Nondestructive Testing Method And Research Of Decayed Cherry Based On Hyperspectral Imaging

(1)Cherry is a kind of high-grade fresh fruit favored by the public,with high nutritional value,and has a good market prospect in recent years.Traditional fruit quality defect detection mainly relies on manual sorting,which is time-consuming and labor-intensive,and its accuracy is easily disturbed.However,non-destructive testing technology gradually replaces traditional testing methods with its advantages of rapidity,accuracy,and non-destructiveness,which is extremely important for fruit quality detection.The research direction of cherry rot defect detection is mainly based on machine vision technology,and technologies such as nuclear magnetic resonance technology,X-ray technology and spectral information technology have problems such as radiation and fruit size limitations,and related applications are less.However,machine vision technology relies too much on image information and is greatly affected by environmental factors.When there are defects similar to the surface of the fruit,the detection rate will be affected.Hyperspectral imaging can collect spatial information and spectral information of fruits,and correlate the physical and chemical properties of fruits with spectral information,which can achieve more accurate detection and avoid the defects of other technologies.Therefore,this paper takes Dalian Sangtian fruit tree plantation Raney,Hongshouqiu and Shamidou varieties of cherries as the research objects and uses hyperspectral imaging technology to achieve non-destructive detection of rotten cherries.The main contents are as follows:(2)This study firstly segmented the decay region of interest and preprocessed the spectral data.The maximum inter-class variance method(Ostu)and illuminance reflection combined with Ostu segmentation algorithm are used to segment the image of the region of interest of decayed cherry.The results show that the illuminance reflection model combined with the Ostu method can segment the defect parts covered by the reflective area more completely;then standard normal variate(SNV)and SG convolution smoothing(Savitzky-Golay,SG)are used to perform spectrum Data preprocessing.The results show that the preprocessing algorithm with SG smoothing is the best.(2)In this study,the feature extraction and screening of cherry decay classification based on hyperspectral imaging was realized.The characteristic wavelengths of three varieties of cherries were extracted by successive projections algorithm(SPA),Competitive adaptive reweighted sampling(CARS)and principal component analysis(PCA)algorithm respectively,which can reduce the amount of data;based on the pixel point defect detection method,the ELM model is established by using the spectral reflectance of the pixel points in the target area of the characteristic band image,and the characteristic wavelength screening is realized.The results show that the CARS algorithm has the best effect;then the block algorithm is used to process the characteristic band image,and the segmented image blocks are processed by Gabor wavelet transform,Gray-Level Co-occurrence Matrix(GLCM)and local binary binary(Local binary patterns,LBP)algorithm to extract texture features;then after the screening of correlation analysis,the original 9 texture features of rotten cherry were screened to 4.(3)This study realizes the classification detection of decayed cherries.Using the spectral value of the characteristic band,the texture feature filtered by the image block of the characteristic band,and the texture feature of the image block and its corresponding central spectral value fusion as the input vector,a Support Vector Machine(SVM),BP neural network(Back propagation neural network,BPNN)and extreme learning machine(Extreme learning machine,ELM)models respectively realize cherry decay detection,and visualize the detection results based on characteristic band spectral values.The results show that CARS+4 texture feature parameters+ELM is the best for cherry decay classification detection,and the overall accuracy rate reaches 94.91%.The results show that the information fusion vector modeling results are higher than any single vector modeling results.To sum up,it is feasible to use information fusion to detect cherry defects.