Design And Experimental Validation Of Fruit Quality Hyperspectral Online Sorting Equipment

As a major fruit producer and consumer,the fruit industry has developed extremely rapidly.However,the problem of low commercial output value of fruit is widespread.One of the important reasons is that the quality of the fruit was not graded.The above reasons have reduced the market competitiveness of our country’s fruits.This paper uses hyperspectral imaging technology as the main detection method.We designed a set of online non-destructive testing equipment for fruit quality through the method of combining spectrum and image.The equipment mainly includes hyperspectral image acquisition system,fruit conveying system,lighting system,sorting device and host computer control system based on LabVIEW.The line hyperspectral image acquisition system of this production can select different characteristic wavelength combinations according to different fruits.This not only reduces the volume of data and the processing time of the system considerably,but also allows the internal and external quality of the fruit to be checked at the same time.There are three sorting stations in the sorting and transposition.The equipment achieves the external quality detection of fruits through the collected hyperspectral image information.When the fruit surface is defective,the fruit is removed from the first station.When there is no defect on the surface of the fruit,the equipment sorts according to the collected spectral information and the predetermined sugar content.The control system accurately controls the results from the corresponding second and third stations.It can be seen that the equipment can achieve simultaneous detection of the internal and external quality of fruits.After testing,the detection efficiency of this production line is 10-15 fruits per minute.On this basis the following work was completed using Gannan navel oranges as test samples.1.Establishing a navel orange brix sorting model.The SNV+SD pre-processing method combined with GA-PLS was used for characteristic wavelength selection and model building.Taking into account the time-consuming sorting decision system,11 characteristic wavelengths and their corresponding sorting models were finally selected.The sorting accuracy was 75.28% for brix greater than 11°Brix(the classification index for navel oranges of exceptional brix in GB/T 21488-200811)and 85.92% for brix less than or equal to 11°Brix(general fruit),with an average difference(absolute)between predicted and true brix of0.45°Brix.2.Development of a defective sorting model for surface ulcer disease of navel oranges.For the external defects of navel oranges,the spectral information and image information of navel oranges were used to determine 60 wavelengths and then 7 feature wavelengths using the cuckoo search algorithm,and the 7 identified feature wavelengths were input into a support vector machine(SVM)for classification modelling.The final classification model was determined to have a recognition rate of 100% for defects,with 93.5% fully identified and 6.5% not fully identified.3.The integrated model,which can simultaneously detect the internal and external quality of navel oranges,is finally validated in combination with a complete set of fruit quality hyperspectral online sorting equipment.The results were 94% sorting accuracy for lower fruit position one(defective fruit),88% sorting accuracy for lower fruit position two(exceptional fruit)and 93% sorting accuracy for lower fruit position three(general fruit),with a mean difference(absolute)of 0.6°Brix between the predicted and true sugar values.