Research On Seed Appearance And Purity Detection Method Based On Multispectral Imaging

Seed quality includes external and internal seed quality.External quality includes seed clarity,weight,health,insect infestation,disease spots,damage,etc.,and internal quality includes seed vigor,authenticity,moisture content,and purity.In the external quality of seeds,detection methods such as clarity,weight,and health are relatively mature,while worm erosion,disease spots and damage are mainly identified by manual identification methods,which are time-consuming,laborious and subjective;In the intrinsic quality of seeds,there are many research results such as vigor,authenticity,and moisture content.However,the detection of purity quality mainly uses biochemical methods.Such methods are destructive,cannot be tested in large quantities,and are expensive.Due to the different shapes of seeds of different crops,in order to avoid the influence of morphological characteristics on the test results,three types of seeds with representative shapes were selected.For example,pepper seeds are relatively small and relatively thin;rice seeds are oval and have a certain thickness;Corn seeds are relatively large and thick.Aiming at these three different types of seeds,this thesis proposes a new method based on multispectral imaging technology combined with machine learning to detect the appearance and purity of seeds.This method has the advantages of fast,non-destructive,and easy to automate.The main research contents of this thesis are as follows:In order to detect the appearance of rice seeds and meet economic requirements,a multi-spectral imaging system was built to take images of rice seeds.At the same time,in order to integrate the acquisition of seed appearance parameters and the classification of seed varieties into one software,a multi-spectral imaging processing software was developed.This software includes a login interface,a seed phenotypic feature extraction system,and a seed variety classification system.The seed phenotypic feature extraction system can realize the acquisition of the phenotypic parameters of the seed,and the seed variety classification system can realize the classification of seed varieties.In the appearance inspection test of rice seeds,a multispectral imaging system was used to take a multispectral image of rice seeds,and then segmented into single-grain rice seed images through image preprocessing,and the two-dimensional convolutional neural network(2D-CNN)Non-defective rice seeds and defective rice seeds(including immature,moldy and damaged seeds)are classified.The test results show that the images collected by the built multi-spectral imaging system can well reflect the difference between non-defective seeds and defective seeds.In the pepper seed purity detection experiment,the onedimensional convolutional neural network(1D-CNN)is used to detect the purity quality of pepper seeds.The full wavelength spectrum data of pepper seeds and the characteristic band selected by the continuous projection algorithm(SPA)are used as the classification model.enter.Compare the classification accuracy of the three classification models of KNN,SVM and 1D-CNN under the full spectrum of the spectrum.Under the characteristic band,use SVM and 1D-CNN to classify and analyze the varieties of pepper seeds;in the corn seed purity quality test,use principal component analysis(PCA)to extract and obtain three principal components with a total contribution rate of 97.6%.Visualization of corn seed classification.The study used the 1D-CNN model to compare the accuracy of the three aspects of the seed spectral reflectance,the seed spectral reflectance combined with the seed shape information,and the seed spectral reflectance combined with the characteristic parameters of the seed shape information;at the same time,the 2D-CNN was used for corn seeds.Multispectral images are tested for classification accuracy.This thesis uses multispectral imaging technology and machine learning methods to study the appearance quality of rice seeds,the purity quality of pepper and corn seeds,and develops a multispectral imaging system.At the same time,a multispectral imaging processing software is designed.This research aims to solve the problem of traditional The method detects the problem of time-consuming,labor-intensive,and costly detection of seed appearance quality and seed purity,which provides technical support for real-time monitoring of seed quality.