Design And Implementation Of Croup Emotion And Cohesion Analysis System Based On Deep Learning

Looking back on the main development achievements in the field of computer in recent years,there is no doubt that the rapid rise of deep learning can occupy a place.No matter in the field of natural language processing,speech recognition or computer vision,deep learning is playing an increasingly important role.On the other hand,with the development of society and the deepening of Internet plus concept in people's daily life,people are getting used to the abstract concepts which are difficult to quantify in the past.The development of deep learning has also brought new directions for the research of character emotion analysis:group emotion analysis and group cohesion analysis are particularly concerned because of their subversion to traditional ways.At the same time,it is of great practical significance to study the group emotion and cohesion analysis system based on deep learning for the organizers who need to understand the real feedback of the masses.The rapid development of deep learning technology and the update and iteration of computer software and hardware also greatly reduce the difficulty of the system from the theoretical stage to the realization stage.This paper will explain the design and implementation of the system from the following directions:Firstly,the algorithm of the system studies the neural network structure Densenet and SE module based on deep learning,and modifies some structures of the network after analyzing the task requirements of the system.Using the open data set EmotiC and EmotiW2018 in the Imagenet pre training Densenet and SEDensenet models,three branches of the global,posture and face are used to carry out the transfer learning training for the three classification problems of group emotion,and the best classification results of the three branches are obtained respectively.Then,on the basis of the best classification results of each branch,we use Emotiw2019 data set and the regression model designed in this paper to carry out regression training of group cohesion index in different branches,so as to get the best regression results of the three branches.Finally,we use the method of model fusion to optimize the results of emotion classification and cohesion regression,and build an innovative neural network model that can analyze group emotion and cohesion at the same time.Then based on the above work,each module of the system is designed,the function of each module of the system is defined,and the interaction process between modules is established.Finally,we use Libtorch to complete the API transformation of the model,and use Qt to write the system interface to realize the interaction between the system and the user.The system can be accessed by video or picture data.On the one hand,it can analyze the general mood trend of the current video or picture group,and on the other hand,it can output the cohesion index of the current group to reflect the team cooperation efficiency of the group.Therefore,the system can effectively make the organizers understand the impact of the current activities on the group,and can play a practical role in evaluating the effectiveness of cultural activities in large-scale cultural activities.