Abstract: In order to achieve the automatic lattice division of themolten copper conductors metallography and corresponding internal feature information description, according to the characteristicsof the metallographic image, such as the not obvious boundary, the confusing internal grain blocks, the irregular grain shape, and the heavy noise of cut surface, the image segmentation algorithm and the feature extraction algorithm fit for the metallography are studied. First, according to the actual situation in the identification work, the basic histogram processing of the molten metallography was performed. Then, based on the histogram specification, compared the edge detection operators and reconstructed the image morphology, and then designed and improved the watershed algorithm for automatic lattice division. Finally, the directional gradient feature of the metallographic image is obtained by the HOG algorithm. The experimental results showed that after the histogram processing and the improved watershed algorithm and HOG algorithm, the lattice area division and internal feature information extraction can be effectively performed.

Key words: elt metallography, histogram specification, edge detection, image segmentation, HOG algorithm；fire investigation