Abstract: The detection of impurities in rapeseed combine harvester mainly depends on labor, which has low efficiency and poor real-time performance, resulting in the lack of basis for the regulation of harvester operation parameters and unstable harvest quality. In order to solve the above problems, this paper proposed a visual recognition algorithm for rapeseed impurities, and developed an on-line detection system for impurities. Based on the HSV color space model, the brightness distribution law of rapeseed image under single side strip light source, double side strip light source and central ring light source in the guided impurity content detection device was explored. The results showed that the brightness variation coefficient of the image under the central ring light source was the smallest and the brightness uniformity of the image was the best. The distribution intervals of color characteristic parameters were compared in the low-order moment of the HSV color space model. The results showed that the characteristic parameters range of rapeseed grains and impurities in the H component were the most significant. Combined with the spherical characteristics of rapeseeds and impurities, a segmentation algorithm considering color and morphological characteristics was proposed. Through the calibration test, the relationship model between the quality of rape grain and impurity and its pixel number was constructed. The number of pixels was converted into the actual quality to realize the on-line detection of rapeseed impurity content. The bench test showed that the precision rate of rape impurities was 91.6%, the recall rate was 89.5%, and the average error of impurity content detection was 14.8%. It can accurately identify the impurities in rapeseeds and calculate the impurity content in real time.