Research On Green Plum Sorting And Detecting System Based On Deep Learning

Green plum is an important fruit resource in China.It has been planted in China for more than three thousand years.It is favored and valued by consumers because of its rich nutritional value.It can also help prevent cardiovascular,urinary,and digestive diseases.However,due to the single industrial structure of the green plum industry,the backward production methods,and the low level of mechanization and automation,the development of the green plum industry is severely restricted.Green plums of different maturity have different acidity content and different deep processing purposes.The maturity of green plums currently used for processing is still judged by fruit farmers or workers based on experience.Workers judge the maturity by judging the color of the green plums or the picking time.This method has strong subjectivity,low detection efficiency,and no reliable guarantee of detection quality,which is difficult to meet the large-scale green plum sorting needs.In addition to sugar and acidity,the surface defects of green plum will also affect the subsequent deep processing.Therefore,based on machine vision and hyperspectral imaging technology,this paper took green plum as the research object,classified its surface defects,predicted the sugar acidity of its internal components,and combined deep learning and machine learning technology to construct a green plum surface defect classification model and sugar.The acidity prediction model,this paper mainly completes the following work:(1)Complete the frame design of the green plum sorting and detection system.According to the needs of the deep processing of green plums,the technical plan for the design of the green plum sorting and detection system was formulated.The overall framework of green plum surface defect classification and sugar acidity testing equipment were designed,the characteristic bands of green plum spectral data were selected,surface defect image preprocessing and spectral data processing were performed.(2)Establish a green plum surface defect classification model.By adding Adam W optimizer and Weighted cross entropy loss function,the Wide Res Net50 model was improved as Wide Res Net50-Adam W-Wce model for green plum surface defect classification.Accuracy and recall rate were used to evaluate the classification accuracy.The Wide Res Net50-Adam W-Wce model was compared with the other four improved networks on performance.The results showed that the model had good performance in surface defect classification with a classification accuracy of 98.95%.(3)A prediction model of green plum sugar acidity was constructed.Using deep convolutional network technology,a green plum sugar acidity prediction model based on Dense Net121 network was built.The mean absolute error and correlation coefficient of the prediction set were used to evaluate the performance of the model,and the prediction effect of the Mobile Net V2 model was compared.The results showes that the mean absolute error of the prediction set based on the Dense Net121 sugar content prediction model was 0.616 and the correlation coefficient was 0.881.The mean absolute error of the prediction set of the acidity prediction model was 0.062,and the correlation coefficient is 0.720.All of them could meet the actual needs of green plum sorting,and the model prediction performance was good.(4)Develop the green plum sorting and detection system based on pyqt.The green plum sorting and detection system had the functions of collecting green plum surface defect for defect classification,dynamic weighing,collecting green plum spectral images for sugar and acidity content prediction and sorting.The green plum quality sorting man-machine interface designed by pyqt was simple and practical,which was convenient for popularization and application.