改进ViBe算法的IC焊点自动光学检测新方法

54 浏览量更新于2024-08-26 收藏 1.9MB PDF 举报

"该文章是关于一种新的基于改进视觉背景提取算法的自动光学检测系统IC焊点联合检测方法的研究论文，发表在IEEE Transactions on Components, Packaging and Manufacturing Technology期刊上。作者包括Nian Cai, Jianfa Lin, Qian Ye, Han Wang, Shaowei Weng和Bingo Wing-Kuen Ling。" 在自动光学检测（AOI）领域，集成电路（IC）焊点的缺陷识别一直是一项挑战性任务。受到计算机视觉中的视觉背景提取（ViBe）算法的启发，作者提出了一种创新的IC焊点检测方法，该方法基于改进的ViBe算法。据作者所知，这是首次将缺陷检测问题视为对象检测问题来处理。首先，他们利用ViBe模型更新方案构建了一个焊点模型。ViBe算法是一种常用于视频分析中的背景建模方法，能有效地去除静态背景，突出动态物体。在本文中，该算法被改进以适应IC焊点的特定环境和特征。接下来，通过将实际的焊点图像与经过充分训练的焊点模型进行比较，可以识别出可能存在的缺陷。这种方法能够捕捉到焊点图像中的异常变化，从而实现高精度的缺陷检测。为了进一步提高检测效果，作者引入了频率图方法。频率图是一种图像处理技术，它能将图像转换为频域表示，以便于分析图像的高频成分，这些高频成分通常对应于图像中的细节和边缘，对于缺陷检测尤其有用。此外，他们还定义了一个名为“缺陷程度”的度量标准，用于评估和量化检测到的焊点缺陷的严重程度。这个度量有助于优先处理严重缺陷，并对检测结果进行量化评估，提高了整个检测系统的可靠性和实用性。该研究提供了一种新颖的、基于改进视觉背景提取的IC焊点自动检测策略，有望提升AOI系统的检测效率和准确性，对于集成电路制造的质量控制具有重要意义。

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

CAI et al.: NEW IC SOLDER JOINT INSPECTION METHOD FOR AN AOI SYSTEM 3

Fig. 1. AOI image acquisition system.

(red, green, and blue) hemispherical light-emitting diode array

illumination.

The red, green, and blue lights irradiate the solder joint

surfaces and are reﬂected onto the camera. Since the ﬂats

and the slopes reﬂect different light components onto the

camera, the 3-D shape information of the solder joints may

be extracted from this acquired 2-D image. For the slopes of

the IC solder joints, a large amount of red lights and green

lights are reﬂected away, while the blue lights are reﬂected

toward the camera, which makes the solder joints appear to

be blue in the images. On the contrary, the ﬂats of the IC pins

and the pads reﬂect the red lights toward the camera, while

green and blue lights are reﬂected elsewhere. Such ﬂat regions

appear to be red in the images. An example of an obtained

IC solder joint image containing 16 qualiﬁed solder joints and

an unqualiﬁed one (the rightmost one) is shown in Fig. 2(a).

For a clear description, we zoom in on the third one on the

left in Fig. 2(a), which is marked by a white rectangle. It is

rotated counterclockwise by 90°, which is shown in Fig. 2(b).

In Fig. 2(b), Region A is located on the ﬂat of the IC pin,

while Region B is located on the solder joint of this pin. That

is, Region B is located between the pin and the pad.

III. A

LGORITHM PRINCIPLE

In the ViBe algorithm for moving object detection, the pre-

vious frames of a video are input in sequence so as to produce

the templates, which approximately represent the model of real

background. The current frame is input and compared with the

trained templates. The parts of the frame failing to match the

templates are considered to be the foreground pixels.

A. Relationship Between Defect Inspection and ViBe

For moving object detection, each frame of a video contains

the background, the foreground, and some noise. To detect the

foreground, we should ﬁrst build a background model. Then,

the foreground is detected by comparing the pixels of the input

frame with those of the model. It means that the input pixels

Fig. 2. IC solder joint image obtained by an AOI system. (a) Image with

17 IC solder joints. (b) Zoomed-in view of IC solder joint image.

mismatching the pixels in the model are considered as the

foreground pixels. This is similar to the problem of defect

inspection in an AOI system. For automatic inspection of

IC solder joints, an unqualiﬁed solder joint shows an abnormal

appearance at the speciﬁc location of the image compared with

a qualiﬁed one. That is, each unqualiﬁed solder joint image

contains the normal region, the abnormal region, and some

noise. Thus, detecting the abnormal region in a solder joint

image can be viewed as detecting the foreground in a frame of

a video. Therefore, we consider the unordered IC solder joint

images as the frames of an inspection sequence. Like moving

object detection, we can ﬁrst use the inspection sequence to

establish a model representing a qualiﬁed solder joint. Then,

we detect the abnormal region by comparing the pixels of

the input solder joint image with the pixels of the established

model. In this paper, we use the ViBe background modeling

method [21] to train the qualiﬁed solder joint model. Since

our proposed inspection method statistically calculates the real

distribution of the hue channel values of a pixel using these

unordered images, we need not sort the frames of the sequence.

Since three channels of the hue-saturation-value model are

independent with each other, it is widely used in the ﬁelds of

image processing and computer vision. In addition, the hue

channel contains the major information for colors perception.

Thus, we use the hue channel of the solder joint image to

establish the qualiﬁed solder joint model. Here, the range of

the hue channel values is [0, 180).

For moving object detection, the background of the outdoor

scene often contains many nonstatic objects, such as tree

branches in the wind and ripples on the surface of the water.

The hue channel values of the pixels in these nonstatic

objects vary signiﬁcantly. The ViBe algorithm models the

nonstatic objects and static scene excellently through several

剩余11页未读，继续阅读

weixin_38677227

粉丝: 4
资源: 929

改进ViBe算法的IC焊点自动光学检测新方法

基于机器视觉的SMT焊点自动光学检测系统研究

焊点缺陷检测必看

机器视觉SMT焊点自动光学检测技术研究与应用

自动光学检测仪

基于计算机视觉的PCB板焊点缺陷检测及分类.docx

基于计算机视觉的焊点缺陷检测系统的设计.pdf

电子政务-基于机器视觉的印制电路板自动检测系统.zip

论文研究-基于神经网络的PCBA焊点缺陷检测新方法 .pdf

AOI的光学系统(自动光学检测)

自动光学检测原理

最新资源