关键词: 极限学习机, 焊点, 分类检测, 图像处理, 神经网络

Abstract: Solder joint processing has a direct influence on the reliability of electronics. Solder joint inspection is particularly important to improve the quality of products. In this paper, the principal component analysis and extreme learning machine are adopted for quality inspection of solder joint. First of all, this paper makes image pre-processing with median filter and watershed algorithm, and reduces the dimension with the principal component analysis after obtaining the contour and area. Then, the paper executes a classification by extreme learning machine with 200 neurons in hidden layer and response function-‘sigmoid’. The result shows that the extreme learning machine can make the precise classification of solder joints. Compared with the support vector machine, k-nearest neighbor and convolutional neural networks, it has a higher detection accuracy and detection time is shorter.

Key words: extreme learning machine, solder joint, classification detection, image processing, neural network