Research On The Classification Method Of Corn Seed Vitality Grade Based On Infrared Thermal Imaging Technolog

In view of the problems existing in the traditional seed vigor detection methods,such as long detection cycle,heavy detection workload,seed damage and so on.In this paper,a seed vitality detection method based on infrared thermal imaging technology was proposed.Corn seeds,a common staple grain,were taken as the test object.The temperature variation characteristics of seeds during heating and cooling process were extracted based on infrared thermal imaging technology,and various machine learning methods were applied to realize the classification of corn seed vitality.At the same time,the activity classification of maize seeds was realized by the combination of infrared thermal image and convolutional neural network.The aim is to establish a convenient,rapid and nondestructive method for seed vigor detection.The main research contents are as follows:(1)Taking Kenhu No.1 corn seed developed by Heilongjiang Academy of Agricultural Reclamation Sciences as raw material,by comparing different artificial aging methods,selecting high temperature and high humidity aging methods to obtain corn seeds of different vitality levels,and building a seed thermal image acquisition device based on infrared thermal imaging technology.Standard germination test was carried out on the collected seeds according to the number,and the reference value of vigor of each seed was obtained.(2)The cooling temperature curves of single corn seeds were extracted,the data were preprocessed by maximum and minimum normalization,and the modeling was carried out by KNN and SVM methods.By comparing various evaluation indexes of KNN and SVM modeling methods,SVM was finally selected as the modeling method,and the accuracy of training set reached 98%,and the accuracy of test set reached 96.5%.(3)The cooling infrared thermal image of single corn seed was obtained by image clipping and stitching.Data set is constructed after image noise reduction and enhancement.In order to improve the network generalization ability,image affine transformation and salt and pepper noise were used to expand the data set.Len Net-5 network model is used to construct classification model.The accuracy of training set reached98.7%.96.8% on the test set.The results show that both the support vector machine algorithm and the convolutional neural network algorithm can be classified in seed viability level based on infrared thermography.And the convolutional neural network algorithm works better.