Image Processing Research On The Visual Location System Of LED Automatic Die Bonder

Along with the conversation and diffusion of environment friendly lighting energy, reliable and energy-saving LED illumination replace the conventional lighting gradually and become major development-direction of future lighting. So the meaning of researching how to improve the performance of LED processing equipment becomes extremely significant. LED automatic die bonder produces the LED lighting cores. Its image processing system gets the wafers'locations and their classification, the control system leads the machine arm to pick them up. As the system's key link in the functional implements of LED automatic die bonder, the image processing affects the entire recognition results of system, and determines the speed and accuracy of processing.The main research of this paper is about the image processing technology system of LED automatic die bonder. The key task is determining the classification target between the qualified wafers and unqualified ones by analyzing features of image. The first step is pre-processing the image obtained so as to remove interference. This process includes filtering, morphological processing and pyramid lessen and magnify.Then extract the edge information of image and confirm the location of LED wafer roughly in the small image and then return the big image and get the accurate location. The last job is distinguishing the qualified wafers, the ink wafers, wafers lacking of edges or corners and the too-close wafers by different judgment criterion.We use multiresolution recognition in order to meet the requirement of wafer identification'timelessness and small size wafer's matching precision, that is, rough-fine identification method. In order to refrain from artificial work on setting the threshold value, an auto-adaptive threshold Canny algorithm based on Otsu was proposed. This algorithm can not only avoid artificial threshold value setting, but also can deduce all kinds of noise interference caused by complicated situation and enhance the accommodation and robustness. An edge model matching was proposed based on morphological operation. In the edge image, a wafer was found when the max match result was got. For many wafers in the image, a multi-objects searching method was put forward. After a max-overall value was found, the relevant region was set as the background and then went on finding the next max value location. According to the classification of different target classification features, the best classification standards method is selected by experimental comparison.A software system was developed based on OpenCV. The algorithms can not only be adapt to the LED wafer recognition, realize the position, classification and identification of wafers but also realize the experiment which can meet different image operation.