Automated Beef Quality Inspection and Grading System

• Main Authors: Chung, Rong-Lun, 鍾榮倫
• Other Authors: Lin, Hong-Dar
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/9mtnsv

碩士 === 朝陽科技大學 === 工業工程與管理系 === 106 === Beef is one of the common foods in our daily life. The nutrients of beef including protein, amino acids and sugars are easily absorbed by human body and become necessary for growth and development. The appearance structure of beef is composed of fat and lean tissue. Different beef marbling levels and lean meat colors significantly affect the grade of beef quality. Excessive accumulation of fat or dark color of lean meat will reduce the quality level of beef. If consumers eat beef with poor quality, it may cause harm to the health of customers. And such defects for the quality of beef processing products also have profound influence. Through the establishment of this study, the proposed system will replace manual inspection operations not only reduce cost waste but also gain a significant benefit in the future. In this study, we select market sold steaks as the testing samples, through the computer with a CCD lens to obtain the beef testing images for processing and analysis. We first use Homomorphic Filter to strengthen the fat portions of a beef image, and apply Curvelet Transform (CT) with high-pass filter method to improve the texture of marbling areas. In the reconstructed image, the background texture is attenuated and the marbling areas are enhanced. Then, we capture the characteristics of marbling textures and lean colors. Beef quality is graded into three levels by Support Vector Machine (SVM) model based upon the feature values of selected lean meat. In the preliminary experiment of this study, 350 beef images were processed to locate the regions of marbling. Experimental results show that the proposed grading system can effectively locate the beef marbling on the surfaces of steak achieves a high 92.68% detection rate, a low 4.97% false alarm rate, a high 94.09% correct classification rate (CR) and the SVM used in beef lean color achieves a high 96.67% classification accuracy.