Study On Prediction Of Soil Organic Matter Based On Digital Image Color Extraction
Wu Cai-Wu,Yang Hao,Xia Jian-Xin,Chang Jia-Ning,Yang Yue,Zhang Yue-Cong,Cheng Fu-Wei
DOI: https://doi.org/10.3964/j.issn.1000-0593(2019)04-1222-07
2019-01-01
Spectroscopy and spectral analysis
Abstract:As an important criterion for determining soil quality, the rapid determination of soil organic matter (SOM) can provide basic data support for the implementation of precision agriculture. Traditional determination method of SOM, through field sampling and laboratory chemical analysis, not only time-consuming and laborious, but also inefficient, cannot meet the largescale demand for soil information in current social development. Although the prediction model of SOM can be established based upon the characteristics of spectral reflectance of soil affected by SOM to realize the rapid prediction for SOM, the spectrometer is of high price and strict operation environment, which limits its wide application. Then visible-light sensor with RGB is cheap and easy to operate. Therefore it is worth exploring and studying the rapid determination of SOM from the perspective of practicality and economy, with the help of many advantages of visible-light sensor. Therefore, in order to verify the feasibility and applicability of extracting color information of digital images for fast prediction of SOM, the paper uses a digital camera to obtain the soil surface color information, analyses the characteristics of soil surface composition, determines the optimal sampling area, compares the correlation between different sample preparation standards (<1 mm and <0. 5 mm) and SOM, selects the high correlation of color variables, and establishes the prediction model of SOM through regression analysis. The results show that the 950X950 pixel as the sampling area can obtain the color of the soil surface more stably and reduce the influence of the edge effect on the sampling result. In the correlation analysis between the soil samples <1 mm and <0. 5 mm and SOM, the RGB bands of <1 mm soil samples have a higher correlation with SOM and are suitable as sample preparation standards for soil color acquisition. In the three bands of RGB, the red band showed the highest correlation with SOM, with a correlation coefficient of -0. 70. The correlation between color and SOM was enhanced by the mathematical transformation of the RGB band and the excessred (ExR) calculation, in which the ExR index shows the highest correlation with SOM with a correlation coefficient of -0. 86. In a single variable modeling process, the best predictive effect is obtained by ExR reciprocal model. In the multivariable modeling, the standard deviations of each color were involved in the modeling, which causes the color information description to be more comprehensive, and the best modeling results are obtained that can better reflect the variation of SOM within the study area, its R-2 = 0. 80, RMSE = 0. 51, the validation result R-val(2) = 0. 84, RMSEval =0. 54. Based on the prediction results of the model for black soil, only the single-variable red band model shows a good prediction effect, and the test results show that the red band is a sensitive band of SOM and has its universality in different soil types. Although the model built in this study cannot be extended to predict other soil types, the prediction of the same soil shows that the digital camera, as a quantitative color imetric tool, has the potential to rapidly predict SOM content.