The Research On CR And Image Process In Nuclear Power Station

This paper studied the blackness under different scanning intensity and found out theoptimal blackness for human eyes and the best condition to form an image. Then we testedsome denoising methods to get a more clear image, after that we extracted the edge of flaws.At last, we extract the character of flaws and use neural network to classify different flaws.Automatic identification can reduce the effect of man and takes less time, when acontroversial image appears we can resort to expert by internet.From the result of the research on blackness, we found that the blackness increase withthe voltage and the scanning intensity. The blackness of the base metal was higher than that ofweld. The curve of blackness fluctuated because the uncertainty of the PSL in the IP plate.In this paper, images were denoised. By testing different denoising methods we find outthat domain filter can filter dust noise better and median filter can deal pepper and salt noisebetter. And the size of the filter also affect the image, images were becoming clear as the sizeof the filter increased. Wiener filter can deal with gauss noise better and can save highfrequency message of the image, but not good for salt and pepper noise.For the edges of the flaws, several edge extracting methods were tested and the resultshows that the LOG method can extract the edges best. Then we created a BP network basedon the types of flaws and the number of input and output. After training the net we made asimulation and get the expected data.