| With the improvement of social living standards,tourism activities have become more and more common.With the popularity of tourism activities and the increase in the number of tourists,there are more and more emergencies in tourism,and it is difficult for the prediction of emergencies in tourism activities.The identification and monitoring of tourism scenes are an effective means to improve the level of response to tourism emergencies.How to perform semantic learning and realize tourism scene recognition and monitoring for these cross-media data is a serious challenge.For the recognition of tourism scenes,on the one hand,it is necessary to improve the accuracy of recognition,on the other hand,to understand the semantics of the scene,and finally to realize scene recognition and tourism scene monitoring.The main work done in this dissertation are as follows:(1)For the problem of low scene recognition efficiency based on scene attributes semantics and time-space limitation of crowd scene analysis,a scene attributes semantic relationship feature extraction algorithm(SASF)and a density field feature based crowd scene analysis algorithm(DFCS)are proposed.Scene attributes semantic relationship feature extraction algorithm can solve the problem of scene attribute semantic sparsity and scene semantic recognition efficiency due to insufficient scene attribute semantics.Experimental results show that the algorithm has higher recognition accuracy and improved feature dividability.The crowd scene analysis algorithm based on the density field feature can effectively analyze the crowd scene,overcome the crowd size estimation and crowd distribution estimation under the time and space inconsistency of the moving crowd scene.The experimental results show that the algorithm can accurately obtain the crowd-distributed state.The size of the crowd,compared to the performance of the current mainstream algorithm has improved significantly.(2)Aiming at the sparsity and imbalance of accumulated attribute semantic data in cross-media data,a cumulative attribute semantic learning algorithm(BDAL)based on deep convolutional neural network is proposed.The deep convolutional neural network structure for cumulative attribute semantic learning is designed.The time-space feature-based crowd counting estimation model is integrated with the cumulative attribute semantic learning model based on deep convolutional neural network,which improves the cumulative attribute semantic learning performance.In order to verify the proposed algorithm and apply it to the crowd counting estimation problem with typical cumulative attribute semantic features,the motion region is obtained by fast moving target segmentation,and the local spatiotemporal features of the super-pixel block in the motion region are extracted to realize the super-depth cumulative attribute semantics learning.The experimental results show that the mean absolute error and the mean squared error index of the proposed algorithm are significantly lower than the current mainstream methods,which indicates that the proposed algorithm can improve the learning performance of cumulative attribute semantics and achieve effective estimation of the crowd counting.(3)Aiming at the difficulty in realizing semantic mapping between cross-media data,and in constructing a cross-media semantic mapping model,a tourism scene recognition algorithm(ARSL)based on cross-media attribute semantic learning is proposed.The semantic relationship topological map is constructed by using the semantic relationship of the scene attributes.The accuracy of the attribute recognition of the scene is improved by the constraint of the semantic relationship topology.The scene recognition model is constructed based on the attribute learning,and the tourism scene recognition is realized.Experiments are carried out on the tourism scene image dataset and the tourist crowd scene video dataset.The experimental results show that the proposed algorithm has higher average scene attribute recognition accuracy than the mainstream method.In the experiment of the video dataset of the tourist crowd scene,the accuracy of the average attribute recognition of the proposed algorithm is improved compared with the current mainstream methods,which improves the performance of the attribute recognition of the tourism scene,and shows the effectiveness of the proposed algorithm.(4)Aiming at the difficult problem of tourism scene emergencies,a scene monitoring algorithm based on crowd scene type identification(CTRNM)is proposed.According to the behavioral characteristics and emotional characteristics and organizational characteristics of the crowd,the crowd scene types are analyzed.The crowd scene type recognition model based on the two-stream deep convolutional neural network is learned,and the scene monitoring rules based on the crowd scene type are obtained.In order to verify the proposed scene monitoring algorithm based on crowd scene type identification,the experimental results show that the proposed algorithm can effectively identify the type of crowd scene and monitor the crowd scene.(5)The cross-media data semantic learning and tourism scene recognition and monitoring system is implemented by synthesizing the scene attributes semantic feature extraction algorithm(SASF),density field feature based crowd analysis algorithm(DFCS),time-space feature and deep convolutional neural network based cumulative attribute semantic learning algorithm(BDAL),cross-media scene attribute semantics learned based tourism scene recognition algorithm(ARSL)and the scene monitoring algorithm based on crowd scene type recognition(CTRNM).The system can realize crowd scene analysis,crowd counting estimation,tourism scene recognition and tourism scene monitoring... |
