Research On Visual Inspection Methods For Precision Parts Processin

Since the 21 st century,precision parts have been widely used in industries such as defense,medical equipment,aerospace,electronics,and household goods manufacturing.However,in the machining process of precision parts,surface defects are inevitably generated,which lead to a decrease in quality and accuracy,and further affect the quality,performance,and service life of the equipment.Therefore,effective detection of surface defects in parts is crucial.Machine vision-based surface defect detection in parts can be divided into two modules: image preprocessing and image recognition.The preprocessing module restores degraded images to feature-clear and easily recognizable images for high-speed and high-precision defect detection in the recognition module.This paper mainly studies the image restoration method under the regularization framework to achieve a balance between noise reduction and detail information preservation and the deep learning-based method for detecting surface defects on metal surfaces to solve the problem of missed or false detection of small targets,as follows:(1)Image edges are the most indicative feature in image analysis and understanding,but noise and blur during the imaging process can result in unclear defect edge features.Therefore,to address the issue of how to effectively preserve edge detail information during the restoration process of degraded images,a hybrid regularization image restoration method is proposed based on the combination of overlapping group sparse total variation and shearlet transform.Firstly,by using the overlapping group sparse total variation and shearlet transform to balance the edge detail protection ability and denoising ability,a hybrid regularization image restoration model is established to achieve a balance between denoising and detail protection.Secondly,the idea of momentum gradient is introduced into the alternating direction method to reduce the number of algorithm iterations and achieve fast convergence.Experimental results show that this method can preserve more detail information while ensuring image restoration effectiveness,at the same time,the number of iterations required is far less than half of the comparison method..(2)A generalized total variation regularization image restoration method based on derivative fidelity term(D-TGV)is proposed to improve the balance between noise removal and detail preservation.Firstly,the fidelity term adopts an equivalent fidelity model in the derivative space,which effectively utilizes the characterization of image gradient features in the differential space to preserve detail features.Secondly,the generalized total variation regularization term eliminates the staircase effect caused by traditional total variation.Finally,the alternating direction method with multipliers is used to decompose the original complex model into several simple sub-problems for fast restoration.Experiments on standard grayscale images show that this method has a good balance between detail preservation and denoising,and its accuracy is one or two points higher than that of the second one.Meanwhile,the number of iterations required is far less than half of that of the comparison method.In particular,when applied to surface images of steel plates,it has a good restoration effect for surface images under impurities and oxidation,making the defect features more obvious and easier to identify,and reducing false positives and false negatives.(3)A metal surface defect detection method based on an improved YOLOv5 s model is proposed to address the issues of missed detection and false detection of small targets in metal surface defect detection.Firstly,the size of the model parameters is compressed through lightweight network compression to reduce the computational cost of the algorithm.Secondly,a deeper level feature fusion module is added to solve the problem of difficulty in extracting features of small targets.Thirdly,a loss function related to the direction between the true box and the predicted box is considered to improve the detection accuracy and obtain a faster and more accurate metal surface defect detection effect.Experimental results show that the algorithm improves the detection accuracy by three points and eliminates the problem of missing detection.In particular,combined with D-TGV algorithm,the precision of small target defect detection is improved.