UAV vision detection method for crane surface cracks based on Faster R-CNN and image segmentation

To realize the remote visual inspection of cracks in unreachable parts of the large crane structures, the structure surface crack detection and measure method based on Unmanned Aerial Vehicle (UAV) taken images is proposed. The images of cranes' complex steel structures are collected comprehensively through the inverted UAV inspection platform equipped with a high-resolution visible light camera. The Faster Region-based Convolutional Neural Network (R-CNN) algorithm is used to classify and detect whether there are cracks, and the positions of cracks are marked with the minimum outer rectangle boxes. The crack length, width, area, and aspect ratio parameters are identified by maximum entropy threshold segmentation, Canny edge detection, projection feature extraction, and skeleton extraction methods. A certain threshold is set for length-width ratio and area to remove the fake cracks, such as paint cracking and water stains. Experimental results show that the proposed method can meet the crane surface cracks' automatic detection and measurement requirements under complex background. The detection accuracy is 95.4%, and the detection speed is 2 frames per second (FPS). The identification error of crack width is about 5.84%. It realizes high-precision intelligent visualization detection of crane surface cracks in dangerous and harsh environments, such as high altitude and high temperature, which is of great significance to the detection and safety evaluation of large metal structures in service.

Automated summary

This study proposes a method for crack detection and measurement of large crane structures based on UAV images. The method collects high-resolution images and uses algorithms for crack detection and parameter identification, achieving automatic detection and measurement of surface cracks in dangerous environments.