ایجاد سیستم طبقه‌بندی بر اساس پردازش تصویر جهت جداسازی کیفی زالزالک طی نگهداری در شرایط مختلف (سردخانه، یخچال و محیط)

Introduction: The crucial sensory characteristic of fruits is appearance that impacts their market value, the consumer’s preference and choice (Afsharnia et al., 2017). The task of fruit grading is vital in the packaging industry because there is a great demand for high quality fruits in the market (Liming and Yanchao 2010). Optical sensors have been used for grading, sorting, and fruit quality detection of different crops (Mohammadi et al., 2015). Today, in various agricultural commodity grading systems, computer vision has become an alternative to visual inspection being objective, consistent, rapid, and economical (Chai et al., 2014). Color is the major attribute to assess quality of agricultural products more than any other single factor (Moallem et al., 2017). It represents the degree of maturity, sugar content, acidity, and taste. For instance, in fresh fruit market such as apples and peaches, darker red color represents higher quality fruit than does light red (Cardenas-Perez et al., 2017). Computer vision is a non-destructive method that can be used for inspection and has found to be applicable in both the food industry and precision agriculture, including the inspection and grading of fruits and vegetables (Wan et al., 2018). This paper proposes an automatic and effective hawthorn fruit grading system based on computer vision-based algorithm. Blasco et al. (2008) developed a computer vision-based machine for detecting and removing unwanted material and sorting the pomegranate arils by color. Liming and Yanchao (2009) developed an automated strawberry grading system using image processing technique and graded the strawberry adopting one or two or three indices among shape, color and size. Okamoto and Lee (2009) developed an image processing method to detect green citrus fruit in individual trees to apply for crop yield estimation at a much earlier stage of growth (Mohammadi et al., 2015).
Material and methods:Grading of hawthorn fruits into three quality grades (A, B and C) was conducted by image analysis technique. Physicochemical and geometrical properties of fruits were determined to compare the results of image analysis and visual classification. Color quality parameters ( ) geometrical parameters, weight loss, firmness, total soluble solid (TSS), pH, treatable acidity (TA) and ripening index (RPI) were the measured factors. TSS content was directly measured from the obtained juice using a refractometer. From the same juice the titratable acidity (TA) was determined. The hawthorn firmness was measured by means of a penetration test using a Texturometer Analyzer. In this study, a machine vision algorithm was developed to capture the images of the hawthorn samples, and then it extracted the feature color value to classify quality grade of the hawthorn fruits. For this work RGB images were captured with a resolution of 1936  1288 pixels and stored on the computer in TIFF format. Different color spaces have been compared in previous works in which it was concluded that the CIELab space is the most appropriate for the measurement of fruit color. From each image, squares from the equatorial area of the fruit were cropped, trying to obtain the greatest possible area but avoiding areas with excessively bright pixels. For the calibration process and image analysis, conversions were performed by using the color space converter plugin for the public domain image processing application ImageJ software. Next, the values of chroma (C*) and hue angle (h*) were also calculated. Principal component analysis (PCA) was used to analyze the dataset obtained from the study of the hawthorn storage. The first set of 200 hawthorns was arranged in a matrix of 14 variables ( ، ، ، ، ، ، ، ، ، ، ،  and ) ×200 averaged measurements. This matrix was used to calculate the Pearson correlation matrix, variable contributions, factorial loadings, eigenvalues, and percentage of variance of the principal components related to the original variables. Later, a second model was built including only the color variables and was used to classify the hawthorns into quality grade.
Results and discussion:Principal component analysis was used to evaluate the correlation between variables. A first correlation was performed between physicochemical and color parameters and variables correlated correctly between each other except for L*, but both described the samples variability with 94.2% reliability. Using only color parameters, the samples were described accurately with 97.4% reliability. Two classifiers based on linear (LDA) and quadratic discriminant analysis (QDA) were used to assess the applicability of vision system. Color parameters obtained by means of CVS under laboratory conditions provided an adequate classification of the quality grade of hawthorn fruits and showed a good correlation with the flesh physicochemical parameters measured. LDA and QDA were capable of classifying hawthorn in their correct quality grade with 99% and 99.5% accuracy, respectively.
Conclusion:The current study used image analysis technique to classify hawthorn fruits into three commercially quality grade (A, B and C). The written algorithm captured an image, eliminated the noises, obtained binary image, removed the noise, and ultimately extracted the color features. This demonstrates that it is possible to create a reliable and objective method for the non-destructive evaluation of quality grade. The selection of variables performed using PCA and the classification model built by means of LDA and QDA allowed adequate classification of the hawthorn according to quality grade using only ، ، ،  and . Thus, it was concluded that external color features of hawthorn fruits can be potentially used to classify the fruits with a proper probability.