| As the most important fruit in our country,apple plays an important role in fruit production,consumption and foreign trade.However,in the process of commercialization,the apple will inevitably be affected by some external forces,resulting in different degrees of bruise.Some apple bruises are not apparent at the beginning of the formation,but the internal quality of the bruised tissues has changed,so it is crucial to detect the external bruises on apples,especially the early minor bruises.In addition,the quantitative and objective description of the bruising degree of apples is not only an important basis for producers and buyers to evaluate apple quality,but also has important significance for improving the postharvest handling of apples.This paper used hyperspectral imaging technology to identify the external bruises and classify the bruising degree of apples,the specific research contents and results are as follows:Firstly,segmented principal component analysis(PCA)for hyperspectral images in a spectral range of 401-1037 nm was carried out,and the near infrared spectrum(780-1037nm)was identified as the optimal spectral region for identifying the bruises.Based on this optimal spectral region with the weight coefficients of principal component images,7 characteristic wavelengths were optimized in this region.On the basis of the PCA operations using the selected wavelengths and image processing operations,an accurate recognition algorithm for apple bruises at different levels was designed.The correct recognition rate of 40 intact samples and 160 bruised samples was 100% and 98.1% respectively,and the average correct recognition rate was 98.5%.Secondly,the average spectra of 157 segmented bruised regions were obtained by applying binary mask,then Kennard-Stone algorithm was used to divide the sample sets to improve the representativeness of the samples.The original spectral data were preprocessed by four different pretreatment methods and their combination methods,and the optimal spectral preprocessing method was determined as the standard normal variate transformation.The competitive adaptive reweighted sampling(CARS)algorithm,successive projections algorithm and the combination of CARS and correlation coefficient method(CCM)were adopted to select the characteristic wavelengths of the spectral data,and based on these characteristic wavelengths,support vector machine models based on grid parameter optimization(GS-SVM)were established to classify the bruising degree of apples.Finally,the optimal bruising degree classification model was determined to be the CARS-CCM-GS-SVM model,and the classification accuracy was as high as 97.5% for test set.Overall,the research results demonstrated that the proposed design scheme can accurately and effectively identify early bruise of apple,and on this basis,realize the accurate classification of bruising degree,which provides a feasible solution for the automation of apple bruise detection,and has positive significance for improving fruit grading technique level and export ratio in apple industry. |
