Research On Image Processing Methods For PCB Defect Detectio

Printed circuit board(PCB)industry has become an important pillar of global economic development and social core technology.With the maturity and rapid development of integrated circuit packaging technology,PCB wiring becomes increasingly crowded and complex,which increases the possibility of defects in PCB production process,and also brings difficulties to PCB defect detection.Machine vision-based detection method plays a very important role in automatic defect detection of PCB because of its outstanding advantages such as fast speed,non-contact and so on.However,degradation factors such as noise,blur and distortion inevitably exist in the process of PCB image acquisition,which makes it difficult to effectively identify the tiny defective features of PCB,leading to miss detection and false detection.Therefore,this paper conducted in-depth research on image restoration and detection methods for PCB defect detection,including the following main research work:(1)In order to solve the problem that the observation matrix is irreversible in the traditional total variation(TV)regularization model and the single TV regularization term cannot guarantee the balance between denoising and detail-maintaining,a two-step double regularization image restoration method based on pseudo-inverse transformation is proposed in this paper.Firstly,by introducing pseudo-inverse,the pseudo-inverse fidelity term is constructed,which eliminates the effect of irreversibility of observation matrix on restoration effect.Secondly,the hybrid double regularization term combining TV regularization term and shear transformation is used to achieve the balance between denoising and detail-maintaining,enhancing the restoration effect.Finally,the idea of two-step iteration is introduced to solve the double regularization image restoration model based on pseudo-inverse fidelity.Experiments show that this method can effectively maintain details while providing good denoising effect,providing feature-complete and clear images for subsequent PCB defect detection.(2)To improve the speed of the two-step iterative solution,an image restoration algorithm based on splitting solution and adaptive threshold is proposed in this paper.Based on the Split Bregman framework,the double regularization image restoration model based on pseudo-inverse transformation is decomposed into multiple sub-problems for iterative solution,and the adaptive threshold function is used to replace the soft threshold function during the iterative solution process,resulting in faster computation and better restoration effect.Experimental results show that this method achieves a faster calculation speed while ensuring the restoration effect,which is beneficial to meet the speed requirements of industrial applications.(3)Aiming at the problem that the small size of PCB defects can easily lead to false detections and missed detections,an improved YOLOv3-based defect detection method is proposed in the paper.Firstly,the spatial pyramid pooling layer(SPP)module is introduced,and a multi-scale fused single feature layer structure is applied to improve the YOLOv3 network framework,making it more suitable for detecting small targets.Secondly,the clustering method that combines K-means++ and genetic algorithm(GA)is used to cluster some priori frames,and the complete intersection over union(CIo U)is regarded as the loss function to improve the accuracy of the target detection algorithm.Experimental results demonstrate the effectiveness of the proposed method in detecting small defects of PCB.