Operating environment: Supports Windows 7 and above 64-bit operating systems, with a minimum display resolution of 1080×720p.
With advancements in image recognition algorithms based on convolutional neural networks and deep learning, the application of artificial intelligence for defect identification using high-resolution inspection images has become more accessible and mature. The specific workflow is as follows:
Step One: Mark collected image data and provide it to the convolutional neural network model for training. The trained neural network model can rapidly detect cracks in images. The recognition rate of the model can be collectively improved through refining the model and gathering more training data.
Step Two: For images in which defects have been identified, a series of processes including enhancement, noise reduction, image segmentation, and edge detection are conducted to obtain results of crack segmentation and measurements at the pixel level.
Utilizing optimization segmentation algorithms based on deep learning, the crack pixel-level overlap rate is high. The algorithm can provide accurate segmentation results even in cases of small cracks, complex backgrounds, and interference from human markings.
Meanwhile, other defects in bridges, such as cracks in steel structural welds, corrosion, loose bolts, and voids, are also identified.