Research On Influencing Factors And Correction Methods For Discrimination Of Moldy Core In Apples Based On Transmission Spectrum

Apple is one of the most important pillars of the fruit industry in China.In recent years,due to the improvement of living standards,people pay more and more attention to its internal quality information,and the demand for apple has gradually changed from “quantity” to “quality”.As an internal disease of apple,moldy core severely restricts the high quality fruit rate and export rate of the whole industry.The method of detecting moldy core in apples based on near infrared transmission spectroscopy has attracted the attention of many scholars at home and abroad,and has obtained a good detection effect.However,most of the existing studies adopt ideal samples for modeling and analysis,ignoring the influence of sample differences such as fruit color and fruit size on the transmission spectrum.The size of the apple,surface color,position,light source distance detection,and many other factors will affect moldy core discrimination in the actual detection.In order to improve the practicability and advancement of the detection model,this paper makes an in-depth study on the influence factors of apple moldy core discrimination based on transmission spectrum,and puts forward some spectral correction methods considering the influence factors.The research contents and conclusions of this paper are as follows.(1)The transmission spectrum detection platform was built and the spectrum acquisition method was standardized.Firstly,the pathogenesis and disease characteristics of the moldy core were analyzed,and the feasibility of detecting apple internal disease based on near infrared spectroscopy was demonstrated.Secondly,according to the characteristics of spectral transmission detection,a transmission spectrum detection platform based on spectrometer was built.Finally,the principles of sample selection method,abnormal sample rejection method,spectral pretreatment method,data compression method and pattern recognition method applied in the spectral analysis were described in detail.(2)The effects of different surface colors,different test sites,different measuring distances and different fruit sizes on the near-infrared transmission spectra of apple were analyzed.According to the single factor variance analysis of the light intensity value,different measuring distances,different test sites,different surface colors and different fruit sizes had significant influence on the spectrum.Among them,the effect of different fruit sizes was the most significant,and the spectral intensity decreased significantly with the increase of fruit size.(3)A spectral correction method considering the diameter transformation and the spectral transformation was proposed to solve the misjudgment problem of large healthy apples and small unhealthy apples.The attenuation coefficient of light inside the fruit was calculated by the formula,and the attenuation coefficient was used to modify the transmission spectrum so as to convert the fruit spectra of different sizes to the same reference.Experimental results showed that both methods can significantly improve the model recognition accuracy.In the SVM algorithm,the discrimination accuracy of the spectrum modified by the diameter transformation in the training set and the test set increased by 7.84 and 5.89 percentage points,and the discrimination accuracy of the spectrum modified by the spectral value transformation in the training set and the test set increased by 5.88 and 5.89 percentage points,respectively.(4)The internal apple disease detection production line system was designed and developed successfully,and the stability of the system was evaluated based on the method of signal to noise ratio and spectral area change rate.The experimental results showed that the transmission spectra obtained when the fruit diameter axis was perpendicular to the direction of spectral collection were more stable than those obtained in other directions.In order to eliminate the influence of fruit placement direction,a global compensation SVM model based on the spectral information of different placement directions was proposed.The results showed that the model established by this method can eliminate the influence caused by the change of fruit placement direction,and the recognition accuracy reached 100 %.