PCB welding spot defect detection method

Abstract

The invention provides a PCB welding spot defect detection method. Firstly, a welding spot image is preprocessed; shape characteristic parameters and texture characteristic parameters of the welding spots are extracted; the optimal RBF kernel function in SVM classification is adopted to construct the SVM classifier, the second SVM multi-classifier based on HOG features is utilized to obtain the final classification accuracy for false detection of welding spots, the classification accuracy of the method reaches 98.46% or above, and the method has certain superiority.

Description

Technical field [0001] The invention relates to the technical field of image processing, in particular to a method for detecting PCB solder joint defects. Background technique [0002] With the continuous development of the electronics industry, PCB boards are an important part of electronic products, affecting their overall quality and reliability. During the soldering process, PCB solder joints may have defects, causing various problems and losses in electronic products. Therefore, the defect detection of PCB solder joints has become a vital link in the PCB manufacturing process and an important means to ensure quality. [0003] The existing PCB solder joint detection methods include: by extracting the solder joint image features and using the Log-Gabor filter for detection, but the classification accuracy is low; by combining the Bayesian method with SVM, but the extracted features are too many, increasing Computational complexity; Probabilistic sorting algorithm detects solder...

AI Technical Summary

Problems solved by technology

[0003] Existing PCB solder joint detection methods include: extracting solder joint image features and using Log-Gabor filter for detection, but the classification accuracy is low; combining Bayesian method with SVM, but too many features are extracted, which increases the Computational complexity; the algorithm of probability sorting detects solder joints by analyzing color solder joint images, but this method has a single feature and uses fewer samples; the Hough transform algorithm is used to detect solder joint defects, but the detection type is single; the extraction of solder joints Three-dimensional features, using SFS method for three-dimensional reconstruction, using SVM for classification, the detection accuracy rate is 92.93%, but the classification accuracy rate needs to be improved; the combination of principal component analysis and extreme learning machine for solder joint defect detection can achieve a higher accuracy rate , but the number of experimental samples is small and there are defects; the automatic optical inspection system is complicated to use and is easily affected by factors such as the external environment and light; the BP neural network detection algorithm requires large samples and slow convergence speed; although the K-nearest neighbor method does not require Learning does not require pre-constructed models, but the classification effect on nonlinear samples is poor; the improved Adaboost algorithm can effectively reduce the false detection rate, but the classification accuracy needs to be improved

[0004] There are many shortcomings in the existing PCB solder joint detection methods, such as: fewer or more sample images, single detection features or types of solder joints, and low classification accuracy, etc.

Images