Creation of image models for inspecting defects in commercial dried fish floss

Abstract

Fish floss is a chopped finely or mashed fish meat boiled in seasonings, then stir FRY until the fish meat is arid and pulverous. In the making of commercial fish flosses, defect inspection is conducted by expertise inspectors using their feeling of contact and sight which could cause misjudgments. When consumers eat fish floss with defects, it may cause harm to the health of consumers. Therefore, this study proposes an automated defect detection method and develop an optical inspection system for commercial dried fish floss. The proposed method applies the curvelet transform with low-pass energy filtering to remove the random patterns of background and delete the angle direction of background texture. The approximated and partial detailed components regarding defects and uniform background are preserved in the low and medium frequency bands. In the reconstructed image, the background random texture is attenuated and the defect areas are enhanced. Finally, the restored image can be easily segmented by an estimated threshold value into two categories namely dark defects, and white background. The experimental results show that the proposed method well balances the trade-off between the recall rate (82.11%) and precision rate (87.62%), and reaches an F-score of 84.78%, outperforming the traditional defect detection techniques in inspection of dried fish floss.