Pork food safety is not optimistic in China. Some pork carrying viruses can damage the liver and kidneys of consumers. At present, the pork traceability system in China mainly relies on the information of the stamp on the pork skin. However, there is no unified stamp standard in China to regulate pork market, and information on the stamp is likely to be destroyed during the circulation process. In this study, deep convolutional neural network (DCNN) was used to extract the features of pork skin, and dynamic inlier selection was used to register pork skin images to achieve accurate traceability of pork. The data set consists of 810 images, which containing images captured from three angles and nine position. The results show that features extracted using DCNN are better than SIFT features, and the average matching rate is 92.59%. Compared with KNN (K-Nearest Neighbor), CPD (Coherent Point Drift), ICP (Iterative Closest Point), our dynamic inlier selection has better registration effect. In our pork skin dataset, the success rate of registration reached 86.67%, which provided a reference for subsequent pork traceability research.
CITATION STYLE
Song, D., Cai, C., & Peng, Z. (2019). Pork registration using skin image with deep neural network features. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11516 LNCS, pp. 39–53). Springer Verlag. https://doi.org/10.1007/978-3-030-23367-9_4
Mendeley helps you to discover research relevant for your work.