AN Baisong1,2, WANG Xuemei1,2, HUANG Xiaoyu1,2, KAWUQIATI Baishan1,2

(1.College of Geographic Science and Tourism, Xinjiang Normal University, Urumqi 830054, China;  

2.Xinjiang Uygur Autonomous Region Key Laboratory "Xinjiang Arid Lake Environment and Resources Laboratory", Urumqi 830054, China)

Abstract：Selecting a reasonable processing method can enhance the effective information characteristics of soil spectrum and improve the estimation accuracy of the model. Taking the Xinjiang Wei-Ku oasis soil as the research object. The spectral data is processed by continuous wavelet transform (CWT) and mathematical transformation to extract characteristic bands. Partial least squares regression (PLSR), BP neural network (BPNN), random forest regression (RFR) and support vector machine regression (SVMR) methods are used to construct the soil heavy metal cadmium content estimation model. The results show that: (1) The trend of the original spectral curve of soil is basically the same. In the range of 600~2450nm, the spectral reflectance decreased with the increase of heavy metal cadmium content, and both are negatively correlated. (2) The combination of CWT and first-order differential (R′) of the original spectrum has the best effect. Its |r| value can reach 0.586, which is a very significant negative correlation (P<0.001), indicating that the processing method combining mathematical transformation and continuous wavelet transform can effectively reflect the spectral details. (3) Comparing the results, it is found that the coefficient of determination (R2) of the CWT-R′-SVMR model is greater than 0.86, the root mean square error (RMSE) is less than 0.02 mg/kg, and the relative percent deviation (RPD) is greater than 2. In summary, this model is effective and it can be used as an optimal model to estimate the soil heavy metal cadmium content in the study area. The continuous wavelet decomposition technology combined with mathematical transformation can effectively extract the potential information in the soil, and provide a reference for the accurate estimation of soil cadmium content.

Key words：continuous wavelet transformation; partial least squares regression; BP neural network; random forest regression; support vector machine regression; heavy metal cadmium

EARTH AND ENVIRONMENT Vol.51, No.2, Tot No.352, 2023, Page 246