Detection Of Beef Freshness By An Electronic Nose

As the economy of our country getting better and better and the daily life of our people being improved day by day these years, the yield and consumption of beef are increasing. Thus, the government and consumers are paying more and more attention to detection of beef freshness and sanitation quality. As the conventional methods for beef freshness detection are so tedious and time-consuming that it is difficult to employ them for real-time beef freshness detection in our daily life. So, there is a great need to establish a method which is simpler, faster, more scientific and objective. Electronic nose is a new burgeoning non-destructive technique of detection. As the computer science and sensor technology become more and more mature, electronic nose developed very fast these years. Nowadays, electronic nose has been employed in food, environment, medical care and some other industries and is a potential way for quality control and detection for food stuffs.This study is dedicated to employing the electronic nose in determination of beef freshness and other chemistry indexes. Analyzing the data obtained by electronic nose through different recognition techniques, the possibility of using an electronic nose in beef freshness detection was checked, the most effective experiment method and pattern recognition technique were founded, and the freshness and other chemical indexes were determined by electronic nose.At first, some experiments were carried out in advance to optimize the parameters of the operation. In different conditions (different temperature, space, sample quantity, space generation time and with or without EDU), samples in different freshness were detected by electronic nose, and the data were analyzed by PCA and variance analysis, and the optimal parameters were: 8℃, 500ml, 25g, 5min, without EDU.In the formal experiment, three batches of beef were detected by electronic nose and at the same time, color, pH, TVB-N, colony count of the samples were also determined. The first batch of sample was fresh beef stored at 2℃ the second batch of sample was fresh beef too and was stored at 5℃;the third batch of sample was frozen beef stored at 5℃.As far as the first batch of sample was concerned, the data of the color parameters were analyzed by PCA and DFA, but no acceptable result was obtained. Since the color parameters were impacted by several factors, such as age of the cow, position, freshness, content of fat and so on, it is difficult to discriminate the samples based on its freshness.Just based on the samples of the first batch, the data obtained by the electronic nose were analyzed by PCA, LDA, DFA and several kinds of ANN. The results indicated that the signals of the sensors changed corresponding to the freshness of the sample which was detected, especially when it came to the sensor 4, 6,7, 8, 10, whose signals correlated with the samples freshness very well. In the score plots of PCA, the samples can be discriminated clearly according to their freshness. The results of DFA indicate that the electronic nose is able to discriminate beef samples with different freshness qualitatively very well. Coupled with GRNN as its pattern recognition technique, picking the data of 15s, 30s, 40s, 50s as the input of the network, the electronic nose can predict the storage time, content of TVB-N, colony count and sensory scqre quantitatively pretty well. The standard errors were calculated and they were 1.36(day), 4.64(mg/100g), 16.12(10~5/g) and 1.31 respectively.Compared with BPNN, GRNN is not only simpler, faster, but more precise and robust In our experiment, it is more effective and yield better results than BPNN. GRNN is a pretty good neural network and pattern recognition technique for electronic nose in predicting qualitatively.Based on the database and network founded in the foundation of the samples in the first batch, the samples in the second batch were detected by the electronic nose. The result was acceptable, the standard error of content of TVB-N, colony count and sensory score were 7.31(mg/100g), 26.59(105/g) and 1.69 respectively. But when the samples in the third batch were detected by the electronic nose, the result was unacceptable, the standard errors of content of TVB-N, colony count and sensory score were 12.76(mg/100g), 60.35(105/g) and 1.93 respectively. This conclusion indicated that different storage conditions have a great impact on the odor and the result of the electronic nose. So, if some samples are detected by the electronic nose, special model or network should be established corresponding to the given storage condition.