Abstract: In order to improve the accuracy of hyperspectral inversion model for heavy metal content in cultivated soil, a genetic algorithm (GA) optimized support vector machine (SVM) was proposed to retrieve the heavy metal content of cultivated soil in a certain area of Yueyang County. After SG smoothing and 10 nm resampling, the first-order / second-order differential, reciprocal logarithm and continuum removal spectral transformation methods were used to enhance the spectral characteristics. The optimal transform spectra were selected by correlation analysis. Pearson correlation coefficient and principal component analysis were used to extract the spectral characteristic variables of heavy metals. SVM and GA-SVM were used to establish soil heavy metal hyperspectral inversion models and their accuracies were verified. The results showed that the correlation between the second-order differential transform spectra and the contents of heavy metals was the most prominent. The visible light bands of 490 nm, 500 nm, 510 nm and 530 nm were the most prominent compared with the SVM regression model. The prediction accuracy was significantly improved, and the verification set R² values of Fe, As and Cr were 0.968, 0.821 and 0.976, respectively. The research results could provide a new reference for the application of remote sensing technology to retrieve the content of heavy metals in cultivated soil.