Freshness Analysis And Research Of Fruits Based On Near Infrared Spectroscopy

As a new-type of detection method,Near-infrared spectroscopy?NIRS?has positive prospect of extensive application in food safety and fruit quality testing,with real-time examination,high accuracy and simple operation.In this thesis,a fruit freshness detection system based on NIRS was designed.Through a series of pretreating algorithm and modeling methods,it can realize variety identification and freshness prediction.The research in this paper are mainly as follows:1.The thesis introduced the principle of near-infrared and explained the generation of NIRS from two aspects:the vibration of functional groups and the reflection or absorption between light and samples.Three collection modes of spectroscopy were introduced,using RMSEP and RMSECV as the index parameters of model evaluation.2.The pretreating methods,data filtering and modeling methods of spectral data were researched,and the SNV method,Standard Normal Variate Transformation and MSC?Multivariate Scattering Correction?were introduced.In order to eliminate abnormal data,the effect of the K-S method on representative data was studied.Beyond that,we've researched Bayes method for classification and recognition.For the freshness prediction of fruits,two modeling methods of PLS and ANN were studied.3.A fruit freshness detection system based on NIRS was designed and built from optical circuit and electrical circuit.In the optical-circuit system,we selected a LED in the range of 640-1150 nm and a spectrometer module,which was adapted to the light source.Electrical-circuit systems were designed for functions such as information acquisition,AD transition and transmission.Through this experimental system,the spectral data of different fruits in different fresh grades were collected.4.In order to improve the accuracy of freshness prediction,the identification experiment of fruit samples was conducted firstly.A classification model was constructed based on training samples data of different level of freshness.The final results show that the accuracy of the fresh-level model is higher than that of the poor-and-corrupt-level models,and the accuracy of apples in the same freshness level is better than pears and fragrant pears.Comparing the spectral correlation of different freshness levels,the spectral difference of apple is the smallest with the change of freshness.The spectral data is influenced by the freshness change,and the recognition accuracy is proportional to the freshness.When the spectral difference of different freshness levels is greater than the spectrum between species,the recognition accuracy is lower.5.Based on the PLS and ANN modeling methods,we conducted the freshness prediction experiment.We built a freshness model directing at each fruit and compared the accuracy and stability of the two algorithm models.Using the RMSEP to evaluate the accuracy of the models,the results show that RMSEP_PLS=0.084,which is larger than the RMSEP of the ANN model,0.079.Compared with the residual distribution,the predicted freshness of the ANN model is closer to standard value.The stability of the model was verified by the leave-one verification method.RMSECV_PLS>RMSECV_ANN,which means the ANN model's stability is better than the PLS model's.