| The thesis aims at image processing and identification problems of strip steel surfacedefects, for steps of technological processes on image processing, due to algorithm theory andexperiments, obtains preferably image processing algorithms on strip steel surface defects,furthermore, build BP Neural Networks classifier in effectively to classify strip steel surfacedefects. Tests are doing in status of off-line, and it’s helpful to realize on-line detection.Research results as follows:(1) On the basis of the noise type of strip steel images, make use of frequently-used imagesmoothing algorithms, filtering image noise in time.(2) Aim at strip steel defect complex causes, different kinds and shapes, build five kinds ofstructuring elements which are similar with the edge of strip steel defects, using correctionalnoise immunity of dilation and erotion operation, which belongs to mathematicalmorphology method, detect the edge of the strip steel defects are more continuous andintegrated compared to traditional edge detection algorithm, especially testing out thecontrast is not obvious as short scratch of strip surface etc defects and irregular edge shape,grey value vary considerably edge as such as phosphating spot etc defects, detect the edgeare continuous and integrated. However, the results of edge detect base on traditional edgealgorithm are not continuous, especially to short scratch of strip surface etc defects, theresult of edge detect are not well, even it can’t detected.(3) Making use of the characteristics of defect image, extract defect area’s shape, grayscale,texture features, set up defects classifier based on the BP neural network, training networkthrough training sample set, at last test BP neural networks effectiveness.Although the thesis achieved certain research results, aim at strip steel on-line test in time,there are some gap. So, at the end of paper, we summarize the paper’s problem and insufficient,offer suggestions for next step. |
PDF Full Text Request