Composite material pore detection method based on UNET deep network

Abstract

The invention discloses a composite material pore detection method based on a UNET deep network, and relates to the field of deep networks. The method comprises the following steps: processing a sliceimage by adopting a UNET deep network to obtain a pixel pore prediction picture; merging pore pixel points in the pixel pore prediction picture through M adjacency to obtain a plurality of pore blocks, performing algorithm grid calculation on each pore block to extract a minimum enclosing rectangle of the pore block as a pore region in the composite material obtained through detection, and calculating the porosity in the composite material; according to the method, non-porous parts, which are difficult to separate by a traditional method, such as pores and foreign matters / scratches can be distinguished, so the recognition precision is improved, and false detection and missing detection are avoided; in addition, dynamic self-adaptive grid division is carried out on pore blocks through a self-adaptive grid algorithm in porosity calculation, the porosity calculation value can be closer to the porosity calculation value of field personnel, and the effects of calculation precision and manual approaching are achieved.

Description

technical field [0001] The invention relates to the field of deep network, in particular to a composite material pore detection method based on UNET deep network. Background technique [0002] In the process of laying composite materials, there is often the problem of air mixing. At present, it is usually imaged through a microscope slice, and then binarized using the traditional vision method opencv, and the background is separated through morphological operations to divide the suspected area. Finally, through the threshold Filter to frame the pore area, so that the area ratio of the pores can be judged and the proportion of the pores in the current area can be calculated, and then whether the paving is qualified or not. However, the existing traditional method will have the following problems: scratches and foreign objects are relatively close to each other on the picture, and after binarization, they are similar to normal pores in the pixel-level representation, resulting...

AI Technical Summary

Problems solved by technology

However, the existing traditional method has the following problems: scratches and foreign objects are relatively close to each other on the picture, and after binarization, they are similar to normal pores in the pixel-level representation, resulting in misidentification and low recognition accuracy.

Images