Nondestructive Testing And Model Optimization Of Moisture Content Of Dried Hami Dates Based On Near Infrared Spectroscopy

Hami jujube is a characteristic dried fruit of Xinjiang,with large flesh,thick purple and shiny appearance,and medicinal aroma.It is a fine nourishing food and medicinal food.Moisture content is one of the important quality parameters of dried Hami jujube.If the moisture content is too small during the drying process,it will increase the hardness and the taste will be worse;If the moisture content is too high,it will make the bacteria proliferate easily.Medium perishable and spoiled.At present,the traditional destructive detection method is also used to detect the moisture content of Hami jujube.Its efficiency is low and time-consuming,which is not suiTab.for large-scale production needs.In response to these problems,this article takes dried Hami jujube as the research object and builds it suiTab.for the dried Hami jujube spectrum collection device establishes a quantitative prediction model of moisture content,and optimizes the model parameters to realize the non-destructive testing of the moisture content of dried Hami jujube.The main research contents are as follows:1.Analyze the characteristics of dried Hami jujube materials to build a near infrared spectrum collection device.Mainly include: NIR spectrometer and supporting software,diffuse reflection fiber,halogen light source,Mitsubishi PLC and computer.And the built-up collection device can adjust the angle of the light source and the distance between the fiber probe and the sample,and can achieve the drying of Hami jujube near infrared at 0 cm / s,1 cm / s,2 cm / s and 3 cm / s Spectral information collection.2.Selection and optimization of acquisition device parameters.Collect the corresponding NIR spectral data under different scanning times,light source angles and the distance between the fiber probe and the sample,and establish the corresponding partial least squares(PLS)and extreme learning machine(ELM)moisture content prediction models.The prediction results of the model under different conditions show that the prediction results are best when the number of scans is 32,the incident angle of the light source is 30 °,and the distance between the fiber probe and the sample is 4 cm.3.Collect the selected 250 samples of dried Hami jujube to collect the NIR spectral information at 0cm/s?1cm/s?2cm/s and 3 cm/s velocity,respectively.In order to simplify the model and improve the prediction accuracy of the model,four characteristic wavelength screening methods(si-PLS,GA,SPA,CARS)were used to screen the characteristic wavelength spectrum under each speed condition,and further used(Norm,SNV,MSC,SG-1-Der).After the four preprocessing methods reduce the noise of the extracted characteristic wavelengths,PLS and ELM prediction models are established.Studies have shown that the prediction results of the PLS model are superior to the ELM model.The Rp of the PLS prediction model after feature wavelength extraction and preprocessing at 0 cm/s?1cm/s?2cm/s and 3 speeds were 18.51,35.7,8.25 and 23.52 higher than the original spectra,respectively.4.Use genetic algorithm(GA)to optimize the weights and thresholds of ELM and BP neural network prediction models.Studies have shown that the prediction result of the GA-ELM model is better than theSVR and ELM models.At 0 cm/s,the prediction result of the GA-ELM after si-PLS+SG-1-Der treatment is the best,and the Rp reaches 0.8208.Increased by 6.36;GA-ELM has the best prediction result after CARS treatment at 1 cm/s,Rp reaches 0.8357;GA-ELM treatment of GA+Norm at 2 cm/s The post-prediction result is the best,Rp reaches 0.8049;at 3 cm/s,GA-ELM has the best prediction result after the CARS+MSC treatment,Rp reaches 0.8181;The research results show that model optimization can improve the stability and prediction accuracy of the moisture content prediction model of dried Hami jujube.The near infrared spectrum acquisition device can realize the adjustment of light source angle and fiber probe and sample spacing,and can effectively collect the near infrared spectrum information of dried Hami jujube;collect the PLS and ELM prediction model established by dry Hami jujube near infrared spectrum under different conditions,and select the device acquisition parameters according to the model prediction results;the combination of different feature wavelength screening methods and preprocessing methods,the number of hidden layer neurons selection,the GA optimization of weights and thresholds of ELM and BP can effectively improve the stability and prediction accuracy of the model;This paper provides some ideas and methods for on-line detection of moisture content in dried Hami jujube,which has certain reference value.