Research On Online Inspection Method For Moldy Core In Apple Based On Multi-source Information

Recently the incidence of moldy core in apples increased year by year,the diseases have been found at major apple production areas,which brings lots of damage to apple industry of China.The current detection method of the moldy core in apples fully utilizes the spectrum technology,and realizes the diseases detection based on the transmission and diffuse mode in Near-Infrared region.However,the detection process is complex,inefficient and low degree of automation,which can’t meet the online classification and sorting demands of apple industry on function and capability.This dissertation focuses on the demand of rapid and massive classification of apples with moldy cores in online sorting and classifying,presented an online inspection method for moldy core in apple based on multi-source information,built a discrimination model based on multi-source information.Furthermore,we developed an online system software of online moldy cores sorting.It brings a new approach to realization online inspection of the moldy core in apples.System tests expatiate that system well meet the online sorting demands of apple industry on function and capability.It is helpful to optimize the management of apple commodity quality and core competition of China’s apple industry.The major work and conclusions are as follows:(1)Analysis on main factors correlated with inspection for moldy core in appleAccording to the pathological feature of the moldy core in apple,the main factors correlated with inspection for moldy core in apple were being researched.Demonstrated the feasibility of using near-infrared spectroscopy and density factor to realize nondestructive detection on apple moldy core.In view of the difficulties of online density testing,an inspection method was constructed by characteristic spectrum,fruit shape index,fruit diameter and weight.Based on spectrum data collection platform,701nm~728nm was confirmed as the moldy core characteristic spectrum within 200nm~1100nm.(2)Study on multi-source information processing methodsBased on online inspection platform for moldy core in apple,multi-source information of 108 Shaanxi red Fuji apples were collected.Directing at the characteristics of apple picture,a fruit diameter calculation method were established,the transverse diameter’s average relative error criterion is 2.07%,the vertical length average relative error criterion is 2.52%.Directing at the characteristics of transmission spectrum data and weight,we choose the average of 3 replicates measurements as the sample data.Multi-source data preprocessing has been completed based on “Z-score” data unification method.(3)Design of the discriminant model of apples with moldy coresBased on nonlinear classification and bipartition characteristics of moldy core in apple,the discriminant model of apples with moldy cores was designed.Based on LDA classifier,a comparative model analysis of different factors combination,result showed that the model based on multi-source factors fusion had the best forecasting accuracy.Based on SVM classifier,a discriminant model of multi-source factors fusion was developed,the model accuracy is 90.91%,and the runtime is 5.951 s.Based on random forest classifier,a discriminant model of multi-source factors fusion was developed,and the influence of the number of decision trees in random forest classifier has been analyzed.the model accuracy is 95.45%,and the runtime is 0.054 s.Combined with the actual requirements of online detection,the discriminant model based on random forest classifier was established as system discriminant model.(4)Design and verification test of online inspection system of moldy core in appleAn online inspection system software of online moldy cores sorting and classifying in apple was programmed by C++ language.Based on the online moldy core in apple inspection platform,online inspection of the moldy core in apples was realized.Test results showed that the system can well realize the requested function.It showed that the accuracy of detection to apple moldy core of Shaanxi red Fuji apple has reached 83.3%,and the single apple detection efficiency has reached 1.5s.