微芯片焊点快速识别与定位方法

Rapid Recognition and Location Method for Solder Pads of Microchips

提出了一种快速、准确识别和定位高速焊线机微芯片图像焊点的新方法。首先使用聚类算法对自制微芯片图像训练集中标注的边界框进行分析,得到锚框尺寸;然后使用卷积神经网络对输入的微芯片图像数据集进行特征提取,对深层特征图进行上采样,与浅层特征图拼接在一起进行特征融合,用于预测焊点位置和类别;最后对多种尺寸图片进行训练,以缓解分辨率的突然切换对模型造成的影响,提高模型的适应能力。采用提出的焊点识别模型在高速焊线机上对不同形状焊点的发光二极管(LED)芯片进行识别,实验结果表明,该模型单张图片识别速度为5 ms,定位精度为99.67%,比传统模板匹配算法高。且该方法对微芯片形状、生产环境以及光照无一致性要求,更适合工业化生产需求。

A new method of fast and precise recognition and location for solder pads of microchip images of high-speed bonding machine was proposed.Firstly,the clustering algorithm was used to analyze the labeled boundary frame in the self-made microchip image training set,and the anchor frame size was acquired.Then the convolutional neural network was used to extract features of the input microchip image data set and the deep feature map was sampled up.The deep feature map was then spliced with the shallow feature map for feature fusion,which was used to predict the position and type of solder pads.Finally,images of various sizes were trained to alleviate the impact of sudden switch of the resolution on the model and improve the adaptability of the model.The proposed solder pad recognition model was used to identify light emission diode(LED)chips with different solder pad shape on high-speed wire bonding machines.The experimental results show that the recognition speed of a single image of the model is 5 ms and the positioning accuracy is 99.67%,which is higher than the conventional template matching algorithm.Moreover,the method has no consistency requirements for microchip shapes,production environment and lighting,which is more suitable for industrial manufacturing needs.