Research On Multi-factor Coupled Discriminant Model Of Moldy Core In Apples Based On Transimitted Spectrum

Moldy core is one of the typical internal disease in apple,If be eaten by mistake, it will affect fertility, carcinogenic and damage the immune system. Most of the disease occurred in the core and vicinity of the apple, it is difficult to diagnose just from the appearance and surface through spectral reflectance. Therefore, how fast, no-destructive, accurate detection of the moldy core has become a major problem to be solved for apple industry safety, quality and development. This paper through analysis of existing test methods and models,present a non-destructive detection method multiple-factor coupled of moldy core in apples which based on transmitted spectrum, built a discrimination model fusion the multiple-spectral transmission spectrum and diameter,weight, provide basic model for realize the rapid and non-destructive detection of moldy core in apples.The main contents are:(1) Moldy core in apple’s relevant factors analysis and built the testing platformBased on the analysis of moldy core in apples symptoms, proposed transmission spectrum, diameter and weight factors is determined feasible to discriminant the moldy core;constructed based on the visible/near-infrared broad band transmission spectrum of the testing platform, used wide-band light source irradiated it from the top of apple, the bottom of the optical fiber connected the USB2000+ spectrometer produced by the ocean optics for obtain the spectral transmittance of core data; used the meinatie electronics vernier caliper to acquire apple diameter, used the leqi electronic balance measures the weight of apples to build the moldy core relevant factors testing platform.(2) Moldy core multiple-factors relevant data acquisition and processingBased on the above testing platform designed experiment project. At first, 1920 Luochuan Fuji samples were tested from the radial direction 120 degrees apart to obtain transmission spectrum data, and adopted dark noise removal, non-linear calibration and stray light correction of the full-band spectrum acquired pre-processing data, used the correlation analysis get in touch with the most relevant moldy core of 12-dimensional spectral data;secondly, measured the diameter three times on the radial position in the same direction,calculated to obtain the average; Directly measurement weight data, accomplish three kinds of relation data collection. Finally, 12-dimensional spectral data, the diameter and weight data were min-max normalized.(3) The research of moldy core in apple discriminant model based on Fisher discriminantStudy the construction method of Fisher discriminant model based on principal component analysis(PCA), Aiming at the data set of normalized processing, The input data were analyzed by 4 kinds of different factors, such as the spectrum, the spectrum and diameter,the spectrum and weight, the spectrum with the diameter and weight. built 4 kinds of rapid discriminant model.40 samples of the same period were used to validate the model, Validation results showed that: spectrum,diameter and weight those three factors combined model accuratelyhas the highest rate, is up to 92.50%.(4) The research of moldy core in apple discriminant model based on genetic BP neural networkStudy the construction methods of two discriminant models based on BP neural networks and optimization weight and threshold value of genetic algorithm BP neural network, 4different factors, the same as the above methods, were used to analyze the input data,Constructed 8 kinds of rapid discriminant model. The model result showed that:in the two models, the combination of the spectrum with the diameter and weight of the 3 factors has the highest model accuracy. Among them, the genetic BP neural network is 97.5%, BP neural network is 90%.In the genetic BP neural network, the combination of the spectrum and diameter and the combination of the spectrum and weight have the same accuracy reaching95%, the single factor spectrum’s accuracy is 92.50%.This paper present a discriminant model constructing method for multiple-factor coupled of moldy core in apples based on transmitted spectrum, The model is constructed by 3 kinds of modeling methods for the combination of 4 factors. The results indicated that the method of multiple-factor associated with the transmission spectrum and diameter with weight may accurately detect moldy core in apples. This research also provides feasible ideas for rapid non-destructive detecting of moldy core in apples.