Document Type : Research Paper

Authors

• Hassan Masoudi ¹
• Seyed Mahmood Asadi ²
• Gholamreza Akbarizadeh ³

¹ Associate professor, Department of Biosystems Engineering, Faculty of Agriculture, Shahid Chamran University of Ahvaz, Ahvaz, Khuzestan, Iran

² Graduated M.Sc. student, Department of Biosystems Engineering, Faculty of Agriculture, Shahid Chamran University of Ahvaz, Ahvaz, Khuzestan, Iran

³ Professor, Department of Electrical Engineering, Faculty of Engineering, Shahid Chamran University of Ahvaz, Ahvaz, Khuzestan, Iran

Abstract

Introduction: In sugar factories, control of sugar crystals growth in the granulation stages is very important to produce sugar grains with a special and required size. Machine vision systems can determine the size of sugar crystals. The main challenge of image processing systems is the lack of capable algorithms to separate contact and overlap crystals accurately. So far, various algorithms have been developed to detect crystals and remove their overlapping. However, these methods have not been able to fully detect and separate the overlap of crystals. The purpose of this study was to provide an appropriate image processing algorithm for determining the size of crystals in sugar baking solution (massecuite), which gives us the characteristics of size and shape for the particles in the baking pot instantly to evaluate and improve the quality of the final product.

Materials and Methods: The massecuite samples were provided from Debal Khozaei Sugarcane Agro-industrial Company, Ahvaz, Iran. After preparation of the sugar crystals samples under lam and lamer (microscope slides), a digital camera with two Megapixel resolution, attached to a ZS9 Olympus microscope, was used for image capturing. Then, using MATLAB image processing toolbox, the color image (RGB) transferred to grey scale. A mixture of structural operations such as erosion and expansion with spatial filters including median filter were used to remove the image noises. The function of histogram local adjustment was used to improve image contrast. Three methods of segmentation including convexity, fuzzy clustering, and multiplicative intrinsic component optimization (MICO), along with their combination were used to segment the image of massecuite crystals. A reference image was used to determine accuracy of the image processing algorithms. To do this, the massecuite crystals image was manually segmented by Image J software. All segmentation algorithms were applied on the reference image, and seven geometrical parameters, including the mean aperture (MA), coefficient of variation (CV), and standard deviation (SD) were calculated for all the sugar particles in the image. Finally, the percent of MA measurement error was calculated for each sugar crystal to find the best algorithm.

Results and Discussion: In manual segmentation, the number of sugar crystals in the selected image was 26. In the manually segmented image, the average of MA, SD and CV for sugar grains in the image were 0.422 mm, 0.157 mm and 37.18% respectively. The relatively large CV of the calculated geometrical parameters indicated the non-uniformity of the sugar particles size inside the massecuite. The convexity method was able to perform well in some areas of the image, and in some other areas, it could not detect the contact between the crystals. The value of the SD and CV of all the geometric parameters determined by the convexity method were greater than the reference values determined by the manual segmentation. This indicates the weaker performance of this method in determining the sugar crystals size compared to the manual method. The values of SD and CV of all geometric parameters determined by the combined fuzzy-convexity method were greater than the reference values, but lower than the values of the convexity method alone. So, the combination of the fuzzy clustering method with the convexity method improved the segmentation performance of crystal images. The SD and CV values of all geometric parameters determined by the combined MICO-convexity method were greater than the reference values, but lower than the values of convexity and fuzzy-convexity methods. This point shows the better performance of the combined MICO-convexity method in segmenting the images of sugar crystals compared to the other two methods. The average of MA, SD and CV for sugar crystals in the image were 0.382 mm, 0.150 mm and 39.23% respectively and had no significant difference with the reference method values in 5% probability level. The mean error of MA determined by the combined MICO-convexity algorithm was 13.24% and Pearson correlation factor was 0.88. As a result, the combined MICO-convexity method was proposed to determine the size of sugar crystals in massecuite.

Conclusion: After applying different algorithms on the selected image of sugar crystals in massecuite, it was found that the combined MICO-convexity method can separate sugar crystals well. Also, the CV obtained for this image segmentation algorithm was not much different from the CV of the manual reference method, so this algorithm can be used in the image processing system of the massecuite crystals.

Keywords

• Massecuite
• Digital microscope
• Image processing
• Convexity method
• Fuzzy clustering method
• MICO method

Main Subjects

• Post-harvest technology