Research On Image-based Classification And Identification Of Rice Seed Varieties

Rice is one of the most widely used agricultural crops in the world,and rice seeds are the source of rice.The purity of the seeds will directly affect the quality of rice.Therefore,for agricultural production,it is very important to check the purity of rice seeds.Through the efficient identification of rice seed varieties,it can prevent some unscrupulous merchants in the market from replacing the superior with the inferior,and improve the purity of the seeds to ensure the development of the agricultural economy.In the past,the identification of rice varieties was performed by experts or inspectors who manually screened a large number of rice seed samples to determine the variety of rice seeds and to evaluate the purity of the sample batches.The degree is not high.Nowadays,with the development of rice hybrid technology,the number of rice varieties is increasing year by year.Among various rice varieties,some seeds are highly similar in shape,color and texture,and it is difficult to distinguish the difference manually,resulting in a bottleneck in the identification of rice varieties..Now machine vision has matured and has been applied to various fields of life.In terms of rice variety identification,image processing methods can also be used to classify different types of rice seeds,which makes up for the time-consuming and labor-intensive shortcomings of manual identification.Accurately identify rice varieties.This paper uses image vision technology to carry out related research on the classification and recognition of rice seed images.The main research contents and conclusions are as follows:(1)In order to improve the performance of the rice seed classification model,a method combining LDA(Linear Discriminant Analysis,Linear Discriminant Analysis)and Bayes(Bayesian)was proposed to classify four kinds of rice seeds(Chujing 7,Ma dam oil stick,jade Yangnuo,jade needle incense)for classification and identification.First,preprocess the original image of rice seeds,and then establish an image data set.Image operations involve image grayscale,image denoising,and image cropping operations.Second,use Open CV to obtain RGB color features of rice seed images.,perimeter area and other morphological features and GMCL(Gray-Level Co-occurrence Matrix,gray co-occurrence matrix)texture features;third,using PCA(Principal Component Analysis,principal component analysis),FA(Factor Analysis,factor analysis),LDA and LLE(Locally Linear Emding,Locally Linear Embedding)analyze and reduce the dimension of the extracted rice seed image features;fourth,use Bayes,KNN(K-Nearest Neighbors,K-neighbors),SVM(Support Vector Machine,support vector Machine),MLP(Multi Layer Perceptron,multi-layer perceptron)classifier to classify and identify the original feature data and dimensionality reduction data.The test results show that the traditional Bayes model has a test accuracy of 97.7% for the four types of rice varieties,and the optimized LDA＿Bayes model has a test accuracy of 99.6% for the four types of rice varieties,and the recognition rate is higher than the traditional Bayes model.(2)Combining convolutional neural network and support vector machine(CNN＿SVM)for 8 types of rice seeds(Chujing No.7,Efengsimiao,Maba Younian,Yuyangnuo,Yuzhenxiang,Bingliangyou 401,5 Xiangyou 398,Taiyou 398)images for classification and recognition.First,single-grain extraction is performed on the rice seed image,and then the image is subjected to data enhancement operation,and finally the original rice seed image dataset is established;second,the HOG(Histogram of Oriented Gradient,Histogram Figure),LBP(Local Binary Pattern,local binary pattern texture),SIFT(Scale Invariant Feature Transform,scale invariant feature transform descriptor)and the feature vector extracted by convolutional neural network for classification and identification;third,to verify CNN＿SVM The adaptive generalization ability of the original rice seed image is randomly added to the original rice seed image by adding salt and pepper noise points to simulate different degrees of noise and adjusting the color saturation to simulate rice seeds of different years,and then the CNN＿SVM model is used to classify the image.The test results show that the recognition accuracy rate of CNN＿SVM model for the original rice seed image dataset is 96.2%,the average recognition accuracy rate for noise interference and rice seed images of different years is95.8% and 96.1%,respectively,and the time to recognize a single image is 4.57%.ms,the efficiency is higher than the traditional CNN and SVM models.To sum up,the LDA＿Bayes model based on data features can quickly and effectively identify 4 types of rice seeds,and the recognition accuracy is as high as 99.6%;the image-based CNN＿SVM model can identify 8 types of rice seeds with an accuracy of96.2%.Compared with the LDA＿Bayes model,the recognition result can be obtained directly after inputting the rice seed image,without the need for secondary processing of the feature data,which is more convenient.Therefore,the above two models can quickly and effectively classify and identify rice seeds,providing a new method for the field of rice seed detection.