Seedling Image Segmentation and Feature Extraction under Complicated Background

Vision system is applied to automated transplanter to increase the productivity in greenhouse. How to separate the seedlings from complicated background and extract the features of them are the two key technologies. Traditional segmentation algorithms based on threshold (Otsu), edge (ES) and region growth (RG) were contrasted in this paper. These segmentation methods are seriously interfered by disturbances. In view of the disadvantages existed in the present algorithms for seedling image segmentation, an improved segmentation algorithm based G-channel region growth (GRG), which utilized G-channel pixel values only, is proposed. Morphological filter was applied to remove noises existed in the binary image segmentation. Then, four kinds of features of seedling leaves image were extracted through this algorithm. Error rates of Eggplant segmentation were 0.48, 0.52, 0.44 and 0.04 for Otsu, ES, RG and GRG respectively, which indicates that the GRG algorithm for seeding image segmentation is better than others. In addition, parameters of seedlings present a correlation and the consistency of average gray value (AGV) can be an indicator for subsequent recognition. The results show that the goals of optimizing operation time and separating leaves unbroken are achieved in this paper.