Image Distortion Correction And Temperature Error Analysis Based On Machine Learning

The cooling water discharged from the operation of the nuclear power plant raises the temperature of the nearby sea area,which causes irreversibly harm to the marine environment.The UAV carrying infrared thermal imager can effectively and accurately monitor the range of temperature rise near the nuclear power station,provide reliable data for marine environmental monitoring,and strengthen the monitoring and protection of the marine environment.However,due to the barrel distortion in the image obtained by the infrared imager,the actual imaging points caused by barrel distortion is to be offset,the accurate target temperature value cannot be obtained in subsequent image processing.In addition,there exists error between the temperature value and the actual value of the infrared monitoring,which requires certain errors and error analysis and accuracy improvement.In view of the above issues,this paper uses,the theory analysis simulation experiments and numerical comparison analysis methods,which based on field experiments to carry out the related contents of machine learning-based image distortion processing and SST monitoring error analysis research.The main work is as follows:(1)This paper analyzes the cause of distortion in infrared imager and the principle of barrel distortion caused by PI640 infrared imager.Combined with the advantages of neural network in image processing,the theory and advantages and disadvantages of LM-BP neural network and RBF neural network are discussed.(2)In addition,this paper uses LM-BP neural network and RBF neural network to correct the image,and correct the accuracy of image distortion by two neural networks through distortion correction error analysis.The error size shows that the accuracy of the RBF neural network correction image is higher than that of the LM-BP neural network,and is better than the template method.Finally,three kinds of correction methods are used to calibrate the infrared image collected on site,which proves that the LM-BP and RBF network is the best way to limit the error and can be applied in this project.(3)What' more,This paper uses UAV to conduct warm water drainage monitoring in the vicinity of Hongyanhe Nuclear Power Station,and the tools,processes and data preprocessing used in monitoring are briefly described.XGBoost has the advantages of high efficiency and precision in the prediction data,and considering the influence of air height,rollangle,pitch angle,air temperature,wind speed and air humidity on the temperature measurement accuracy of unmanned aerial vehicle,XGBoost is used to predict the temperature by python,and the prediction value that is close to the ideal value is obtained.The factors affecting the infrared temperature measurement error are analyzed one by one,including the attitude angle,air temperature,wind speed,air humidity and so on.It is proved that all kinds of factors have great influence on the accuracy of temperature measurement,which cannot be ignored.