Research On Video Captioning Of Abnormal Events Based On Attention Mechanism

In the coal industry,how to intelligently understand the semantic information in the mine underground monitoring video and describe the abnormal events is one of the effective means to prevent and mitigate the frequent occurrence of safety accidents.The data quality of coal mine surveillance video is poor,which will have a certain influence on the detection of abnormal events in the video and the semantic description of text in the video.Traditional video anomaly detection methods are mostly rely on manual characteristics,the identification accuracy is not high,low degree of intelligence,the problems such as insufficient power of expression,use deep learning network can complete independently extracted features and strong generalization ability,but there are still in the study of abnormal events video semantic description of abnormal events recognition rate is low,the text described the problem of accuracy is not high,to solve these problems of abnormal events in the coal mine monitoring video text description methods are studied,the main research content is as follows:(1)Aiming at the problem of abnormal event detection in coal mine video scene,a VGGNet-LSTM video anomaly detection method based on attention mechanism is proposed.Firstly,the video frame sequence is preprocessed to reduce the impact of video quality on the final prediction classification result;secondly,the global feature representation of video frame is obtained by using VGGNet network and input into LSTM network;then,visual attention mechanism is used to fuse into VGGNet and LSTM network to assign different weights to the targets in the video frame;finally,softmax is used to obtain the global feature representation of video frames Make classification forecast.The experimental results in the coal mine video data set show that the proposed method can effectively improve the accuracy of abnormal event detection.(2)Aiming at the video caption problem of abnormal events in coal mine video scene,a video text description method based on the fusion of attention mechanism and multiple features were proposed and applied to the coal mine monitoring video containing abnormal events.Firstly,this paper extracts the global features,local features and video features of the video frame sequence to obtain more visual feature information,and weights the features to achieve early fusion.Secondly,the fused features are trained on the text semantic description model.In the decoder part,an attention mechanism is introduced to automatically assign weight to the generated words to realize the later fusion.Finally,the text semantic description of the video exception event is output.The method proposed in this paper is verified by experiments on video data sets in underground coal mines,and the results show that the method proposed in this paper can effectively improve the accuracy of text description of video abnormal events.