The Research On Image Emotional Semantic Classification And Language Description Method

With the rapid development of the Internet,the social network is gradually becoming mature.At the same time,the content that people published on the social platform includes a lot of emotional information,which may has a great influence on the whole network environment and public opinion communication.At present,the research of image emotion analysis is still in its infancy.The contextual information and detail information of the image have important reference value for image emotion analysis.At the same time,the advantage of deep learning method also provides a good technical basis for image emotional analysis,and the combination of language description features can effectively promote the efficiency and accuracy of image retrieval.Based on the deep learning method,we have studied the model and algorithm of image emotion analysis and description.The details are as follows:To solve the problem that it is difficult for traditional methods to detect similar emotions,an emotional detection model based on convolution neural network is proposed.The model takes the multi target detection and classification as the main task.Make thinning detection of emotional targets in each sample,and the loss function is rebuilt by adding the weight information of each emotional category.Improved the accuracy of the discrimination of similar emotions,and added a rotation detection algorithm to achieve the accurate detection of irregular emotional targets.Finally,several control experiments were set up and verified the validity of the model.In order to deal with the problem of image emotion expression through language description,an emotional language model for multimodal data is proposed,and the language description with emotional characteristics is realized.On the structure,the emotional language model uses the multimodal data as input,integrates the emotion detection model and realizes the extraction of the emotion target feature.Then,it realizes the vector fusion of multimodal data in the alignment network,and finally describes the emotional content of the image in the form of language.Compared with the existing model methods in benchmark datasets,we get better emotional analysis results on the basis of language description.