The work aims to optimize the light source of the machine vision recognition system based on the defect features of the food inner packaging, in order to improve the recognition rate of the inner packaging defects and reduce the amount of defective packaging. Based on computer vision recognition technology, the common defect features of inner packaging paper under different light sources were detected by spots. The type, shape and angle of light source were determined in order respectively by means of Matlab three-dimensional drawing and correlation analysis for application verification. The result was that, the infrared light source was the most suitable light source type for the defect feature recognition of paper aluminum composite inner packaging paper. The strip light source was extremely significantly correlated with the inner packaging paper, with the defect recognition rate of 96.95%.The 60° high angle illumination position was significantly correlated with the inner packaging paper, with the defect recognition rate of 96.96%. For the infrared strip light source under high-angle illumination, the defect recognition rate was 99%. Compared with LED ring light source, the infrared strip light source under high-angle illumination is applied to the on-line detection of paper aluminum composite inner packaging paper and its visual recognition rate of defect features is increased by 0.51 percentage points.