Convolutional neural network-based concrete structure surface microcrack feature extraction method

Abstract

The invention discloses a concrete structure surface microcrack feature extraction method based on a convolutional neural network, and belongs to the field of concrete structure surface damage feature extraction. The method includes following steps of: establishing an image classification data set containing microcracks and backgrounds on the surface of the concrete structure; training and verifying a convolutional neural network used for concrete structure surface microcrack area identification; testing the trained and verified convolutional neural network; performing, by adopting a trained and verified convolutional neural network, microcrack area identification in the concrete structure surface image; skeletonizing the identified micro-crack area in the image; and extracting the characteristics of the microcracks according to the microcrack region skeleton. Compared with a traditional image processing method, the method has better robustness and generalization ability. Compared with a concrete structure crack feature extraction method based on a deep learning semantic segmentation algorithm, the method has the advantages of being simpler in data set manufacturing, smaller in calculation amount and higher in precision. The precision and efficiency of concrete structure surface microcrack feature extraction are improved.

Description

technical field [0001] The invention relates to the field of feature extraction of surface damage of concrete structures, in particular to a feature extraction method of micro-cracks on the surface of concrete structures based on a convolutional neural network. Background technique [0002] Concrete is widely used in various large-scale engineering structures due to its advantages such as wide source of materials, low price, low defect rate and strong plasticity. Among the many surface damages of concrete structures, cracks are the most harmful. Rainwater will directly contact the surface of steel bars through cracks, causing corrosion to steel bars, greatly reducing the service life of the structure and increasing the risk factor. Regular detection of cracks on the surface of concrete structures and maintenance at the initial stage of cracks can reduce capital investment due to structural reconstruction and increase the service life of the structure, which is in line with t...

AI Technical Summary

Problems solved by technology

The deep learning image semantic segmentation method can realize the effective detection of wide cracks on the concrete structure surface, but this method cannot detect the micro-cracks on the concrete structure surface in the image, and the training of the semantic segmentation network model required by this method needs to establish a pixel-level The crack data set requires a large manual input

In addition, the training of the semantic segmentation network model requires a large amount of data calculation, which has high requirements for the hardware equipment used for network model training.

Images