Research On Detection Of Total Number Of Mold Colonies In Tea Based On Hyperspectral Imaging Technology

The total number of mold colonies directly reflect the degree of mold contamination of tea,which is closely related to the quality,drinking safety and exports of tea.Therefore,it is of great significance to realize the rapid and non-destruct ive detecting of the total number of mold colonies in tea.The current detection of the total number of mold colonies in tea is mainly based on traditional chemical detection methods.But they all have the disadvantages of complicated pre-processing procedures,large consumption of reagents and long time-consuming,which are not conducive to practical promotion and application.At present,the non-destructive detection methods of the total number of mold colonies in tea mainly include spectral technology,image processing technology and hyperspectral imaging technology.Because of its relative ly small point source sampling method,spectral technology cannot contain all spatial information of the sample;the image processing technology can only describe the microscopic changes of the tea surface,but can not characterize the changes in the content of the internal components of tea,hyperspectral imaging technology combines traditional imaging and spectroscopic techniques to obtain spatial and spectral information of the sample.Therefore,a rapid and non-destructive detecting method was proposed for the total number of mold colonies in tea based on hyperspectral imaging technology in this paper.The main contents and conclusions are as follows:(1)The mechanism and verification of the detection of the total number of mold colonies in tea by hyperspectral method was analyzed.The physical changes are mainly reflected in the color change of the tea surface,and the chemical changes are mainly reflected in the changes of the organic content of tea.Then,the change of the average spectral curve of tea samples under the ten groups of different total number of mold colonies were analyzed and demonstrated.(2)A method for extracting spectral information and image information from hyperspectral image of tea samples was discussed.In the aspect of spectral informa t io n extraction,the method of segmentation with band ratio combined with threshold was used to implement the segmentation of the background of hyperspectral image of tea samples and the extraction of regions of interest(ROI).Then the average spectrum within ROI was extracted as the spectral information of samples.In the aspect of image information extraction,24 color features of the image were extracted by using the RGB and HSV color space models,and 15 texture features of the image were extracted by using the GGCM method.(3)In the end,the paper quantitatively analyzes the total number of mold colonies in tea from three perspectives of image information,spectral information,spectrum and image fusion information.First,the spectral information was modeled and analyzed.The BP and BA-BP quantitative detection models were established before and after the variable selection based on spectral data.The results show that the IRIV-BA-BP model has the best prediction effect.~2is 0.9332,and the RMSEP is 0.0262.Second,the model was established and analyzed with the image information.The BA-BP modeling analysis was performed on the data before and after variable selection.After variable selection the results show that the texture and color fusion feature model has a better effect,and~2is 0.6118,RMSEP is 0.2140.Third,the model was established and analyzed with spectral and image fusion information.BA-BP quantitative detection model was established based on the data before and after variable selection.The final result shows that the modeling results based on spectral,texture and color fusion information are the best.~2is 0.9520,and the RMSEP is 0.0214.Compared with the modeling results from three perspectives,it is easy to see that the prediction effect of the model based on spectral and image information fusion is the best.Compared with single image information or spectral information model,the prediction effect has a certain increase.Therefore,the method based on hyperspectral imaging technology can realize the rapid and non-destructive detection of the total number of mold colonies in tea.