Detection On Internal Defects Of White Radish Based On Hyperspectral Imaging

Radish(Raphanus sativus L.)is annual or biennial herbaceous dicotyledonous plants belong to cruciferae radish genus and its cultivation and planting area is large,it is China's third largest vegetable.With the improvement of people's living standard and the development of production,consumers' demand for radish products has also improved.However,the main threats to the high quality and high yield of radish are plant diseases,insect pests and internal defects.Generally,the main internal defects are hollowness and black heart,hollowness usually presents lignification and internal moisture loss;black heart shows the smooth surface and decay or dry rot heart.At present,radish quality detection is mainly using the manual detection in China,i.e.sampling inspection about abnormal with visual analysis.Visual analysis method is not only time-consuming,but also ineffective.When people find skeptical radish,they must cut radish and proceed inspection.As a nondestructive detecting technology,hyperspectral imaging system was used for inspecting internal defects of radish in this paper.In order to detect the internal defects of sample,we use hyperspectral imaging system to acquire the hyperspectral images of radish,region of interest(ROI)was extracted and optimal wavelengths were selected to build prediction models for evaluating hollowness and black heart from normal samples.1.System of hyperspectral imaging detection for radish' internal defectsHyperspectral transmittance,reflectance and half-transmittance acquisition units were established respectively for the detection of internal defects,and their parameters were decided through preliminary experiments,including light intensity,exposure time,speed of conveyor,height and angle of the light source and so on.2.Analysis on the reason of white radish hollowness development during storageTo explore the relationship between quality parameters and hollowness of radish,in this section,we measured the water content,hardness,crispness,soluble solids content,crude fiber,lignin and other physical and chemical indicators of radish,which were stored in different hollowness grade.Then,the correlation between physical and chemical indicators and hollowness degree was analyze.The correlation coefficient between moisture content and hollowness degree was 0.982.The correlation coefficient between lignin and hollowness degree was 0.985.On the basis of the results and existing literatures,it concluded that the moisture content and lignin content was the main factors to cause the hollowness.3.Detection of white radish hollowness by hyperspectral imaging technologyHollowness is a common defect found in radish postharvest storage.In the study,a prototype hyperspectral imaging system was designed for evaluating internal quality of radish.Three different detecting models including semi-transmittance,reflectance and transmittance were built and used to extract the hyperspectral imaging data of radish,in order to improve the ability to distinguish hollowness radish,different spectral preprocessing methods were compared on the whole spectral band before modeling process,by using the SPA method for optimal wavelengths selection.Four characteristic wavelengths including 623,685,747 and 623 nm were decided.Compare the results of all wavelength and selected wavelength,then,PLS-DA,SVM,ANN algorithms were used to establish the hollowness radish identification models,finally build an identification model to recognize five hollowness degrees.The results showed that hyperspectral transmittance imaging achieved the best prediction results among the three different detecting models.All band model is superior to the characteristics of band model.ANN algorithm was the optimal to build hollowness discrimination model.Hyperspectral transmittance imaging with the combination of ANN got the best results for detecting internal hollowness of radish,with the prediction accuracy of 94.9%and 95.6%on calibration and prediction sets respectively.With the prediction accuracy of 64.3%and 60.2%on calibration and prediction sets respectively,identification model has a low accuracy rate to recognize five hollowness degrees.4.Detection of radish black heart by hyperspectral imaging technologyRadish with black heart shows normal smooth surface,and people can't observe decay or dry rot heart from the appearance,consumers can't accept this kind of disease.By using the hyperspectral transmission detection mode to test the radish,and the method of SPA to select optimal wavelengths,five characteristic wavelengths,including 580 nm,673 nm,747 nm,805 nm and 877 nm were identified.Compare the results of all wavelength and selected wavelength,then combined with PLS-DA,SVM and ANN algorithms,black heart radish recognition models were established.As a result,characteristics band modeling not only to remove redundant information,improve the operation speed,also improves the model accuracy.ANN algorithm was the optimal to build black heart discrimination model,the prediction accuracy of calibration sets and prediction sets were 99.1%and 97.6%,respectively.