Measurement Method of Injury Rate on Fish Skin Using an Image Processing System

Abstract

Nowadays, the conveyor belt system is the most effective method to transport items in factory such as transporting fishes from harbor to factory or transporting fishes on a basement of an image processing system. If the fish is put on the basement of only the image processing system or the conveyor belt system without control, the center position of the fish cannot be matched with the center position of an image processing system. Thus, an effort of designing a conveyor belt system and developing a method measuring injury rate using an image processing algorithm to transport the fishes effectively is very necessary. To do these tasks for this thesis, the following problems are considered. The first one is to present system description of the fish conveyor belt system to be constructed for this thesis. The second one is to design a control law to control a desired angular position of driving pulley of the fish conveyor system. The third one is to develop an algorithm using image processing method to measure injury rate on fish surface. Firstly, the full system includes image processing system and a fish conveyor system. A simple dynamic modelling simplifying the fish conveyor system is used for this thesis. However, the dynamic modelling parameters of the system such as viscous friction, rotation friction and break coefficients are considered as unknown variables. The second Newton law is used to obtain the dynamic modeling of a fish conveyor system. Secondly, a model reference adaptive control (MRAC) law is applied to solve the uncertain parameters of system issue. Two controllers based on the MRAC are described as follows. The sub-controller is to obtain the output signal of a reference model to track a desired angular position of the driving pulley. However, this system is of second order. Therefore, it is difficult to get the stability of the reference model using Lyapunov stability theory. A simple method using the transfer function of the reference model is proposed. The main controller is to obtain the output signal of a real system to track the reference model output signal. The stability of the main controller is surely guaranteed by the Lyapunov stability theory and Babarlat’s lemma. Thirdly, there are several color spaces used in food industry to develop an algorithm using an image processing method. L*a*b* and HSV color spaces are used generally. Therefore, a direct measurement of injury rate on fish surface compared to L*a*b* and HSV color spaces is implemented. Based on this comparison, the K-means or Fuzzy C-means clustering algorithm on an image is considered to show clean injury and body shapes of fish. Median, Gaussian and bilateral filters are used to reduce several kinds of noises such as salt and pepper, random, and Gaussian noise. Moreover, Candy edge detection algorithm is used to detect clearly the boundary between injuries and body fish on color spaces. Furthermore, the Fuzzy C-means clustering algorithm is applied to solve overlapped data issue at the boundary between injury and fish body and measure the injury rate on fish skin more exactly. Finally, the experiments are done on fishes to test the proposed image processing algorithms. The real data of the injury fish is measured by Photoshop software and compared with the experimental results of the proposed image processing method. Simulation and experimental results are shown to verity the effectiveness and performance of the proposed controller and the proposed image processing algorithm.