Research On Apple Quality Detection Method And Device Based On Hyperspectral Imaging And Deep Learning

Apple is recognized as one of the highest nutritional level of healthy fruits,and crisp taste sweet,excellent flavor,is one of the world’s four largest fruits,is also the largest fruit in China.Apple quality nondestructive testing and classification technology is to promote the industrial development,enhance the effective means of value-added products,aimed at the present stage of apple quality inspection and grading index of a single,problem such as low efficiency,general with red apple as the research object,this study based on compose a specular image,deep learning and machine vision technology to carry out the apple internal and external quality inspection and grading research,and development testing software and grading device,the main research content and conclusions include:（1）Nondestructive testing of apple soluble solids（SSC）and hardness.Firstly,the spectral data,SSC and hardness values of Red General apple were collected.The abnormal data were eliminated by Monte Carlo partial least squares method and the data set was divided by SPXY algorithm.The characteristic wavelengths were selected by the competitive adaptive weight sampling method（CARS）and random leapfrog algorithm（RF）,and the multiple linear regression（MLR）prediction model was established.The results showed that CARS-MLR model had the best effect on SSC detection,and the main prediction parameters were²=0.80,RMSEP=0.57,RPD=2.11>2;RF-MLR has the best prediction effect in hardness prediction model,and its main prediction parameters are²=0.80,RMSEP=0.83,RPD=2.29>2.A visual distribution map of SSC and hardness was generated according to the optimal model,laying a data foundation for nondestructive testing of postharvest apple quality.（2）Detection of early hidden injury in apple.The hidden damage of apple was made by the device of apple damage generation.The observation showed that the samples within 6 hours of the damage were in the state of hidden damage.Spectral information of healthy samples,1h damage samples and 6h damage samples were collected respectively.Through PCA analysis and waveform similarity matching,we could observe the early hidden damage of apples that could not be detected by naked eyes.The continuous projection algorithm（SPA）was used to select the characteristic bands,and the LIB-SVM implicit damage detection model was established.The overall prediction accuracy was 84.44%.（3）Apple surface defect detection.First acquisition health,minor defects,serious defects sample images,YOLOv3 algorithm was used to build the deep learning model,and Dark Net53convolutional neural network was used as the feature extraction network.After the iterative training,the model detection effect was evaluated,and the results showed that the model detection accuracy was 93%.（4）Non contact detection of apple coloring degree and fruit diameter based on machine vision.Color image segmentation technology is used to segment complex background samples,h-component threshold segmentation is used to calculate the surface red coloring rate,Canny edge detection algorithm is used to extract the contour,and the minimum circumscribed rectangle algorithm is combined with the size calibration coefficient to calculate the actual Apple diameter.The maximum absolute error between the measurement results and the actual measurement is 2.01mm,which meets the classification requirements.（5）Research and development of apple quality detection and sorting device.Based on the quality nondestructive testing model,the quality testing and grading device is developed.The device has the functions of trigger spectrum and image information acquisition,original three machine position image display,non-destructive detection of soluble solids and hardness,non-contact measurement of fruit diameter,coloration detection,automatic grading,traceable two-dimensional code generation and printing,and the overall grading accuracy reaches 90%.The results show that the designed apple quality detection and sorting device can quickly and accurately detect the comprehensive quality of apples with high classification accuracy,which is of positive significance to promote the development of apple industry.