Research On The Sorting Of Defects,SSC And PH Of Green Plum Based On Artificial Intelligence

Green plum contains a variety of natural acids such as citric acid,which is indispensable for human metabolism.It is a rare alkaline fruit that has been discovered so far.Green plum is rich in amino acids,lipids,inorganic salts,vitamins and trace elements.It has high nutritional value and medicinal value.It is extremely beneficial to the normal progress of protein composition and metabolic functions for the human body and has obvious preventive and curative effects on widespread diseases like cardiovascular,urinary,and digestive diseases.In actual production,the composition control of raw plums is mainly based on the experiences of workers by controlling the picking time.Generally speaking,the plums for producing plum essence are picked at a 70% ripe stage,the plums for green plum wine are picked at an 80% ripe stage and the plums for fresh fruit consumption are picked at a 90% ripe stage.At present,the defect classification and composition determination of green plums are still done manually,and there are problems such as low efficiency and low quality,which can hardly meet the production needs of green plum products.In addition,the appearance quality of green plums also has a great influence on the classification of green plums.Different appearance qualities of green plums are used in different occasions.Therefore,this paper took green plum as the research object,and based on artificial intelligence technology,studied the classification of surface defects of green plum and the rapid nondestructive testing method of internal sugar acidity.This paper mainly completed the following research contents:A new hyperspectral imaging system was built to ensure the stability of the system in the process of sample data collection.The green plum samples were selected to collect the green plum hyperspectral image data,white and black correction was carried out on the image and the spectral characteristic curve of each green plum was extracted as the spectral information of the green plum.The results of SSC and p H of green plum were collected by traditional physical and chemical tests,which were used as the composition information of green plum to establish the foundation for subsequent model study.A static surface image acquisition platform of green plum was built to provide early technical support for the final realization of fast dynamic defect classification and sorting.On the basis of VGG network model,the random weighting average optimizer SWA(Stochastic Weight Averaging)and w-softmax loss function were used to adjust VGG network to meet the requirements of the classification of green plum feature.Migration learning was used to classify and train intact green plums and defects such as rain spots,rot,scars,cracks.The recall rate,accuracy and F1-Measure were chosen to judge the effect of the classification model.The classification result was compared with the original VGG and Res Net-18 network.The results showed that the average accuracy rate of the green plum defect classification network reached 93.56%,the loss value decreased rapidly and the obtained loss value got lower during training.The model convergence speed was fast,which verified the superiority of our green plum defect classification network.Based on the XGBoost model,the XGBoost model was improved by adding Kernel Principal Component Analysis(KPCA)and Linear Discriminant Analysis(LDA)to predict the acidity of green plum.The correlation coefficient and root mean square of prediction set and cross validation set were selected to evaluate the prediction effect of the model.The results were compared with the prediction effects of different kernel functions in KPCA,XGBoost,KPCA-XGB,LDA-XGB and KPCALDA-XGB models.The results showed that the sugar content and acidity prediction model based on KPCA-LDA-XGB had good performance in the prediction of sugar content and acidity of green plums.The correlation coefficients of the prediction set based on KPCA-LDA-XGB model were 0.956 and 0.829,respectively,and the root mean square errors of the prediction set were 0.056 and 0.107,respectively.The automatic sorting system of green plums was designed.The green plums were graded according to the results of defect classification and sugar acidity prediction,and the grading results could be directly displayed on the interface.