The invention discloses a leaf cadmium content visible light-near infrared spectrum (Vis-NIR) spectral feature modeling method based on wavelet transform-partial least squares support vector regression (WT-LSSVR). Effective spectral sensitive band data extraction is performed on visible light-near infrared spectrums (Vis-NIR) of tomato leaves under stress of different concentrations of heavy metalcadmium, an atomic absorption spectrometry is utilized to measure heavy metal cadmium contents in leaves to obtain a leaf heavy metal cadmium content label set, and extracted sensitive characteristicwavelength and a WT-LSSVR feature modeling algorithm are utilized to finish leaf cadmium content Vis-NIR spectral feature modeling. The modeling method has the advantages of quick detection speed, high cadmium content predicting accuracy, high cadmium content predicting efficiency, no damage to leaves, ability in quantitatively detecting different cadmium residue contents of the same variety of tomatoes, ability in predicting cadmium residue contents of unknown leaves and the like.