Online Discriminant Model Of Blood Spot Eggs Based On Visible Spectrum Technology

Egg is the world's most extensive food which contains rich nutrition.It is also a lot of food processing raw materials.Therefore,the internal quality of eggs related to the people's livelihood and people's health.Blood spot is the detection index of the internal quality of eggs,blood spots detection is the basis to ensure the quality of eggs.In order to ensure the quality and quality of eggs in the field of circulation,the study of blood spot eggs detection has a direct practical significance and research value.In this study,eggs were used as the research object.Online discriminant model of blood spot eggs was studied by optical technique and pattern recognition method,so as to identify the blood spots inside of the eggs.The main contents and conclusions of this paper are as follows:(1)Based on the literature research of blood spot egg,the transmittance of blood spot eggs and normal eggs in the range of 500 nm to 599 nm was analyzed.It was found that there were much differences between this two kind of eggs in this wavelength range,Based on the spectral characteristics analysis,this paper chooses 500nm-599 nm as the full-wave bands,which provides the theoretical basis for the subsequent band feature extraction and modeling.(2)Three kinds of algorithms are used to eliminate the abnormal samples,which are clustering,similarity coefficient and box value anomaly detection.If two or more than two methods judged the sample as abnormal,the sample would be deleted.Based on breaking the abnormal samples to analyzing the characteristic.(3)Data preprocessing method was used in this paper,which including data enhancement transformation,smoothing,derivation,multiple scattering correction and standard normal variate.Finally,the best preprocessing method was determined is standard normal variate by modeling effect.After the spectrum otherness analysis of the three velocity patterns,first derivative and standard normal variate was combined to effectively reduce the error due to the longitudinal deviation of the spectrum.(4)In order to simplify the spectral characteristics of the whole spectrum,the characteristic wavelength is selected by three methods.iPLS algorithm was used to select wavebands,CARS and SPA algorithm were used to select characteristic variables.The comparison of the extracted spectral characteristic variables in different methods shows that the characteristic variables of CARS have the best modeling results.Hence,the characteristic variables of CARS can be used to replace the full wavebands for further analysis.(5)Considering the special requirements of on-line detection,select 5 kinds of method to establish the linear model,including partial least squares discriminant analysis(PLSDA),support vector machine(SVM),Bayes discriminant analysis step by step(SBDA),binomial Logistic regression(BLR),linear discriminant analysis(LDA).The best model should be choice by the accuracy and the running time of model,then many factors suitable for online should be taken into account.The best result was SBDA,it was implanted into online testing procedures,blood spot eggs could be quickly identified based on ensuring accuracy of discrimination.Therefore,this article finally chose SBDA as the final blood spot eggs online discriminant model.(6)Through the update of the model under the uniform condition,the optimal parameter matching setting is found.By updating the model under the speed condition,it is concluded that the parameter combination is more suitable for 2500 / hour detection mode.Through the replacement of the new sample into the model to verify the applicability of the model,the spectra and the internal situation of the wrong discrimination samples based on 2500/hour detection mode has the opposite attribution trend.So the the wrong discrimination samples were added to the training process to optimize the discriminant model with specific and guiding training.Finally,the optimized model is:Y1 is Blood spot eggs and Y2 is normal eggs:Y1=0.242*X512+0.659*X531+3.482*X558+1.187*X566-2.030*X568-0.402*X580+2.359*X586+4.972*X598-6.327Y2=-1.212*X512-0.961*X531+4.888*X558+3.949*X566+1.298*X568-3.335*X580-0.239*X586+2.598*X598-9.964The discriminant results were as follows: 60 blood spot eggs could be all correct,the accuracy rate of 100%,9 normal eggs was discriminated wrong in 118 normal samples,the accuracy rate is 92.4%,the overall accuracy rate is 94.9%.Therefore,the optimized model was considered to implanting to the actual on-line detection of blood spot eggs.