Abstract
This paper provides an efficient procedure for the application of machine vision in solder defects detection, which has long been a problem in electronic industry. The major issues of this task are unstable images due to specular reflection of solder surfaces and large variations of solder shapes. Thus, this paper is composed of two steps. The first one is the illumination design, through the simulation of solder surfaces, suitable illumination method could be induced by observing the synthetic images, and finally the adoption of a single-layer circular illumination with incident angle about 25°is concluded as the best illumination for extracting characteristic features from the images. The second stage is the feature extraction from real solder images under the designed illumination, and the feature set proposed includes the partitioned regions and weighted areas.The major contributions of this work are: (1) The designed illumination can effectively and robustly highlight solder image features according to the solder types; furthermore, this simplified lighting can also help in speeding up lighting operation and/or image processing. (2) The proposed feature set and classification method can not only recognize typical solders precisely, but also identify the ambiguous solders sensitively. Even in the practical application involving ambiguous solders, the recognition rate of the proposed method remains as high as 95.16%. Based on these conclusions, this research has great potential in industrial applications.