摘要：Carrot grading is a labor-intensive and time-consuming task. In order to improve the efficiency and effect of carrot grading, algorithms were proposed to extract the key parameters of surface defects such as green-shoulder, bending, fibrous root, surface cracked and broken based on machine vision. The detection algorithm of green-shoulder is obtained by binarizing the H component. The recognition of bending carrots is realized by extracting the skeleton of the carrot on the H component and the shape of the skeleton. The detection of fibrous root is realized by the slope of carrot edges on S component. And the algorithm of surface cracked detection is gotten by binarization on G subtract B component. Broken carrots is detected by calculating the slope of carrot ends’ edges on H component. On these bases, five quantitative indicators, i.e. green shoulder ratio, bending degree, fibrous root number, surface cracked degree and broken degree, are defined. 720 carrot images selected randomly were tested. The experimental results show that the correct rate is 97.4%, 85.4%, 92.6%, 80.8% and 93.2% respectively, and the overall recognition rate is 90.9%. The algorithm proposed in this paper has positive significance for the following carrot surface defect detection and on-line classification.
