Construction of the automatic quantification system for the phenotype of Amygdalus mira seeds based on HSV space and fitting ellipse

Abstract: Extracting the phenotypic characteristics of Amygdalus mira seeds is to measure the size of a physical object that needs to operate a large number of peach seeds. However, some phenotypic data is still difficult to obtain at present. In this study, an automatic multi-feature extraction system was proposed for peach seeds using HSV color space and edge point detection. The system included three parts. The first part was the collection and image acquisition of Amygdalus mira seeds. Specifically, the Amygdalus mira seeds were collected from the scientific research institutions, and then seed images were captured using a small studio and digital camera. The second part was the image processing of peach seeds. First, the region of interest was obtained on the original image of peach seed, then converted from the RGB to the HSV color space. The threshold segmentation was then selected using the HSV space, in order to remove the seeds from the original image. The purpose of threshold extraction was to determine what threshold range of H space was used to segment the seed kernel and background and then determine the best segmentation of S space under the H threshold range. Finally, the V space threshold was selected in the threshold range of H and S space with the best segmentation, in order that all pictures were the same set of segmentation thresholds, further to realize the preliminary segmentation of peach seed. Binary morphological operations were then utilized to revise the under- and over-segmentation. The third part was the feature extraction and quantification of seeds. First, the morphological features were achieved, including area, shape index, and seed tip state. Specifically, the edge points of seed kernel images were detected to draw the fitting ellipse and separate the tip of seeds. Among them, the tip state was evaluated using the area and sharpness of the seed tip. Subsequently, the color and texture characteristics of the peach kernel were obtained using low-order moments and gray-level co-occurrence matrix. As such, the quantitative analysis was realized for the nucleus phenotype of Amygdalus mira seeds. Additionally, the extracted color features included the first-, the second-, and the third-order moments. The texture features included contrast, energy, homogeneity, and correlation. A comparative experiment was conducted to evaluate the RGB and gray threshold. It was found that the HSV threshold presented a better segmentation, indicating the highest accuracy rate (99.7%), average accuracy rate (98.9%), and Intersection over Union (IoU) (97.4%). In addition, the extraction experiments of morphological, color, and texture features were carried out to further verify the performance of the system. The results showed that there were quite different phenotypic characteristics of different seed individuals. At the same time, the H-mean and S-mean moment showed a downward trend, as the color depth of seed gradually deepened, compared the extracted color features with the visual. The same comparison experiment was also performed on texture features. The contrast increased, while the homogeneity decreased gradually, as the depth of grooves on the seed increased gradually. The energy and correlation decreased gradually when the surface texture of seeds was much clearer. In summary, the extracted characteristics of color and texture were more consistent with that of the visual, indicating the quantitative texture of Amygdalus mira seed kernel. Consequently, this system can be expected to realize the extraction and quantification of kernel tip state, color, and texture features. The finding can also provide the data foundation and technical support for the breeding research of Amygdalus mira.