Research On Intelligent Detection Technology Of Peak Honey Adulteration Based On Three-dimensional Fluorescence Spectroscopy

Honey is a precious natural product,which is of great health and nutritional value.However,adulterated honey damage seriously the credibility of the honey industry,restrict the healthy,rapid and sustainable development of bee industry.At present,many analytical methods have been developed to detect adulterated honey,such as stable isotope ratio method,chromatography,mass spectrometry,etc.However,these methods are often expensive,long testing cycle and complex operation,which are difficult to meet the needs of on-site inspection.Aiming at the limitations and shortcomings of the existing honey adulteration detection methods,this paper uses three-dimensional fluorescence spectroscopy technology combined with deep learning algorithm to study the honey adulteration intelligent detection algorithm based on three-dimensional fluorescence spectroscopy,in order to realize the rapid and on-site detection of adulterated honey.This dissertation carries out research in several aspects:1.The 3D fluorescence spectra of honey were studied,the excitation wavelength range and emission wavelength range suitable for the measurement of 3D fluorescence spectra of honey were determined,and the influence of honey concentration on its fluorescence intensity and spectral distribution was analyzed.The results showed that there was an obvious fluorescence quenching phenomenon in high concentration honey.When the concentration of honey was more than 4%,the fluorescence intensity and spectral distribution showed nonlinear changes with the increase of concentration.According to Lambert-Beer's law,it is determined that the 1%concentration of honey satisfies the additive principle of spectrum.At this concentration,the three-dimensional fluorescence spectrum of adulterated honey is equivalent to the linear sum of the three-dimensional fluorescence spectra of honey and syrup.2.According to the principle of additive principle of fluorescence,a honey 3D fluorescence spectrum data set for deep learning algorithm training and testing was constructed.Based on the convolutional neural network,a network model suitable for the three-dimensional fluorescence spectrum analysis of honey was proposed.By using the convolutional layer to "observe" data from multiple different scales,richer input characteristics were obtained.Adam optimizer and ReLU activation function are introduced to reduce the training time of the network and effectively alleviate the problem of gradient disappearance.The effects of the hyperparameter,such as network depth,convolutional kernel size,position and value of dropout,on the recognition effect of honey 3D fluorescence spectrum were studied systematically,and the model was optimized.The test results show that the model is 97.09%accurate and has good generalization ability.3.Based on struts-spring-hibernate framework,a network platform for intelligent detection of honey adulteration is built.DL4J deep learning framework is used to transplant the above convolutional neural network model for honey adulteration detection to the network platform,so that the platform not only has the function of online detection of honey adulteration,but also has the ability of algorithm optimization.This platform can make full use of the huge computing resources of the server to move the detection of honey adulteration from the three-dimensional fluorescence collection terminal to the Internet cloud,providing a remote,fast and accurate detection platform of honey adulteration.