Research On Detection Method Of Early Mildew Of Corn Grain Based On Hyperspectral Image Technology

Grain is not only an important strategic material of a country,but also the basic material for the production and life of the people.Grain security is closely related to social stability and development,political tranquility and peace,and national economic development.China has always regarded grain security as a key livelihood issue in the grain industry.Among them,as one of the important grain crops in China,the importance of corn quality and safety is self-evident.Corn is easily infected by fungi during storage,resulting in mildew,and then lose its edible and use characteristics,which will bring serious losses to the grain industry and farmers.Early detection of corn mildew and timely response measures are very important to improve its storage stability,delay quality deterioration and reduce loss and waste.Therefore,this thesis uses hyperspectral imaging technology combined with machine learning algorithm to develop recognition methods of different mildew states of corn grains.Firstly,hyperspectral images of corn grains of various grades were collected in the spectral range of 400-1000 nm,and different types of corn grains spectra and image data sets were established.Based on the measured number of fungal spores,the corn grain mildew state was divided into four grades: safe grain,slightly moldy grain,moderately moldy grain and severely moldy grain.Secondly,the Tamura algorithm is used to extract the color and texture features in the image,and the variable combination population analysis(VCPA)method is used to extract 16 spectral characteristic wavelength variables.Combined with the extreme learning machine(ELM),partial least squares regression(PLSR)and support vector machine(SVM)algorithm,the VCPA-ELM,VCPA-PLSR and VCPA-SVM models are established to distinguish the mildew grade of corn grain.Among them,the VCPA-ELM model has an ideal effect on the recognition of mildewed corn grains,and the correct recognition rates of correction set and test set are94.21% and 93.68% respectively.Thirdly,in order to deeply enhance the precision of this model,the swarm intelligence optimization algorithm is used to optimize the VCPA-ELM model.By analyzing and comparing the optimization results,the particle swarm optimization algorithm has a better effect on enhancing the precision of model recognition,and the accurate rates of correction set and test set reach 98.42% and 97.89% respectively.Finally,the different mildew grades are visualized at the pixel level and object level on the corn grain image.In summary,the research show that the hyperspectral imaging technology based on VCPA-ELM model performs well in the early detection of corn grain mildew.It is expected to be used as a reference method for batch and on-line detection of corn mildew and improve food storage security.