Visual Detection Technology Of Metal Surface Based On Image Fusion

With the development of science and technology and the improvement of social productivity,machine vision becomes more and more important in the field of industrial application.In the field of machine vision,metal is famous for its unique texture features and variable defect features.Conventional visual optical system and algorithm are difficult to be applied to metal surface defect detection.In this paper,three typical defect characteristics of metal surface are selected: scratch,scratch and crush.As the research object,a unique combined optical lighting system and image fusion scheme are adopted.The main work contents are as follows:1?Select the metal steel sheet with abrasions,scratches,compression and other defects;According to the field of vision,detection accuracy and defect characteristics,the hardware selection of machine vision system is completed by combining the knowledge of optics,camera and lens.Set up a combined optical lighting system,so that the industrial camera in turn to each metal steel plate for four times of image acquisition,the four images into a group,which are recorded as: up,down,left,right.2?Aiming at the defect characteristics of metals,this paper is a classifier based on ANN neural network.Firstly,the neural network model is built according to the collated data set,and then loss and optimization functions are performed to make the built neural network model more stable and more accurate in detection during operation.Finally,the network model was trained and tested to achieve accurate classification of scratch,scratch and crush defects.The traditional visual detection method is: first detection and then classification,but this paper adopts the reverse thinking: first classification and then detection,accurate classification and efficient and stable detection of metal defects.3?For the bruise defect image after classification: firstly,image fusion is carried out to concentrate the defects of four images on one image,and then the fused image is used as the new target image;Then fast Fourier transform and gaussian filter are used to perform convolution operation on the new target image,and then automatic threshold,morphology,feature extraction and other methods are used to achieve accurate detection of scratch defects.4?For the classified scratch defect image: firstly,image fusion is carried out to concentrate the defects of four images on one image,and then the fused image is used as the new target image.Then the new target image was edpreprocess,including automatic threshold and edge extraction.Finally,the edpreprocess image was extracted by feature extraction.The feature extraction was mainly based on morphological processing and texture features to achieve accurate detection of scratch defects.5?For classification of crushed after defect images: first,image fusion processing,this stepof image fusion method is implemented based on cphotometri stereo method,according to the design of visual optical system can define three array: sample data array,the output light source array + Z axis Angle,light source in the XY plane array according to the X axis deflection Angle,on the basis of the three arrays cphotometri stereo method processing.The normal vector of the image surface is obtained by the cphotometri stereo method,the normal vector is convolved with the derivative of the gaussian function,and then the convolution image is filtered.Finally,the dynamic threshold method can be used to detect the characteristics of the crush defect.6?The last chapter of this paper is the experimental results and analysis.In this chapter,the experimental hardware platform construction of the visual system,the simulation experiment analysis of the classifier and the simulation experiment analysis of the visual system are explained in detail.