A Study On Near Infrared Spectral Information Processing Technology For Apple Detection

Near-infrared spectroscopy is widely used in the quality inspection of agricultural products due to its advantages of fastness,accuracy,non-destructiveness and low cost.Near infrared spectroscopy has the characteristics of high sensitivity,large number of variables and redundant information.Therefore,choosing an effective processing method to preprocess spectral information can reduce the amount of irrelevant information,improve the validity of data,reduce the influence of interference information on data analysis,and improve the prediction ability of mathematical model.China is a big country in apple production,ranking first in the world and increasing year by year.But the export volume has not increased,and exports lag far behind some developed countries in Europe and America because of China's backword post-harvest commercialization technology.In order to improve the market competitiveness of China's apples and meet the diversified needs of consumers,we should improve the post-harvest processing level of apples.Fruit quality testing is a key step in fruit post-harvest treatment,while soluble solid content is an important indicator of fruit quality.This paper takes apple as the research object and uses near-infrared spectroscopy to detect the soluble solid content of apple.This study takes Red Fuji apple as the research object,and the near-infrared spectrum information of apple was processed in this study.The main work of this study is listed as below:?.We have explored the basic principle and acquisition method of near-infrared spectroscopy and soluble solid content,and collected the spectral data of 440 samples and the corresponding soluble solid content.?.We have used wavelet packet threshold denoising,Savitzky-Golay smoothing filtering,multivariate scatter correction,principal component analysis-mahalanobis distance method to process the spectral data Based on the performance of these methods,the combination of wavelet packet threshold denoising,principal component analysis-mahalanobis distance and multivariate scatter correction(WPD-(PCA-MD)-MSC)is proposed to serially process the spectrum.?.We have used the discrete-degree analysis method based on wavelet packet transform,continuous projection method,feature extraction based on wavelet packet analysis and partial least squares to screen the characteristic variables of the spectrum.In order to further optimize the characteristic variables,the ant colony algorithm was used to select the optimal combination of variables based on the root mean square error of the prediction model.?.We have designed he“pre-modeling spectral data processing algorithm”and“applied near-infrared spectroscopy processing flow based on soluble solid content detection”,and used the original spectral data and spectral data processed by“pre-modeling spectral data processing flow”to establish the partial least squares prediction model.The experimental results show that,compared with the original data,the model correlation coefficient(R_P)of the data training after processing is increased by 0.4196,and the root mean square error(RMSE)is reduced by 0.6299.The BP neural network was used to train the prediction model of soluble solids in apples,and“pre-modeling spectral data processing algorithm”was further verified.This study has achieved the goal of improving model prediction by spectral data processing.