The Research On Early Recognition Of Apple's Tiny Bruise Based On Optical Parameters

Fruits are susceptible to bruise during the picking and transportation process due to external forces.Machine vision technology can grade fruit damage according to defects of the fruit surface with the development of machine vision and image processing technology.But the problem of fruit bruise,especially the early bruise problem is difficult to solve with machine vision.The tissue with early bruise occurs in the fruits 10 mm radial distance of the skin.The bruise tissue's color is not much different with normal tissue,and difficult to distinguish with naked eyes because of the low contrast.Especially for mature fruits that the damage is usually red or yellow,the result of detection is worse by mechanical visual method.The tissue of bruise is easily oxidized and tissue's color becomes brown and will be soft to rot over time,even will affects the surrounding good fruit.In this paper,taking the Fuji apple as example,the single integral sphere system was used to measure the optical parameters of the apple tissue with no early bruise and two different degrees of early bruise.The models of support vector machine(SVM)and probabilistic neural network modeling(PNN)were set up to identify from the processed data of optical parameters.Except one apple,the others were validated.The identification accuracy is more 98%.Here are brief summaries about the main content.(1)The mechanical properties of apple tissue and the stress level when the damage was achieved were measured by the texture analyzer.The finite element model of Ansys-Workbench was established by combining mechanical parameters and geometric model of Solidworks.The maximum stress of simulating and the stress value of the experiment were compared.The results showed that the apple did not get hurt in cutting process.(2)This section introduces the single and double integrating sphere technologies and this integrating sphere device.The validation of this system is the emphasis to describe.The absorption coefficient and reducing scattering coefficient of the system are verified by pure water and 10%-Intalipid respectively.The result proved that the system is reliable.(3)The whole process of optical parameter measurement is described.First,the parameters of the integrating sphere system were determined,the refractive index,diffuse reflectance and diffuse transmittance of the sample were measured,and the optical parameters of the sample are obtained by the inverse doubling method(IAD).The reason of the absorption coefficient with on peaks at 680 nm and the influence of physiological saline to optical parameters were discussed.(4)Two kinds of preprocessing schemes,MSC and SNV,were used to deal with the baseline deviation of the original data.The Principal Component Analysis(PCA)based on linear transformation and Stochastic Neighbor Embedding(SNE)based on nonlinear transformation are used to reduce the dimension The results of the principal component analysis are better than the stochastic domain embedding method(5)The Support Vector Machine(SVM)and the Probability Neural Network(PNN)had get high accuracy after the date processing with SNV and PCA.The PNN has the advantages of higher speed,the power of overcoming the baseline drift and better performance adjustment than support vector machine.This study shows that the optical parameters have good consequence in recognition of the early tiny bruise of apple.SVM and PNN have high accuracy to identify 3 types of apple with different bruises.This study shows the application prospect of optical properties in identifying and predicting the early tiny bruising of apples.