Speckle Image Recognition And Location Based On Faster R-CNN

Stealth capability is a key indicator to measure the combat radius and survivability of modern fighters,and radar absorbing materials play a key role in improving the stealth capability of modern fighters.However,these absorbing coating materials are usually composed of a multilayer structure,and defects may occur in the production process,such as: debonding,wrinkles,cracks,impact damage,etc.,which will reduce the mechanical properties of radar absorbing materials,or even limit Its ability to absorb radar waves has serious consequences.Therefore,it is of great significance to study large-scale,highprecision,and intelligent defect recognition and location technology.The internal debonding defects of radar absorbing coating materials are small in area and extremely concealed.The detection methods such as ultrasonic measurement,ray detection and infrared thermal imaging can not meet the detection requirements,but the laser shearing speckle interferometry technology has the advantages of non-destructive,large-scale,high-precision,etc.,and has become popular in the field of non-destructive testing.This detection technology applies heat loading to the wave-absorbing coating material containing defects to deform the surface,and then collects the two speckle interference images before and after the deformation and makes the difference to obtain an image with interference fringes.Using deep learning-based object detection algorithm can improve the recognition and positioning accuracy of speckle images,so as to achieve a large-scale,high-precision,and intelligent detection effect to replace traditional manual defect recognition.The main work of this paper is as follows:1.The principle of laser shearing speckle interferometry to detect defects is deeply studied,and the mathematical model of shearing speckle interferometry for detecting debonding defects of absorbing coating materials is constructed.2.Design a defect detection system based on Michelson laser shear speckle interferometry technology,optimizing the external thermal loading technology,and realizing the thermal loading of large-area,complex and non-uniform structures.Design the host computer software to control the high-speed camera to collect the speckle interference image in real time,and control the phase shift controller to introduce the 4+4 phase shift algorithm to extract the phase image of the speckle interference image,and realize the recognition and location of the speckle image defect.3.Analyze the noise characteristics of the speckle image,and use the sparse threedimensional transform domain collaborative filtering and denoising algorithm(BM3D)to filter the speckle image.The algorithm have very good effect of filtering out random background noise and retain the texture contours of interference fringes.The speckle image is enhanced through operations such as flip and rotation,random shift,and grayscale transformation,and the speckle image data set is made with the annotation tool.4.According to the characteristics of speckle images,the Faster R-CNN object detection algorithm is improved:(1)A dual-path residual network is proposed to reduce the interference of background noise in the speckle interference image on the defect characteristics;(2)Introduce Context Ro I structure and cascade detection mechanism to improve the global positioning accuracy of defects;(3)A feature pyramid network based on the self-attention mechanism is proposed to adaptively weight multi-scale features to boost the detection accuracy of smallarea defects.