The Research Of High Precision Recognition Algorithm For Golden Ball Wire Bonder

The recognition of object is key technology for automatic gold wire bonder. Although manual gold wire bonder has been self-produced in our country, we realize automatic producing not yet, especially challenges still exist in the high speeding, high precision and stability. So far, many methods of image recognition have been put forward, but the high precision recognition algorithm in the image processing system of gold wire bonder is still need further research, therefor it's extremely important to propose a recognition algorithm which can extract invariant features of image when it is translated, rotated and scaled.This paper proposed new algorithms about preprocessing, moment feature extraction, then the invariant moments are taken as input information of wavelet neural network to realize classification. Because the limitation of production technology and image environment, it is still difficult to many recogniton algorithm to realize high precition. Aiming at resolve the problem in preprocessing, a new edge detection algorithm based on improved SUSAN algorithm is proposed in this paper, and experiments demonstrated the advantages of the new algorithm in robustness and high speeding.In this paper, comparisons are also made among Hu momoents, structure moments which can decribe objects'globle feature and wavelet moments that can extract both global and local feature. Structure moments can recognize complicated images, and wavelet analysis can be analized in local time and frequency domain, to combine the advantages of structure moment and wavelet analysis, the method of wavelet-structure moments are put forward. Through make experiments by extracting the Hu moments, structure moments and wavelet-structure moments invariants when the chip images are translated, rotated and scaled, this paper has demonstrated that wavelet-structure moments has higher recognition precition than other methods, this is of great significance for the image process system of gold wire bonder.Based on the advantages of wavelet-structure moments, it's better to combine wavelet-structure moments and wavet neural network to recognize objects, and experiment showed that wavet neural network is better than BP neural network in acturation of classfication.