| 摘要 |
PURPOSE: To detect a surface defect of a steel plate highly accurately by extracting multi-dimensional characteristic amounts such as a width of the surface defect, an area of the defective part, a lengthwise/breadthwise ratio, a maximum reflection intensity, etc., from a still image, thereby judging the defect. CONSTITUTION: A reflecting light of a light projected onto a steel plate 1 from a light source 13 such as a halogen lamp or the like is caught as a regular reflection light and an irregular reflection light. The image data is processed by averaging and averaging in XY directions, and also treated to obtain an average of a movement in a widthwise direction, and a threshold value, following a change of the surface of the steel plate in a longitudinal direction, whereby as many as about 40 kinds of multi-dimensional characteristic amounts are extracted. Among the characteristic amounts, particularly, lengths in XY directions, an area, a lengthwise/breadthwise ratio, a moment, the center of gravity and the like as size data, and a maximum intensity, a minimum intensity and an average intensity as reflection intensity data are extracted. Then, flaws are classified. A characteristic amount per flaw is extracted, thereby to judge a defect. An alarm is generated and an image of the defect is monitored. A classification accuracy for flaws is thus improved. |
