Research And Experimental Analysis Of Image Filtering Denoising And Edge Detection Technology

As a non-contact inspection technology,visual inspection is a comprehensive cross-cutting technology integrating mechanical engineering,computer technology,image processing,automatic control and other disciplines.It has multiple functions and a wide range of applications.In the field of industrial production,as the focus of intelligent manufacturing,visual inspection technology,driven by the Made in China2025 strategy and the huge manufacturing market demand,has ushered in an unprecedented golden development period.Visual inspection technology has great practical and economic value for realizing intelligent inspection of workpieces,so it is necessary to carry out research on related technologies in visual inspection.Image filtering,denoising and edge detection are important steps of visual inspection,which affect the accuracy of workpiece size measurement.The related technologies of traditional visual inspection have the following problems: traditional filtering and denoising algorithms can only effectively filter out salt and pepper noise within a certain concentration range.When the noise concentration exceeds a certain value,the denoising effect is not ideal;traditional edge detection operators are easy affected by external environmental factors and the surface of the workpiece.The robustness is poor.The detected false edges affect the measurement accuracy of the workpiece.For this reason,based on the analysis of traditional image filtering algorithms and edge detection operators,this paper improves the image filtering and denoising algorithms,proposes an edge detection model based on convolutional neural networks,and verifies the effectiveness of the algorithm through visual measurement experiments.The specific research process is as follows:(1)The thesis analyzes the traditional image filtering algorithm and the evaluation method of denoising effect.For the salt and pepper noise in the image,a hybrid algorithm based on median,mean and interpolation fusion is proposed.In this paper,Matlab is used to simulate the flange image with a salt and pepper noise concentration of 0.1?0.9 using four different filtering and denoising algorithms.This paper combines subjective and objective evaluation methods to evaluate and analyze the denoising effects of four different filtering algorithms.(2)The traditional edge detection operator is studied,and five common edge detection operators are used to simulate and analyze the flange image.The traditional edge detection operator has the problem of poor robustness.The VGG16-based edge detection model has long training time,high training cost,and large computer memory usage.Based on the convolutional neural network,a small and portable edge detection model is built.After training the model 1500 times with the BSD500 data set,the universal pre-training model is obtained.For specific workpieces,Labelme software is used to make corresponding edge labels.The pre-training model is fed for retraining to achieve fine-tuning of model parameters and obtain an edge detection model based on a specific workpiece.(3)In order to realize the function of workpiece edge refinement and threshold adaptive adjustment,this paper studies the non-maximum suppression and dual threshold processing in Canny operator.In order to reduce errors caused by manual threshold selection,the OTSU algorithm is introduced into dual-threshold processing to adaptively set high and low thresholds in the image.Matlab is used to perform nonmaximum value suppression and double threshold processing on the flange edge detected by the model,and adaptively solve the high and low thresholds of the image.(4)In order to verify the effectiveness of the algorithm in this paper,this paper uses the flange as the measurement workpiece to build a vision measurement platform.Combining radial distortion and tangential distortion,this paper studies the principle of camera imaging and completes the camera calibration experiment.This paper compares the measurement data obtained by classic image processing methods and analyzes the measurement results of the two algorithms.According to the measurement error produced in the experiment,the source of the error is analyzed.