Research On Technology Of Gas Monitoring With Infrared Hyperspectral Video

With the continuous development of human society,many gas pollutants are generated in the process of industrial production,which is harmful to humans and environment.However,many gases are odorless and invisible to human eyes,and usually diffuse rapidly with time.These factors make it difficult for ordinary imaging sensors to monitor their distribution.Owing to the fast development of thermal infrared remote sensing technology,in recent years,researchers have been able to acquire hyperspectral video images of thermal infrared band with high temporal resolution.Since many gases have unique spectral characteristics in the thermal infrared region,based on the processing of thermal infrared hyperspectral video images,researchers can effectively monitor the distribution of gas existing in the imaging scene.The main work of this paper is to study the monitoring methods of gas target in infrared hyperspectral video images,which mainly consists of the following three aspects:First of all,starting from the infrared hyperspectral gas imaging model,the basic principle of thermal infrared radiation is studied,and the imaging process of hyperspectral image in thermal infrared band is studied.By analyzing the influence of gas target on infrared radiation transmission,a corresponding radiation spectrum mixing model is constructed.For the hyperspectral video data,in order to realize its effective expression and utilization,a tensor-based video data representation model is introduced.Then,the gas detection methods of infrared hyperspectral video are studied.Based on the radiation spectrum mixing model,the single-frame gas detection methods based on subspace model and spectral unmixing are studied respectively.In order to utilize the inter-frame correlation information of hyperspectral video data,it is considered as a fourth-order tensor that expands with time.A gas detection method based on cumulative tensor CP decomposition is proposed,which can effectively improve the detection performance of gas target in each frame,and provide key frame information for subsequent gas target tracking task at the same time.Finally,aiming at the dynamic tracking of gas target in video data,firstly we apply the classical robust principal component analysis to the processing of video data.Considering the problem that the assumed background is always fixed in the robust principal component analysis method,the gas tracking method based on incremental nonnegative matrix decomposition is studied,which realizes the dynamic update of background in the gas tracking process.In order to realize the comprehensive utilization of time-spatial-spectral information contained in hyperspectral video data,a gas target tracking method based on sequential tensor decomposition is proposed.By decomposing the third-order tensor corresponding to each frame,the tracking performance of gas target is effectively improved.