FFT-CARS-SVM

Soil organic carbon (SOC) is crucial in precision agriculture as well as carbon cycle, and it is significant to utilize hyperspectral technology to achieve low-cost, efficient and non-destructive prediction of SOC content. Aiming at the current stage of organic carbon hyperspectral inversion process, there are problems such as overfitting and large computational volume, based on 202 southern subtropical regional forest red soil sample library, using the content gradient method to analyze the effect of different spectral transformation methods on the spectral curve and soil organic carbon content, the first proposed to the Fast Fourier Transform (FFT) combined with the Competitive Adaptive Reweighting Algorithm (CARS) of the spectral bands selection A hyperspectral prediction model of SOC content in the region was established by machine learning algorithms, and independent samples were used for validation to explore the effects of different modeling methods on the accuracy and stability of the model.