Fruit Detection Research Based On Near-infrared Spectroscopy

The fruit sown area in China is the first in the world,and the total fruit production also increases year by year.However,the export volume of China's fruits only accounts for 1.8% of the total fruit production.This is ultimately due to the backwardness of China's fruit quality testing technology.In order to meet the fruit export demand and the increasingly diversified demands of domestic consumers on the quality of fruits,it is necessary to improve the level of fruit quality testing in China and achieve accurate fruit quality testing.Therefore,this paper researches on fruit quality detection based on near-infrared spectroscopy analysis technology.The main research contents and conclusions of the paper are as follows:(1)According to the characteristics and experimental requirements of three near-infrared detection methods,a near-infrared diffuse reflectance experimental system was designed,the experimental platform was built.The five points which divided the banana into copies were the data collection point,the three points along the equator(approximately 120° apart)of the cherry tomato and jujube were taken as data collection point.The four points on the center circle of apple that are symmetrical to each other and one point on the central axis are data collection points.(2)Spectra of different individuals in the same sample of apple,jujube and banana were studied,as well as the spectral difference between the whole sample and the slice.And the spectral changes of bananas(jumpy fruit)and fresh jujube(non-climacteric fruit)putting for one week were studied.The experimental results show that the experimental data can completely reflect the internal structure of fruits in bananas,jujube and apple.The wavelengths corresponding to the three sample peaks are similar.Although there are differences in the spectra of different individuals,the waveforms are similar and the crest positions are basically the same.The internal information reflected by the slice is more comprehensive.The spectral fluctuations of the leapfrog fruit after picking are greater,and the absorbance shows a trend of rising first and then decreasing,but the spectral change of the non-impulsive fruit in the short term is not obvious.(3)Taking cherry tomatoes as research object,the sugar content of cherry tomatoes was predicted based on near-infrared spectroscopy quantitative analysis technology,including pretreatment,model establishment and model verification.Preprocessing: Savitzky-Golay method was chosen for smoothing by comparing 5 kinds of smooth denoising methods,and four data with cumulative contribution rate higher than 99.99% were randomly selected according to the principal component analysis results to obtain the input matrix of the neural network.The BP network model was established: the pre-processed spectral data was used as the input of the network,and the sugar content of the cherry tomatoes was measured as the output.Multiple models were obtained by adjusting the dimension of the input layer,the number of neurons in the hidden layer,and the network training function.The cross-validation of the model determines the coefficient,average absolute deviation and the resulting map of the predicted and measured values were compared.And the model with the best predicted effect is 80-12-1.At this time,the cross-validation coefficient of the model is 0.8328,the average absolute deviation is 0.5711.Validation of the optimal model: pretreatment of the spectral data of twenty cherry tomatoes was taken,the first 80 principal components after the principal component analysis were used as the input of the network,andit was simulated through the optimal model,the values of the predicted and measured sugar content are similar,and the cross-validation determination coefficient is 0.9280,so the optimal model can achieve sugar content prediction for the cherry tomatoes.