การพัฒนาระบบประมวลผลภาพสำหรับการตรวจสอบเส้นดำในการผลิตกุ้งแช่แข็ง

Abstract

The so-called "vein" in shrimps is a digestive tract filled with grit, sand and sediments, easily viewed as a dark string stretching along the abdomen. Presence of vein is restricted in various products, as detailed in the United States Standards for Grades of Fresh and Frozen Shrimp. Deveined shrimps with remnant vein longer than specified therein are considered disqualified. Currently, removing vein and discriminating improperly deveined shrimps are manually handled. Scarcity of local labor drives the industry to increasingly rely on uncertain supply of migrant workers. It has come to the point where necessity of automated quality inspection technologies cannot be denied anymore. This research aims to develop an image-based approach for detection of improperly deveined shrimps. Two hundred shrimp images were experimented by a sequence of image processing tools; preprocessing to gray scale images, segmenting the region of interest (ROI), and extracting significant features. These features include shape measurements, i.e., area and length, and pixel value measurements, i.e., average, standard deviation, minimum, 25th, 50th, and 75th percentiles. In this research, disqualified shrimps were identified by two different classification techniques: discrimination analysis (DA) and support vector machine (SVM). Merit of Principal Component Analysis (PCA), a dimensional reduction technique, on classification performance was investigated. However, it was found that in this research slight contribution of PCA was observed. Highest classification accuracy was obtained from the SVM with a linear kernel function. The success of this research not only fills a void left by past studies, but also assures a promising future of fully automated shrimp quality inspection development.