Akademik Veri Yönetim Sistemi
Journal of Aquatic Food Product Technology, cilt.21, sa.1, ss.72-85, 2012 (SCI-Expanded)
In the second part of the study of the quality evaluation of pollock roe by image analysis, methods to quantify the color defects (green spots, dark strips, dark color, and uneven coloring due to freezer burn) were developed. Dark roes can be detected by their average L* value. Dark strips can be detected by quantifying the percentage of pixels that have an L* value below an L * threshold. Since there is wide variation among the average colors of the roes, this L * threshold value must be auto-adjusting to the color of the individual roe. Green spots can be detected by their darker color and by ignoring red blood vessels by setting an upper a * threshold. In this study, identifying pixels with L* values less than the L * threshold = 66% of the L * average of the roe, and a* values less than an a * threshold = 20 successfully detected dark strips and green spots. Detection and quantification of uneven color and freezer burn required a smoothing of the roe colors to reduce details. The color primitives method was used, with a setting of a color threshold (CT) = 75. The resulting images were analyzed by setting L * threshold values of 60, 65, 70, 75, 80, and 85% of L * average of individual roes. More surface area of the roe was judged as defective with increasing L * threshold. With proper selection of L * threshold, a * threshold, and CT value, image analysis can accurately quantify the color defects of pollock roe. Practical Application Abstract: Automation of pollock roe sorting by color would streamline the operation, reduce error rates, and help with standardization of quality. Combined with other capabilities of machine vision such as sorting by weight, this technology can be used for multiple purposes simultaneously. © 2012 Copyright Taylor and Francis Group, LLC.