Aiming at the problems of large size, high cost and destructive detection of the traditional method of heavy metal cadmium detection in vegetable leaves, a method for detecting heavy metal cadmium in vegetable leaves based on visible light-near-infrared spectroscopy was proposed, and a method was built without pretreatment. The heavy metal cadmium detection instrument with fast detection speed, small size and easy to carry can be suitable for mobile onsite detection. Nutrient solution of four heavy metal cadmium stress gradients (0mg/L, 1mg/L, 3mg/L and 5mg/L) was configured, lettuce samples were cultivated under each cadmium stress, and leaf reflectance data was collected through a hyperspectral imaging system. Three characteristic bands (550nm, 680nm and 800nm) were selected by using principal component analysis (PCA), and a heavy metal cadmium detection model was built by using partial least squares regression (PLSR). The correlation coefficient RP of test set was 0.9149, and the root mean square error of test set was 0.5271mg/kg. The designed detection instrument for heavy metal cadmium in vegetable leaves included: light source part, signal processing part, display part, power supply part and control part. The size of the instrument was 50mm×70mm×60mm. Using self-made instrument for calibration experiment, selecting A/D acquisition voltage as reference, calibration data were used for modeling, model training set correlation coefficient Rc was 0.8581, training set root mean square error was 0.4975 mg/kg, test set correlation coefficient Rp was 0.8432, root mean square error of the test set was 0.5526mg/kg, and the model prediction performance was better. Finally, the detection accuracy of the portable heavy metal cadmium nondestructive testing instrument was verified. Totally 30 groups of cadmiumstressed lettuce leaves were selected for real-time detection, which were not related to the modeling. Compared with the standard physical and chemical values, the root mean square error was 0.32 mg/kg, and the absolute measurement error was -0.69~0.66 mg/kg, the average absolute error was 0.26mg/kg. The results showed that the instrument can realize real-time nondestructive detection of cadmium content in lettuce leaves.