Abstract:
Abstract: Pod beak and constriction play an important role in the cultivar identification of peanut. However, little research has been done in the identification of pod beak and constriction based on image processing. To identify pod beak and constriction automatically and evaluate the feasibility of image process techniques for quantifying peanut DUS traits, the identification of pod beak and constriction and quantitative analysis of DUS traits based on Freeman Chain Code were proposed. Twelve peanut varieties from Heibei and Shandong provinces (Rizhao, Weifang, Qingdao, Yantai city) were used and 50 peanut pods were randomly selected from each variety. A scanner was used to capture images of the peanut samples. Preprocessing including image segmentation, graying, enhancement and binarization were conducted to the peanut images. To make the proposed method robust to rotational change, a rotational operation was performed to make the pod in vertical direction. Then, image edge was extracted from the binary image and eight-direction freeman chain code was applied to code the edge. Approximate curvature method and local maximum method were used to detect inner corner points and exclude the pseudo corner points. Local maximum method was used to remove the adjacent corner points and keep the point with the largest curvature within a neighborhood. Pod beak and constriction were identified by their locations. Meanwhile, some other corner detection methods, such as Moravec, Harris and SUSAN algorithms, were used for comparison with the proposed method here. Finally, 3 DUS testing traits were extracted and quantified using image process techniques. Degree of pod constriction was qualified by the location of pod constriction and graded by the value of the pod constriction degree. Distinctness degree of pod beak was qualified by the curvature where the pod beak was located. The length of pod was qualified by the length of its minimal outer rectangle. To evaluate the proposed method, 600 peanut samples were selected for identification of pod beak and constriction. Results showed that the accuracy of pod beak and constriction identification based on Freeman Chain Code were 93.1% and 95.5%, respectively. The false alarm rates of pod beak and constriction identification were 7.3% and 5.8%, respectively. Comparably, the Freeman Chain Code method was better than the other three with the accuracy for pod beak identification of 53.4% with Moravec, 85.5% with Harris and 83.4% with SUSAN, and the pod constriction identification accuracy of 48.5% with Moravec, 82.3% with Harris, and 84.2% with SUSAN. Meanwhile, the processing speed was also higher (468 ms) than the others (1436, 738, 567 ms). Based on quantified DUS testing traits using Freeman Chain Code, the peanut grading accuracy could achieve 92.4% for 524 samples. The grading accuracy of the distinctness degree of pod beak was 91.7% for 436 samples. The relative error of pod length measurement was 5.8%. In sum, the Freeman Chain Code method is effective in image recognition of peanut beak and constriction and also for quantification of peanut DUS trait.