Research On Image Recognition Technology And Its Application On LED Wire Bonders

LED(Light Emitting Diode) lights have advantages of low energy consumption,high conversion efficiency,little pollution and long service life, so almost all countries in the world are strongly popularizing that using LED illumination replaces traditional illumination. An automatic LED wire bonder is a kind of LED packaging equipment, and it is mainly used to finish bonding process when producing LED. It is very expensive to import LED wire bonders, and price seriously influences rapid development of semiconductor industry; therefore the government widely sponsors to research and develop semiconductor packaging equipments which have own intellectual property. Because image recognition is the core technology of software system in LED wire bonders, it has vital effect on machine performance.Firstly, the paper analyzes characteristic of capturing images and image recognition technology about automatic LED wire bonders; secondly it builds machine vision system which is suitable to industrial production for the machine and designs images preprocessing algorithm and image recognition algorithm; finally the thesis develops image recognition module and debugs by testing. The dissertation completes four major tasks as follows:(1) Building machine vision systemIllumination is very important for capturing images. Selecting an appropriate illumination directly affects images' quality. According to experiments, the system uses half of circular LED blue light whose wavelength is from 430nm to 480 nm and makes circular center coincide with lens center so that it can keeps capturing images angle best. At the same time, the system chooses proper enlargement of the multiples lens,CCD(Charge-coupled Device) and image card to make sure to gain best images.(2) Image preprocessingBecause of some noise from internal and external system when producing and some poor quality chips, images are not very good and objects are not obvious. So images should be processed before matching for aim identifying easily. Gauss filter linear transformations for the gray value and homomorphic filter usually apply for preprocessing. Images preprocessing takes different measures according to different chips in order to get the best effect.(3) Principal research and application about difference with much value match algorithmFirst of all, the thesis does some research about ABS(Absolute Balance Search) algorithm,matrix match algorithm,SSDA(Sequential Similarity Detection Algorithm) and NC(Normalized Correlation), and balances their advantages and disadvantages. On the basis, it proposes difference with much value and image integral gray match algorithm and then does some elaboration about algorithm derivation and algorithm programming. At last, the paper does some work about images search algorithms,programming and testing.(4) Structuring experiment platform and developing images recognition softwareAccording to LED wire bonders' characters, the paper builds machine vision hardware system and develops software platform. Basing mechanical module and circuit control module, then some testing and comparing experiments are done, later the thesis verifies accuracy,speed and algorithms stability and transplants program to LED wire bonders system.