Extracting capacitor parts from an electronic circuit board by image processing using support vector machine assuming recycling task

In the task of recycling wasted televisions, extraction of harmful capacitor parts from an electronic circuit board is inevitable. In order to robotize this task, it is necessary to recognize these parts by image processing. The problem of this task is that the size and the position of these parts are different from one board to another according to the product company, the age of product, the kind of product, etc. Moreover, stained parts are included, and it is difficult to keep a lighting condition in a recycling factory. To clear these problems, a two-stage recognition method is proposed. In the first stage, circular edge images are selected as the candidates for capacitor parts by a template matching procedure, which is speeded up by interlacing and binarizing the image. In the second stage, the candidates are classified further into two categories of capacitor parts and others by Support Vector Machine. Sample images are prepared, for which the application of the usual template matching using a normalized cross-correlation is difficult. The recognition was successfully performed for these images, i.e., the false rejection rate (FRR) was 0% for the first stage, and the false acceptance rate (FAR) was 0% for the second stage. The time needed for dealing with a 640 × 480 pixels image was approximately 7 s.