基于文本-图像增强的突发事件识别及分类方法研究

Research on Emergencies Identification and Classification Method Based on Text-Image Enhancement

[目的/意义]丰富的互联网数据为洞悉真实事件提供了多维视角,快速识别突发事件并准确判断其所属类别,有助于各级政府及应急管理部门高效地管理应急情报资源。[方法/过程]文章构建了基于文本—图像增强的突发事件识别及分类的理论模型;通过文本卷积神经网络、视觉几何群网络搭建深度神经网络共同组成Multi-DNN模型;最后以真实的自然灾害类突发事件数据进行实例验证。[结果/结论]通过文本、图像相互增强,多模态特征融合能够提升突发事件识别及分类的准确率,同时在小样本数据的任务处理中仍有良好效果,证明不同模态的数据能够相互补充、相互印证,对其融合处理能够提供比单一模态更为准确和全面的信息分析。

[Purpose/significance]The rich internet big data provides a multi-dimensional perspective for insight into real events,and the rapid identification of emergencies and accurate judgment of the categories to which they belong can help governments and emergency management departments at all levels to manage emergency intelligence resources efficiently.[Method/process]The study firstly constructs an emergency event identification and classification model based on Text-Image Enhancement,and then constructs a Multi-DNN model by TextCNN,VGG-16 and DNN.Finally,validates it with real natural disaster class data by example.[Result/conclusion]Mutual enhancement through text and images,multi-modal data fusion can improve the accuracy of emergency event recognition and classification,while still having good results in the task processing of small sample data,proving that data from different modalities can complement and corroborate each other,and their fusion analysis can provide more accurate and comprehensive information processing than single modality.