| The rapid and accurate quantitative monitoring of soil nutrients in black soil is of great significance for the effective utilization and protection of black soil resources.Soil organic matter(SOM),total nitrogen(TN),total phosphorus(TP)and total kalium(TK)are the main nutrients in soil,which are important indexes to measure soil nutrients.The traditional laboratory analysis method has the problems of time consuming,energy consuming and pollution,which cannot meet the demand of soil dynamic monitoring.With its extremely high spatial resolution and rich spectral information,hyperspectral remote sensing technology can achieve repeated observations in a short time,which has an absolute advantage in the study of soil component content.A typical black soil area of Heilongjiang Province was selected as the study area.,this paper collected soil samples,to obtain the SOM,TN,TP and TK content and ground spectral information.According to the sampling interval of CASI/SASI aerial hyperspectral images,the ground spectrum was re-sampled(RS),and on this basis,7forms of spectral transformation such as spectral differentiation,scattering correction,and continuum removal were performed.By calculating the correlation between all spectral forms and nutrient content,the characteristic bands that reach extremely significant correlation levels are extracted as the independent variables of the model,and four nutrient content prediction models were constructed by partial least squares regression(PLSR)and BP neural network(BPNN).Substitute the CASI/SASI image into the optimal model for mapping,and use the histogram matching method to modify the mapping results,and finally obtain the spatial distribution of soil nutrients.The study found that a variety of spectral variable models based on two methods can successfully predict the quantitative content of SOM and TN.Among them,the SD-BPNN model has the best prediction effect on the two nutrient contents.The model verification set R~2 is 0.872 and 0.867,and the RPD is 2.618 and 2.5.TP and TK are relatively low in content,and the characteristic band is not significant.There is a gap between the prediction effect of the model and the SOM and TN models.The RPD of the optimal prediction model is between 1.4~2,which can achieve a rough estimate of the two nutrient contents.The difference between the reflection spectrum of the image spectrum and the ground spectrum will cause the image mapping accuracy to decrease.After the histogram matching correction can effectively improve the remote sensing prediction accuracy,the corrected hyperspectral remote sensing mapping can realize the quantitative prediction of SOM and TN,as well as the rough estimation of TK content,but cannot effectively predict the TP content. |
