The rapid and precise detection of heavy metals in soil is the key to the efficacious prevention and remediation of soil contamination. Employing a portable X-ray fluorescence spectrometer facilitates the in-situ, non-destructive, and rapid detection of typical heavy metals. This advanced analytical technique also obviates the need for elaborate sample digestion procedures. However, the accuracy of the XRF-based heavy metal detection technique is significantly influenced by the soil matrix effects, which considerably limits the accuracy of such measurements. Calibration against standard soil with a similar matrix is imperative. As a result, this study combined pattern recognition and the standard curve method to achieve a precise analysis of typical heavy metals in various soils. The dataset comprises the X-ray fluorescence spectra and heavy metal contents across six characteristic soil types collected within China: humid-thermo ferritic, paddy soils, black soils, flavor-aquic soils, yellow-brown earth, and yellow-red earth. The spectral data is refined using a five-point, three-times window movement smoothing algorithm and a min-max normalization approach, followed by principal component analysis (PCA). Post-PCA dimensionality reductions first five principal components are employed as input feature variables, with soil types serving as labels. A predictive model based on a Radial Basis Function (RBF) kernel for Support Vector Machine (SVM) is constructed to categorize soils by matrix similarity. The models hyperparameters are optimized using the Horned lizard optimizer algorithm, yielding an optimized kernel function (g) of 0.038 1 and a penalty factor (c) of 7.852 9, with a correct classification rate of 100% under a five-fold cross-validation. The quantitative analysis utilizes the standard curve method. For the six soil types, the correlation coefficients for Chromium (Cr) ranged from 0.994 7 to 0.999 3, for Nickel (Ni) from 0.986 8 to 0.999 4, for Copper (Cu) from 0.992 9 to 0.999 9, for Zinc (Zn) from 0.984 1 to 0.999 8, and for Lead (Pb) from 0.987 7 to 0.999 6. Furthermore, the correlation coefficients of Arsenic and Lead (As & Pb) ranged from 0.961 3 to 0.999 5. The above results indicate a favorable linearity for heavy metals within the same matrix. Subsequently, the established RBF-SVM model and standard curves are applied to a prediction set of 24 samples. The predictive outcome corroborates a 100% classification accuracy for the six soil types. Upon classification, corresponding standard curves are utilized for quantitative analysis. The results show that the average relative prediction errors for Cr, Ni, Cu, Zn, Pb, and As are 2.24%, 3.66%, 2.72%, 2.15%, 2.13%, and 5.55%, respectively, below 6%. These findings prove the excellent applicability of the RBF-SVM model in combination with the standard curve method for the rapid detection of typical heavy metals in soil. This algorithm will facilitate the rapid quantitative detection of typical heavy metals in natural soil.