Abstract: Aiming at the image characteristics of solder joints in SMT（Surface Mount Technology） production line, a solder joint defects recognition algorithm is proposed. A series of image preprocessing methods（including median filtering, iterative threshold method and Sobel operator etc.）are adopted to suppress noise interference and enhance image contrast, in order to extract better image characteristics. Finally, RBF neural network is applied to identify four different solder joint defects. Simulation results show that RBF neural network can overcome some disadvantages of BP neural network in training process such as initial dependence and possible local convergence, and possess faster operation speed and better detection results.

Key words: Printed Circuit Board（PCB）, solder joint defects identification；RBF neural network

摘要： 针对生产线上的SMT（表面贴装技术）焊点图像的特点，研究基于图像处理的焊点缺陷识别算法，采用中值滤波、迭代阈值法、Sobel算子等一系列的图像预处理方法，有效抑制了噪声干扰，提高了图像的对比度，提取出较好的图像特征。采用径向基函数（RBF）神经网络对四种焊点缺陷进行识别。仿真结果表明，RBF神经网络很好地克服BP神经网络训练过程收敛依赖于初值和可能出现局部收敛的缺陷，具有较快的运算速度和较好的检测结果，基于图像处理的焊点识别方法是有效的。

关键词: 印刷电路板, 焊点缺陷识别, RBF神经网络