Research On Anomaly Detection Algorithm Of Water Flow Data Based On Video

Petroleum can be used as fuel,lubricating oil,solvent,etc.It is an important chemical raw material,but many toxic and harmful polluting gases will also be generated during its mining and generation process.In order to protect the environment,it is urgent to monitor pollution sources.It is a new research trend to monitor the water flow through intelligent video monitoring technology to achieve the purpose of gas pollution source monitoring.This topic provides a solution for the detection of abnormal water flow data,and is of great significance to the detection of abnormal water flow data and environmental protection.The main research results and contributions of this article are:Aiming at the problem that there is no dataset related to water flow anomaly detection in the existing research,a dataset for water flow anomaly detection was established through field collection with the help of relevant experts.The dataset contains 50,000 data samples,which are divided into two types,normal samples and abnormal samples.The abnormal sample data is affected by the gas pollution source,and the degree of change in the water flow pattern between the two frames of data before and after the data is more obvious than that of the normal data.The selection of abnormal samples was carried out by experts with long-term relevant work experience.The experts comprehensively considered the detection results of chemical reagents and the degree of change in water flow patterns on the time axis.In order to solve the problem that the traditional feature extraction method cannot describe the change process of the water flow shape and the abnormalities are difficult to enumerate,a water flow anomaly detection algorithm based on Multivariate Gauss is proposed.The algorithm first segmented the water flow data,and then extracted the spatiotemporal characteristics of the water flow.Then according to the degree of change of the spatiotemporal characteristics between the two frames before and after,a Multivariate Gaussian model was established,and finally according to the abnormal probability of the Multivariate Gaussian model judgment the abnormal samples.Experimental results show that the detection accuracy of the algorithm reaches 93.1%,and the detection speed can reach 28 frames per second,which can meet the needs of water flow anomaly detection tasks.Aiming at the problem that the Multivariate Gaussian modeling method does not fully consider the distribution information of water flow shape and the problem that abnormal data cannot be fully enumerated,a water flow abnormal data detection algorithm based on shape flow is proposed.The algorithm first combines the change value of the water flow shape in the time domain with the current shape distribution information,and proposes a new feature extraction method which named shape flow;then in order to enhance the learning ability of the neural network,the generator part of Ganomaly is added fusion layer;Finally,in order to solve the problem of gradient instability during the training process,the residual network is used to optimize the network structure.Experimental results show that compared with Ganomaly,the optimized algorithm improves accuracy by 5% in water flow anomaly detection tasks,and improves accuracy by 1.9% compared with Multivariate Gaussian based algorithms.At the same time,experiment shows that the shape flow based algorithms can overcome a small amount of the influence which caused by fog.