Research On Recommendation Of Rice Classification Based On Text And Images

At present,most of the rice quality detection methods on the market are distinguished by the naked eye,which consumes a lot of time and manpower,and the detection results contain human subjective colors,resulting in low detection results of rice.With the rapid development of computer technology,AI With the advent of the times,more and more work done by people will be replaced by machines.As a kind of computer technology,machine analysis technology has been widely used in various fields and has shown good results.We applied machine analysis technology to rice field,which can make up for the problem of strong subjective emotions when artificially distinguishing rice,Machine analysis technology can be more fair and objective when analyzing rice grades.Traditional rice classification mainly focuses on images,and has made certain research progress.Rice reviews can reflect other users' comments on the rice,but the amount of text in the comments is very small,and there is a problem of data sparsity of the comment information.The research on rice classification system based on multimodality is relatively few,which also makes rice,text and video and other multi-modal data classification methods become a new direction of research and exploration.Aiming at the multi-modality of the data of rice commodities,this paper proposes a rice classification method based on the fusion of typical correlation analysis images and text features,and uses rice image and text feature fusion methods to classify rice commodities and improve the accuracy of rice classification.By extracting the contour features of the rice image,four parameters of rice perimeter,area,length and width are obtained,and the unsupervised Kmeans clustering algorithm is used to cluster the extracted features into three clusters and find the Euclidean distance of feature vectors in the training set.In the extraction of rice text features,first of all,we need to use the word bag feature method to convert rice text comments into text feature vectors,and then use the word frequency inverse word frequency(TF-IDF)method to find the importance of each feature word in the text feature vector Score,and finally use the chi-square statistical method to calculate the distinguishing rice text features.Based on the Canonical Correlation Analysis(CCA)model for the fusion of rice image and text features,a random forest model is used to fuse feature vectors after CCA fusion.Experiments show that the model used in this paper improves the classification accuracy compared to the single text rice classification and single image rice classification models,solves the problem of sparse data of single images and single text,and improves the accuracy of rice classification.