Research On Infrared And Visible Image Fusion Algorithm For Power Equipment Fault Detection

In power equipment fault detection,multi-source images can reflect the characteristics of the equipment from multiple angles.Infrared images can capture the temperature changes of the equipment more accurately and are not restricted by environmental conditions.Visible light images can clearly show the details of the device’s appearance.The fused image has the local temperature information of the infrared image power equipment and the shape profile information of the power equipment,so it can be applied to the fault detection.In order to meet the needs of practical applications and realize the fault location of power equipment,this thesis studies the fusion algorithm of infrared and visible image for the status detection of power fault,and the specific work is as follows:(1)Aiming at the problem that traditional checkerboard calibration plate can’t make the clear image under infrared camera and can’t realize infrared visible binocular camera calibration,a kind of infrared and visible binocular camera calibration plate is designed and produced.Infrared calibration plate has the advantages of clear imaging and simple structure.Finally,the visible infrared binocular camera is used to collect the images from the calibration plate,and the Matlab Calib toolbox is used to detect the corners and characteristics of the images,in order to achieve the calibration of the visible infrared binocular camera.After reprojection error analysis,the calibration plate has high calibration accuracy and can be used to calibrate the infrared visible binocular camera.(2)In order to solve the problem of the poor recording effect due to differences in resolution and imaging between infrared and visible images,a method of recording by infrared visible binocular camera based on Harris characteristic points was proposed.First,the collected infrared image and the visible image were detected by Harris Angle points,then a number of pairs of characteristic points were selected to calculate the projection transformation matrix.The projection transformation model was used to describe the geometric transformation of the image to be recorded.After that,the recording error of horizontal and vertical pixels was checked to verify the recording effect.It is proven that the recording error is low,and the recording method has the advantages of more accurate recording and less calculation.(3)In order to solve the problems of insufficient extraction of thermal radiation information in infrared images by SCNN algorithm and loss of partial texture details in fused images,an infrared visible image fusion method based on SCnn-SE Net was proposed.This method is improved based on Liu’s SCNN model.By adding an SE Net module and a maximum pooling layer to each branch of the original SCNN network model,SE Net can better obtain the features in different channels of the feature map and realize the effective extraction of the original image detail features.A Loss function composed of Softmax and Cross Entropy Loss is used to assign weights to the relative sharpness of each pixel in the infrared and visible images.Finally,subjective evaluation and objective evaluation with other algorithms are carried out on the TNO data set and the infrared visible data set of substation power equipment.The results show that the fusion method of infrared and visible image based on SCNN-SE Net has a good performance in the two evaluation results.(4)Aiming at the problems of poor dependence between pixels and the need to design complex fusion rules,a fusion method of infrared and visible images based on SCNN-Vi T is proposed.On the basis of the above method,the original SCNN network model is adjusted to consist of five convolution layers up and down,which is convenient to fully extract structural information from the training image.At the same time,Vision Transformer(Vi T)module is introduced.The Vi T module is mainly composed of channel vision converter and spatial vision converter.Channel vision converter is used to preserve channel dimensions and calculate the relationship between different channels,and spatial vision converter is used to enhance depth characteristics.The addition of Vi T module enables the network to effectively extract the original information and global information of the image,and avoids the problems of small receptive field of the feature map and poor dependence between image pixels in the traditional SCNN network.At the same time,the SCNN-Vi T network model is end-toend,which solves the problem that the SCNN image fusion method needs to choose the decomposition or fusion strategy manually.The SCNN-Vi T algorithm is compared with other algorithms on the TNO data set and substation infrared visible data set.The experimental results show that the image fusion algorithm based on SCNN-Vi T can obtain better fusion results,which is convenient for related personnel to carry out fault detection of power equipment.