The incidence of black-heart disease in post-harvest storage and transportation of potato in China is high, the internal quality is also uneven, and the detection and sorting technology lags behind, which seriously restricts the development of potatoes’ staple food industry. Simultaneous online non-destructive testing of internal quality such as potatoes’ black-heart disease and starch content is of great significance for promoting the strategy of main potato diet in China. Based on the principle of visible/near-infrared diffuse transmission spectroscopy, this study uses a non-destructive on-line detection system built by the laboratory (detection speed is about 4/s), and carries out black heart disease with potato black-heart disease and starch content as internal quality test indicators. Simultaneous non-destructive testing with starch content. The original spectra of 121 healthy potatoes and 116 black-heart potatoes in the 600～1 000 nm band were averaged. The absorbance values of black potato samples in the 600～900 nm band were significantly higher than those of healthy potato samples, and the influence of black heart tissue was observed. The characteristic absorption peak of chlorophyll near 663 nm and the characteristic absorption peak of water near 760 nm of healthy potato were significantly higher than that of black heart potato. Partial Least Squares Discriminant Analysis (PLS-DA) was established based on the original spectrum of healthy potato and black heart potato. At the same time, SG-Smoothing, Standard Normal Transformation (SNV), Multiple Scattering Correction (MSC), First Derivative (FD), SG Smoothing and First Derivative (SG+FD) and other pretreatment methods were applied to the 121 healthy potato spectra. And combined with CARS algorithm to screen the characteristic wavelength, established a Starch Content (SC) Partial Least Squares (PLS) quantitative prediction model. The results showed that the correctness rate of the correction set and verification set of the PLS-DA model of the black-heart potatoes was 97.74% and 98.33%, respectively, and the total discriminant correct rate was 97.89%. The original spectrum was preprocessed by SG smoothing plus first derivative, and then combined with CARS. The PLS model of potato starch content was optimized by algorithm screening. The correlation coefficients of the calibration set and prediction set were 0.928 and 0.908, respectively, and the root means square error was 0.556% and 0.633%, respectively. Finally, the model was built into an online inspection system and externally verified using 50 samples that were not modeled. The correct rate of potato black heart disease was 96%, the correlation coefficient between the starch predicted value and the standard physical and chemical value was 0.893, and the root means square error was 0.713%. It is indicated that potato black-heart disease and other internal quality can be simultaneously detected by on-line non-destructive testing based on potatoes’ diffuse transmission spectroscopy, which provides a technical reference for potatoes’ post-harvest qualities testing and promotion of potatoes’ staple food industry.