| Pork has abundant nutritional value, it has become more and more important role in residents’ daily meat diet in our country. Freshness has become the important index to assessthe quality of pork, and the fresh recognition technology is vitalmethodand assurances to insurethe securityof pork quality. The pork tenderloin was used as the research object, the computer vision technology was regarded as the core, and the pork freshness determination model research was carried out based on color feature optimization, and the mobile application was developed based on the Android platform, and to supplydependable, exact and real-timeway to determine pork freshness was the ultimate goal. The primary study content is as follows:Firstly, elaborated the image preprocessing algorithm used in the paper. The image processing algorithm mainly involves the following several steps.And processing results of the relevant methods under each step were analyzed and contrasted, and then picked up the most suitable method. Finally, the target area image can be completely separated from the background image. In addition, the image acquisition system and the used color model were also be introduced.Secondly, studied the pork specular image highlights removal algorithm. MSF diagram calculation, adaptive threshold calculation and the distinguish condition of diffuse pixels were used in the highlights removal algorithm, and an improved specular remove algorithm based on the actual light chromaticity estimates; Structure similarity method was used to evaluate the processed image quality, and the results show that the bigger SSIM value and less distortion of diffuse image was got when used the improved algorithm than directly used the primary algorithm.So the modified specular remove algorithm could be used to provide certain reference for color feature extraction directly of cover plastic wrap pork.Thirdly, put forward the color feature parameter optimization selection method in the process of pork freshness prediction. 12 commonly used color feature parameters R, G, B, H, I, S, L ~*, a ~*, B ~*, R, G, B from pork tenderloin were extracted, which were combined into five kinds of characteristic parameter combination: RGB-HIS, RGB-L~*a~*b~*, rgb-HIS, rgb-L~*a~*b~* and HIS-L~*a~*b~*. And the pork freshness both BP and the SVM prediction model were built, and the above five kinds characteristic parameter wererespectively used as the input values of the model to determine the freshness level. The prediction results show that when chose the RGB- HIS combination as the input parameters of both BP model and the SVM model, the pork freshness prediction accuracy were the highest, 88.89% and 95.56%, respectively. So it is suggested that choose RGB-HIS combination as the entered values of the neural network when to predict the level of pork tenderloin freshness.In the end, a pork freshness identifying system based on the Android platform was built. For the pork freshness determination mostly executed in the laboratory or other fixed places, to achieve no longer limited to execute in a certain place and be more close to the actual needs, built the pork freshness recognition system based on Android. Described some related system functions, such as image acquisition and transmission in mobile devices, image analysis, identification in the server. The function and the identification accuracy was tested, and it turned out that the function can reach the design goal, the overall recognition rate is 93.33%. |
PDF Full Text Request