Research and application of image recognition-based identification for flesh browning of loquat fruits

【Objective】Loquat [Eriobotrya japonica (Thunb.) Lindl.] is a kind of fruit tree of the genus Loquat in the Rosaceae, maloideae, and its fruit is tasteful, rich in nutrients, and reputed as‘the first fruit of the early spring’. Browning of flesh can affect the quality of fresh -cut product of the fruit. So far there has been no report on the research of fresh- cut loquat fruit. Exploring the fast and efficient identification and evaluation of the flesh browning of loquat fruit is conducive to the efficient screening of browning- resistant germplasm resources of loquat.【Methods】The mature fruits of five whitefleshed loquat resources, including Zhongbai, Sanyuebai, Baixuezao, Guifei, and Guofenben, and five red-fleshed resources, including Zhongshudaxiang, Huangjinkuai, Ruisui, Muluo, and Yanhong collected from the National Loquat Germplasm Resource Nursery (Fuzhou, China), were used as materials. After the loquat flesh was freshly cut, it was placed in a simple soft light photographic light box with fixed light source and temperature. The browning phenotypes of loquat flesh cuts were photographed and recorded in 0 min, 10 min, 30 min and 60 min. And then the Photoshop software was used to pre-process the background purification of the original photos taken by the camera, and the rgb2lab and rgb2hsifunction algorithms of the MATLAB R2022a were used to convert the color space of the pre-processed pictures of the cut surface of the fruit flesh, the recognizabilities of the loquat fruit flesh under each color component of the three color spaces of RGB, Lab, and HSI were compared, and the suitable color spaces were accordingly chosen for measuring loquat flesh. Then based on the MATLAB edge detection algorithm, the Sobel operator was used for binary segmentation of the loquat flesh cut image, to extract the change of Lab value of the flesh image pixels at different time points after cutting, and calculate the browning index according to the formula of color change value. Additionally, the MATLAB ROI function was used to select the irregular representative loquat flesh browning areas, extract the color feature value, and complete the browning area segmentation of the original image by using the Euclidean distance. Finally, according to the characteristics of the Lab value change, the browning index and browning area during the browning process of loquat flesh section, the principal components were extracted and ranked by the affiliation function, by which the browning resistance grading of the 10 loquat resources was comprehensively realized.【Results】A method based on the MATLAB image segmentation algorithm was established to rapidly identify loquat flesh browning, and the CIE-Lab color space transformed by MATLAB could accurately identify the flesh browning phenotypes of the different resources, which would be mostly close to the results determined by colorimeter. The L value and a value could effectively distinguish the flesh color characteristics of the two major types of red flesh and white flesh, and could be used for the flesh color phenotype identification analysis of the loquat germplasm resources. Using the color difference formula to calculate the browning index (BI) of the 10 resources at different time intervals based on Lab values, the browning indexes of the loquat resources with white flesh were significantly higher than those of the red flesh resources, and the browning indexes of Baixuezao, Guifei, Zhongbai, Zhongshudaxiang, Ruisui, and Yanhong increased with the prolongation of time post cutting, whereas thosed of Sanyuebai, Guofenben, Huangjinkuai, and Muluo loquat reached the threshold for browning after 30 min of fresh-cutting, and then tended to keep stabile. The Euclidean distance algorithm indicated that the percentage of browning area in the white flesh type was significantly higher than that in the red flesh type. Among the 10 resources, the browning process of Guofenben was the fastest and the browning area was the largest, while the browning process of Yanhong was the slowest and the browning area was the smallest 60 min post cutting. The study indicated that splitting the browning area could be possible to distinguish and localize the phenotypic differences of browning between the red and the white flesh types more precisely. According to the membership function ranking by principal component analysis, the browning resistance of the 10 loquat resources from strong to weak was: Yanhong, Ruisui, Huangjinkuai, Zhongshudaxiang, Muluo Loquat, Zhongbai, Guifei, Sanyuebai, Baixuezao, Guofenben.【Conclusion】The MATLAB image segmentation algorithm had a wide recognition range and fast computational speed, which would be suitable for quantitative analysis of the color change process of intensive resources. In the evaluation of the loquat flesh browning, the browning index and browning area of the MATLAB algorithm indicated the browning situation from different dimensions, and the combination of the two methods could maximize the characterization of the browning resistance of the loquat resources. The MATLAB image segmentation technique could be used to accurately and rapidly identify the browning resistance of the loquat flesh, and the technique could be also applicable to the identification and evaluation of the other color traits of the loquat germplasm resources.