Research On Non-destructive Prediction Of Pig Lean Meat Percentage Based On Machine Vision

Food safety and quality problem has become the focus of the general public. As the largest market share of meat products, pork is under higher level of requirements on the quick assessment of its quality by many institutions. The most concern of pig purchase is the lean meat percentage(LMP). Two major assessment are adopted to evaluate the LMP of pigs in domestic market: the first one measures the proportion separately after the segmentation of the pigs within the slaughtering process; the other one is nondestructive testing by involving equipment like back-fat analyzer or LMP determination instrument, etc. The first weighing process is destructive testing while the second method is not applicable for small and medium enterprises due to the high investment of equipment and high involvement of manual work in the actual operation for one by one testing.Machine vision technology aims to capture the target image by an image pickup device, after digitization, the detection and judgement of size, color and shape can be made base on the provided information of brightness and color. Machine vision technology with more accurate judgment and faster detecting speed is seen as an effective substitute to replace the artificial senses. Despite that the machine vision technology is a new subject, it has been wildly applied to many fields and gains more weight rapidly in the modern industry which has a trend of automation and intelligentization.This paper investigates the research progress of pig LMP estimation in the whole industry, made a comparison between different equipment and testing methods and set up this project design with feasibility and practicability after a comparative study. The set of nondestructive testing system design is based on the machine vision technology and adopt the pig shape characteristics as parameters to optimize the efficiency and effectiveness. The proposed project focuses on following primary purposes:1) To define the shape characteristic dimension scale for preliminary parameter selection based on the content of the animal husbandry and biology.2) To establish a model and quantitively investigate the relationship between the LMP of pigs and the corresponding parameters. The statistical analysis based on collected data of various characteristics that acquired and selected from pig entity detection is supported by SPSS.3) Digital camera is used to obtain the side and back image of pigs. Key points can be extracted from the outline of the pigs by digital image processing and analyzing. The key points are then applied as the input of the calculation to get the corresponding parameters for model establishment.4) To verify the feasibility of applying image processing technology to pig shape characteristics detection by comparing the data that extracted from the image with the measured data.5) Set the shape characteristic parameters and the LMPs as input and output, respectively. Then estimate the LMP by establishing RBF neural network model. Meanwhile, run a feasibility test on testing sample and adjust the model correspondingly.6) To establish a nondestructive testing system for LMP based on machine vision technology.The testing result presents that the prediction value of the average accuracy rate reached to 94.5%, which meets the basic needs of domestic market. It suggests the developed method adopting pig shape characteristics is applicable. This non-destructive testing method with higher efficiency and lower cost can provide basis for pig buyers and raisers while making related decisions.