Research On Defect Detection Based On Image Processing For AFP

With the application of using automatic laying technology is more and more widely in aeronautics and astronautics,it has been paid much attention to how to guarantee the laying quality effectively in the process of automated laying.Aiming at the defects such as bubbles?wrinkles and FOD(Foreign object debris)arose in the automated laying process,based on the Image processing technology,this paper has carried on the related research on high qualitative image acquisition,pre-processing of prepreg image,image segmentation and defect recognition technology.The mainly works of this paper are as follows:(1)Based on the principle of light in this machine vision,the lighting scheme of ‘ring LED and dome lampshade' has been built.By establishing the attenuation model of the light,the illumination uniformity simulation has been analyzed on the platform of Matlab,and the best location parameters of light has gained for achieving high qualitative image acquisition of prepreg defects.(2)Considering the problem that the gray scale of prepreg image captured by camera is uneven due to the shake of placement,this paper has proposed a image preprocessing method combined gray compensation algorithm with median filter to compensate the image matrix by calculating the average gray curve of image.The experiment showed that this method is not only improving the uniformity of image,but also enhancing the image contrast.In addition,a rapid method for detection which were based on the complexity in RPVC has been put forward,meanwhile,it can enhance the detection efficiency of system.(3)Aiming at the problem that the defect image of prepreg can't be detected in RPVC method,based on the subtraction algorithm,this paper has introduced the concept of limiting error to point out the maximum error,and established standard background,then substracted the defect image of prepreg.At last,by comparing with the substracted results of defect image in Mass centroid algorithm and Otsu algorithm,proposed algorithm has good segment results and low time-consuming.(4)Considering complicated structure and random distribution of the prepreg image,in order to fully characterize the image defect information,this paper has extracted and filtered the gray-scale feature that could reflect the global characteristics of the defect image,chose the geometric features and moment invariants features that could reflect the geometric characteristics of the defect area in the prepreg images,and by using PCA method,extracted the redundant features,taking 10 effective features as the input of the defect recognition system.(5)In this paper,by introducing the neural network to classify the image of prepreg defects,the pre-processing,image segmentation and the classification of prepreg image has been realized.and the goal of the modular design in the prepreg defect system were achieved by using GUI of Matlab.