Infrared Thermal Wave Image Deblurring Based On Depth Residual Network

Infrared thermal wave detection is a specific application of non-destructive testing technology.It applies controllable thermal excitation to the object,and reflects the influence of the physical structure characteristics and boundary conditions of different media materials on the infrared thermal wave transmission in a certain way on the temperature field change of the media surface.The infrared thermal imager is used to continuously observe and record the temperature field change of the object surface and the realization of.This kind of detection method will not damage the internal structure and material characteristics of the object itself,but through the difference of defect location and structure,resulting in the change of surface temperature.In the process of infrared thermal wave imaging,because of the lateral thermal diffusion,the influence of the object thermal radiation energy on the infrared band,the inconsistency of the imaging distance and other problems have a great impact on the infrared thermal wave imaging,which leads to a certain blur of the imaging results and a considerable detection error.It is of great significance to propose an effective infrared image deblurring method for such problems.Convolutional neural network has developed rapidly in recent years,and has achieved excellent performance in the application field of image processing.However,the related research on infrared image deblurring is relatively scarce.Therefore,the research focus of this paper is to design an effective deblurring algorithm model for infrared thermal wave blurred image:(1)This paper studies the characteristics of convolutional neural network in deep learning,including the working principle,design concept and other related knowledge of convolutional neural network,studies the specific network models applied in the field of de fuzzy at present,and provides a theoretical basis for the follow-up research by analyzing the advantages and disadvantages of different models.(2)The relevant principle of infrared thermal wave imaging is studied,and the infrared thermal wave imaging system is used to collect the image of artificial defect samples,obtain the original infrared blurred image,and make the infrared image data set by manually calibrating the defect position.Through the corresponding data enhancement operation on the collected data,the amount of data can meet the training requirements of convolutional neural network.(3)According to the characteristics of infrared image data,combined with the characteristics of convolutional neural network,a deblurring algorithm model based on depth residual network is designed.The network model is divided into two parts,fuzzy kernel distribution estimation network and deblurring network.Through the combination of the two,the deblurring operation of infrared image can be effectively realized.This paper also improves the loss function used in the network training process from the perspective of image gradient,which can effectively improve the robustness and accuracy of the network.