of the soil sample includes: collecting original spectrum data of the soil sample through a spectrometer; carrying out fractional order differential processing on the original spectrum data to obtain processed spectrum data; and carrying out standardization processing on the processed spectrum data to obtain the enhanced spectrum data. In a possible implementation manner of the first aspect, the performing fractional order differential processing on the raw spectrum data to obtain processed spectrum data includes: Performing inverse transformation processing on the hyperspectral reflectance curve function corresponding to the original spectrum data to obtain a processed hyperspectral reflectance curve function; The processed hyperspectral reflectivity curve function is carried out in a preset interval And performing step-difference transformation processing to obtain the processed spectrum data. In a possible implementation manner of the first aspect, the predicting, by a joint inversion model, the organic carbon content in the soil sample based on the enhanced spectrum data, the ph predicted value and the nitrogen content predicted value, to obtain an organic carbon content predicted value includes: Respectively carrying out standardized treatment on the enhanced spectrum data, the pH value predicted value and the nitrogen content predicted value to obtain the enhanced spectrum data after treatment, the pH value predicted value after treatment and the nitrogen content predicted value after treatment; Carrying out fusion treatment on the treated enhanced spectrum data, the treated pH value predicted value and the treated nitrogen content predicted value to obtain a fusion feature vector; And predicting the content of organic carbon in the soil sample through the joint inversion model based on the fusion feature vector to obtain the predicted value of the content of the organic carbon. In one possible implementation manner of the first aspect, the enhancement spectrum data is taken as an independent variable X, the organic carbon content is taken as a dependent variable Y, and the predicted ph value and the predicted nitrogen content are taken as intermediate variables M; determining the intermediate variable M from the enhanced spectral data and the auxiliary inversion model by the following formula: ; determining a dependent variable Y from the intermediate variable M, the independent variable X, and the joint inversion model by the following formula: ; Wh