An Architecture Design Of Food Quality Traceability System And Data Fusion Algorithm For Dingfengzhen Co.,Ltd

In recent years, there has been a lot of domestic issues about the safety of food quality which have created a widespread concern. This project designs a food quality supervision retrospective demonstration platform based on the four-tier structure of the Internet of Things and proposes a data fusion algorithm based on Bayesian decisions which is supported by the Science and Technology Development Program of Jilin Province (project number:11KZ03，A platform of food quality and traceability control depending on the Internet of Things) and cooperates with Dingfengzhen food limited liability company. The main purpose of this system is to collect and record production data, predicate the equipment failure and retrospect the quality issues.The quality traceability system architecture consists of five design elements, including data acquisition method and communication networks as hardware, and coding system, database construction and quality retrospective strategy as software. These five design elements are combined according to the four-tier structure of the Internet of Things to achieve six functions. They are, respectively, the storage management of raw materials and products, the processing flow management, the environmental monitoring management, operator management, logistics flow management and retrospective event management.The quality traceability system is all about data collection, processing and storage. In the four-tier structure of the Internet of Things, the object sensing layer is responsible for data collection. We use RFID, two-dimensional code reader and the temperature and humidity sensors and other equipment to collect raw data, then extract relevant information from different data format according to the coding protocol to do the processing and upload. The structure of network decides the data fusion protocol should be cluster-based. Each cluster has its own primary data fusion program and contains a small database to let it achieve the prediction of equipment failure without the server’s support. The information integration layer is responsible for the process of data fusion, which is done by the center server. The center server processes the data that uploaded by each cluster to meet the need of the application layer and stores the useful data into the main database preparing for the later usage.Data fusion algorithm is to deal with the problem of massive data processing. In this system, the storage of production data uses data fusion algorithm to achieve data compression, the equipment failure prediction system uses data fusion algorithm for fault judgment, while the quality traceability system uses data fusion algorithm to provide decision support.In fact，there are many factors affecting the quality of puddings, and the main impact factor of raw materials has a high degree of relationship among its samples. According to this feature, we propose a developed data fusion algorithm to accelerate the the system training speed in the decision-making level of data fusion. The algorithm is based on Bayesian classification decisions, combined with a secondary collection of sample data. This process is to discover the hidden relationship between sample data and the training samples, equivalent to an increase of the amount of training data which has effect on the distribution of sample space.In the end, the paper uses two methods to verify the superiority of the developed algorithm. One is to verify that the developed algorithm can achieve a higher capacity in decision-making under the same amount of training samples. Another is to verify that the developed algorithm needs less training samples to reach the same level of decision-making capacity. Then we conclude that the developed algorithm is superiority than the Naive Bayesian decision and the Bayesian decision with minimal risk decision in this case.