Research And Application Of Industrial Smoke Monitoring Method Based On Deep Learning

With the rapid development of industry and the increasing concern of the country on environmental protection,it has become an important research direction to use deep learning technology to conduct smoke detection for industrial monitoring video.However,there are many problems in the field of smoke detection.The quality of data used by researchers is uneven,and there is no authoritative image data set.Currently,two-dimensional convolutional neural network is mainly used to build the model.It is easy to ignore the spatio-temporal features between video image frames,and the accuracy rate in actual scenes is relatively low.At the same time,there are few studies on the evaluation criteria of video smoke emission levels at home and abroad.Aiming at the specific scene of industrial monitoring,in order to overcome the shortcomings of current smoke detection methods,an industrial smoke monitoring method based on deep learning is proposed,and the following work and contributions are made:(1)A image data set based on industrial smoke scenes is established.Compared with the currently open smoke data sets,the monitoring scene is richer and has higher quality,which lays a foundation for the training and testing of the smoke detection model.(2)A smoke detection algorithm based on U-Net convolutional neural network is proposed.The algorithm uses three-dimensional convolution to build a deep learning model based on U-Net,which effectively solves the problem that the two-dimensional convolution cannot extract the spatiotemporal features between video image frames.At the same time,the model structure is simplified according to the actual scene,the number of parameters is reduced while maintaining the performance.And the "residual block" structure is introduced in the model,which further improves the detection efficiency and accuracy of the system.(3)A series of smoke emission degree indicators are proposed.After detecting the smoke,extract information such as area features,color features from the smoke area,calculate the growth rate of the smoke area and relevant indicators,and comprehensively analyze the smoke emission degree.The paper makes a detailed theoretical study of industrial smoke detection methods,designs and builds a complete deep learning model,and designs an industrial smoke monitoring application on this basis.Finally,the system is implemented and tested.The results show that the proposed deep learning model improves the accuracy of industrial smoke detection,and the missed detection rate is low.The indicators of smoke area can be used to assist the analysis of emission degree.The whole application can be applied to actual industrial smoke monitoring scene and has certain research value.