Objective To achieve precise non-destructive detection and sorting of blueberry quality.Methods An online evaluation system for blueberry quality based on near-infrared spectroscopy and visible light image fusion is designed. The system consists of a fruit box chain conveyor module, a linear array image detection module, a near-infrared spectroscopy detection module, and a control system. A diffuse reflection optical path is designed by the near-infrared spectroscopy detection module, and a partial least squares regression model is established to predict the soluble solids content in blueberries through S-G convolution smoothing, second-order derivative connection preprocessing, and sequential projection algorithm for feature variable extraction.Results The developed system shows an average relative error of 0.093 in determining the individual diameter of blueberries and the accuracy of 91.85% in measuring fruit diameter uniformity. The predicted correlation coefficients for superior, first-class, and second-class blueberries are 0.843 4, 0.782 2, and 0.723 7, respectively, with predicted root mean square errors of 0.831 6, 0.951 0, and 1.070 5, respectively.Conclusion The overall evaluation accuracy of the decision tree quality evaluation model for blueberries, integrating both external and internal quality, reaches 89.55%.