Surface Defects Inspection System Based on Machine Vision

Abstract

A machine vision based tinplate surface inspection system was developed. The system was composed of two parallel line scan CCD cameras, a special designed wide field illumination, which can overcome the vibration of tinplate, and a software based on SOM (Self-Organizing Feature Map) neural network. The images of tinplate were captured by cameras. All kinds of defects candidates such as pinholes, scallops, dust and scratches were found out, and their features can be extracted and selected from images. These candidates were distinguished by the SOM neural network to find out real defects. The inspection speed reached up to 1.4 m/s, and the resolution was 0.1 mm, and recognition rate was 95.45%.