Soil Organic Matter Hyperspectral Estimating Model Based On Fuzzy Recognition

The land is an important basic resource which is irreplaceable and non-renewable, andhuman’s survival and development must rely on it. Its limited number and scarce supply makethat we should make full use of land resources. The soil information’s accurate access is thepremise of precise management and precise fertilizer in modern agriculture, and is theimportant basic data which can ensure the agriculture production. The traditional nutrientmeasurement methods have many shortcomings, such as slow, low accuracy, and so on.Hyper-spectral technology has many bands, high spectral resolution and can obtaincontinuous spectral images; therefore, it can provide a favorable means for the monitoring ofthe soil organic matter content. This paper based on the soil samples collected in Shanxiprovince, Hengshan county and National Natural Science Foundation of China “Theapplication of imaging spectroscopy in the land’s dynamic monitoring and evaluation”(projectnumber:40271007) to get the hyper spectral data of the soil reflectance and organic mattercontent, and then we use linear regression, neural network and fuzzy recognition methods toestablish the soil organic matter hyper spectral estimating model and discuss these models.In the first place, we analyze the hyper spectral characteristics of different types andorganic matter contents’ soil, and also analyze the soil organic matter and water content’sinteraction on the soil spectral reflectance. We draw the following conclusions: the soil ofdifferent types has different soil reflectance; for the soil of same type, the soil reflectance wasnegatively correlated with the organic matter content; the existence of the soil organic matterand water content’s interaction on the soil spectral reflectance.Secondly, we carry out the spectral data transformation and draw the correlationcoefficient figure; at last we found that the correlation coefficient is the best in some bandafter the number of first-order differential transformation and choose this transformation asthis paper’s method; based on single correlation analysis to choose393nm，443nm，502nm，1453nm，1936nm and2186nm as inversion factors; based on the scatter plot andR2’s valueto exclude the abnormal samples.Finally, we establish the fuzzy nearness inversion model of the soil organic mattercontent in this paper, and compared with the results of linear regression and neural network model we can see: the linear regression and BP neural network models accuracy are almostthe same accuracy and very low；the fuzzy nearness inversion model’ accuracy are high, oneof these is nearly90%, the rest are more than90%; in addition, according to the diversities ofthe spectral inversion impact factor and the dynamics of the soil organic matter content wepropose the interval-value fuzzy inversion model, the forecast result is that6samples’prediction interval is accurate,1sample’s prediction interval exists error, and the errorsample’s bias is3.22%. This shows that the fuzzy recognition model solves nonlinear andfuzzy effectively, and using the fuzzy recognition model to predict the soil organic mattercontent is valid.