Study On The Detection Method Of Yellow Peach Bruises Based On Hyperspectral Imaging Technology

As one of the many fruits,yellow peaches can supplement the body with some vitamins and other beneficial ingredients,and it has the function of preventing diseases such as anemia and beauty care.Because of this,yellow peaches have become increasingly popular in recent years and they have a considerable economic value in market.However,due to the soft texture of yellow peaches,they aren’t resistant to store,in order to make the fruit good for sale,a series of treatments such as selection,grading,and packaging are carried out on the picked fruit,and the mechanical damage inevitably occurs in the process.In actual life,for damaged yellow peaches,firstly,bruised fruits are screened out,then,the degree of bruising is tested on the screened damaged fruits,and finally,the storage time of the same damaged fruits is tested and classified.Through the above process,the different damaged yellow peaches can be processed into various forms according to the actual situation and give full play to the commercial value of yellow peaches.Therefore,the paper uses the advantage of hyperspectral"union of imagery and spectrum"to extract the spectral information and image information of bruised yellow peaches,and then uses image segmentation algorithm and machine learning algorithm to detect and analyze the bruised yellow peaches.It provides a theoretical basis for hyperspectral imaging technology in fruit bruise detection.The main contents and conclusions of this paper are as follows:（1）Detection of early bruises on yellow peaches based on hyperspectral imaging combined with band ratio and improved Otsu method.The principal component cluster analysis was used to analyze the three regions of Vis-NIR（390.0-981.3 nm）,Vis（390.0-780.0nm）,and NIR（780.0-981.3 nm）,respectively.It was found that both Vis-NIR and Vis spectral regions along PC1 could distinguish between healthy and bruised tissues.The key wavelength images corresponding to the two regions were selected according to the load curve,respectively,and the PC images and band ratio images were established.After comparison,the band ratio image Q_608.9/689.4 was found to be the most suitable for detecting early bruises of yellow peaches.The global thresholding,Otsu and improved Otsu method algorithms were used to segment the bruised areas on yellow peaches,respectively,and the noise was removed using an adaptive median filter（AMF）.The results showed that the improved Otsu algorithm had the best segmentation effect.However,since the stem-end and calyx area have similar intensity characteristics with the bruised area,the wrong segmentation is caused,the stem-end area and calyx area were removed by curvature-assisted Hough transform circle detection（CACD）algorithm.And all test set samples were used to evaluate the performance of the proposed method,and the overall accuracy of it was 96.0%.（2）Discrimination of storage time of slight bruised yellow peaches based on hyperspectral imaging technology and machine learning algorithms.The pulp firmness,soluble solids,titratable acid and chromaticity values of healthy and different bruised degree yellow peaches samples under different storage times were determined and statistically analyzed.The partial least squares regression（PLSR）models were developed based on the physicochemical indicators and spectral data,and the R_P and RMSEP for pulp firmness,soluble solids,titratable acids,and chromaticity values corresponding to the best predicted model results were 0.7141 and 0.4928 N,0.7176 and 1.4829 ^oBrix,0.7171 and 0.0589%,0.8395 and 2.8948.The value of ripeness index（RPI）was used to characterize the impact damage degree of yellow peaches.The ripeness index（RPI）was used to characterize the bruising degree of yellow peaches.The value of RPI of damaged yellow peach was defined as moderately bruise,and the value of RPI less than 1.2 was defined as severely bruise.The extreme gradient boosting（XGBoost）and random forest（RF）models were built based on the raw and pre-processed spectra.The results show that the XGBoost model under the raw spectrum had the best performance,and the correct rate of the prediction set is 93.1%.（3）Discrimination of storage time of slight bruised yellow peaches based on hyperspectral imaging technology and machine learning algorithms.The spectra of the sample bruise region were extracted as spectral features,and the hyperspectral images were processed using principal component analysis（PCA）.Eight single-wavelength images were selected based on the weight coefficient curves of PC1 images,and the grayscale values of these images were calculated as image features.In addition,texture features such as contrast,correlation,entropy,homogeneity,and angular second-order moments of the PC1 images in four directions（0°,45°,90°,and 135°）were extracted using a gray-level co-generation matrix（GLCM）.The random forest（RF）and extreme gradient boosting（XGBoost）models were built based on spectral features,image features（gray value features and texture features）and spectral features combined with image features.The RF model based on spectral features combined with texture features was optimal,with an overall accuracy of 96.25%.To simplify the model,The competitive adaptive reweighted sampling（CARS）algorithm was used to screen the wavelengths of the normalized spectral features and fused with the image feature data again and build RF and XGBoost models.The RF model based on the screened spectral features combined with texture features had the best discriminative effect with an overall accuracy of 97.25%.