Study On Automatic Grading For Beef Color Based On Computer Vision

Implementation of the beef grading system is important to develop beef cattle industry. The trade standard of beef carcass grading has been conducted in our country. The beef color is one of important quality indexes of beef. The color grading of beef is still done by human visual inspecting at present. There are series of problems in manual operation due to the subjectivity, low-level efficiency, and some other shortcomings. Therefore, studying automatic grading method of beef color based on the computer vision technology is of great importance, which can resolve the problems brought by manual activity and promote automation level of our beef cattle industry.A lighting device for capturing system of beef image was designed and then images of beef rib-eye cross-section were collected using this device. Following pre-processing of images, the extraction of the color characteristics was carried out. Finally, two classifiers designed by author were used to classify the beef muscle color and fat color. The main research work in this dissertation was outlined as follows:1. A lighting device for capturing system of beef image was designed and produced. The uniformity, luminance and ability of color manifestation given by this device were analyzed and studied. Experimental results showed that good, uniform and color reproduction illumination could be provided by this device.2. Image processing technique was applied to the pre-processing of obtained images, involved in denoising, removal of the background, the segmentation of the effective determining area etc.. Firstly, space-domain low passing filtering method was adopted to reduce the noise of beef rib-eye cross-section images. Secondly, the method of edge-tracing was used to segment the background of image. Lastly, some image processing methods: OSTU, all direction erosion and expansion techniques and "AND" operation were used to segment the effective determining area for beef color grading. Results showed that satisfied segmentation was obtained.3. Segmented beef images were represented in both the RGB and HSI color spaces. The average and standard deviation were computed for each of the six color components R, G, B, H, S, I. Total 12 color characteristics parameters were used to describe the beef color quantitatively.4. BP Neural Networks and Support Vector Machine classifier were designed to automatic grading of beef muscle and fat color. For the testing samples of beef muscle and fat color, the predicted precision of BP Neural Networks classifier was 95% and 97.4% respectively. At the same time, the predicted precision of Support Vector Machine classifier was 97.5% and 97.4% respectively. Experimental results showed that both BP Neural Networks classifier and Support Vector Machine classifier were effective tool in predicting beef color grade, and the efficiency of Support Vector Machine classifier was better than BP Neural Networks.