The Research Of The Key AOI Inspection Technologies For The Keyboard Defective Characters

Character detection is a very important step in the whole keyboard production process. Ifthe characters are completely detected by labor, not only the detection efficiency becomes verylow, but accuracy is not high enough, and it doesn’t meet modern production requirements. Withthe development and popularization of Automatic Optical Inspection (AOI) technology, as anuseful method to identify the defective characters, AOI is capable of avoiding the disadvantagesof inspecting manually. But because of the characteristics of the keyboard itself, such as toomany keys and the size of the keyboard characters, there are also some differences betweenkeyboard characters and other products in terms of AOI technology. Therefore, this paperconcentrates on the defect characters on the keyboard, and makes a research on the keytechnologies the inspection process involves. Besides, we develop an appropriate detectionsystem on the basis of the technologies to realize the function of automatic inspection.We divide the keyboard inspection technology into five parts in this paper, includingpreprocessing, obtaining region of interest, feature extraction, SVM classification and recheck.In the preprocessing period, we improve the iterative method. As a result, we get a better binarysegmentation threshold value after calculation. This paper also puts forward a novel method toget the sorted keyboard character areas, which is also called region of interest (ROI). It is acombination of mathematical morphology, boundary tracking technology and sorting andmerging algorithms based on the key arrangement characteristics, which is designed andpresented only in this paper. And the obtained ROIs meet the requirement that only one ROI canbe created in one key. In the feature extraction period, we extract8features associated with thekeyboard characters, and they embody the changes between normal characters and defectivecharacters by the variation of the feature values. Besides, support vector machine is successfullyapplied to identify the character defects in the paper, and some improvements have been made.First, because there may be over-fitting in original SVM, we design a feature selection algorithmbased on a distinctive parameter, which can reduce the possibility of over-fitting. Second, thispaper presents a method combining fuzzy membership to improve the support vector machine’sclassification performance. We find a relationship between the classification membership anddecision values, so the method solves the classification problem when the decision values are inthe gray zone. Finally, recheck again based on the classification work achieved above, and thedetection accuracy is further improved, which makes the normal character among the defectivecharacters be classified correctly.A keyboard AOI inspection software is developed after a great amount of test by makingfull use of the technologies and sample images. And the results show that the software’s worstclassification accuracy can reach92.8%, therefore, the validity and usefulness of the methodsand technologies presented in this paper is fully verified.