| As China is a great poultry country, hatching has been an important task in poultry farming. Because of their owned deficiencies and the bad effects of the hatching conditions, some hatching eggs would become dead embryonic eggs. In the same time, some unfertilized eggs were put into incubator by mistake. A serious problem was turned out. This dead embryonic eggs and unfertilized eggs were easy to breed bacteria and other pathogens which were seriously impact the normal development embryonic eggs. In this case, more and more embryonic eggs would become dead if these bad eggs were not rejected from the incubator. Supposing that phenomenon happened in large poultry farms, the economic losses were inestimable.Machine vision technology was used in this paper. A reasonable discriminative model was set up, and a non-destructive detection system of hatching egg’s survival, which could content with physical production, was designed. With this detection system, the heavy physical labor work, such as egg candling by hand, would be replaced gradually. It all because the system was able to automatic reject the ineligible eggs on the production line.Problems this thesis was trying to address include:1. Confirming the position of single egg in the image including many hatching eggs. First of this study, the single egg was extracted from the image including42hatching eggs. Completion of this task relied on co-operation between the software and the hardware. In this research, the CCD camera which used in the device had the function external triggers. And there was a photos witch fixed on the external circuit. It made certain that eggs of each incubator tray were taken photograph at the same location every time. So in the images, each egg was in a fixed position. And then, the eggs’position was found according to the pixel coordinates of the centre of each egg. In the meanwhile, each egg was deal with as a circle. With this method, the eggs were extracted from the images.2. Dynamic image processing. In order to fill the need of detection on line, dealing with the images which collected when the eggs was moving was an important problem. In this test, the eggs moved on the conveyer on the speed of2.5cm/s. So the broad images due to moving need to process first. That was the image restoration. In this paper, Inverse Filtering, Wiener Filtering, and Least Square Method were compared, and Least Square Method finally picked to process these images. And sharper images were gained for this study. 3. Setting up reasonable discriminative models. It could be found that different kinds of the hatching eggs with a significant difference from the color. Extracting the statistics values of the images-r-component, g-component, b-component, H-component, S-component, I-component, and comparing them for different incubation period. In the end the maximum subordination principle and the Bayes discriminance was draw into the survivor detection model for7-day-old hatching eggs. For the11-day-old and18-day-old hatching eggs, specific threshold values was found to distinguish the three, live embryonic eggs, dead embryonic eggs and unfertilized eggs. After model testing, it found that the inspection precision of the model was above90%.4. Union of the detection system and classification system. In order to detection and classification did in the same time, and considering with the data processing advantage of computer and the industrial automation reliability and sttong adaptability of PLC, this design adopted computer and PLC communication to complete the classification task. And a industrial automation software was written based on Visual C++. It realized modularization of image collecting, image processing and communication with PLC. According to the classification message from computer, and PLC drove corresponding executive body. Deal with30000eggs per hour was the requirement of this system. |
PDF Full Text Request