Research And Application Of Defects Detection Of Large-scale Penstock Based On Visual Detection

With the increasingly serious energy problems, hydropower as a clean renewable energy will be taken seriously more and more. Penstock as an integral part of the station, during the normal operation and maintenance of the power plant, especially the production of early and after years of running, we need to conduct a comprehensive safety inspection in order to eliminate hidden problems. It is significant for station security to detect and identify defects timely. In recent years, with the rapid development of image detection technology, visual detection technology has also been considerably developed. Visual detection is an interdisciplinary research containing the machine vision technology, communications technology, embedded technology, image processing technology and computer technology.As the background of the defects detection of large-scale penstock, this subject analyzed the common defects of penstock, researched and designed a set of defect detection algorithm.This paper analyzed and researched the commonly used method of removal of uneven illumination from photos in detail, according to the shortcomings of mutations in background brightness and the losses of defects details, it proposed a method based on the average template linear filtering and wavelet-based to preprocess image defects. Simulation results showed that the algorithm removed the uneven illumination of the image and enhance the defect details well.It described the commonly used image segmentation algorithms, based on the characteristics of large-scale penstock defects image, it used threshold segmentation algorithm for image segmentation.With the use of morphological filtering to the segmented images, the noise was reduced well. According to the characteristics of defects of large-scale penstock, it selected minimum bounding rectangle to extract the defect feature, then K-means algorithm was used to classify image defects.Finally, it realized the proposed algorithm on the hardware platform SEED-DVS6446, the experimental results showed the expected goals were achieved.