Open‐source deep learning‐based air‐void detection algorithm for concrete microscopic images

Abstract

Analysing concrete microscopic images is difficult because of its highly heterogeneous composition and the different scales involved. This article presents an open‐source deep learning‐based algorithm dedicated to air‐void detection in concrete microscopic images. The model, whose strategy is presented alongside concrete compositions information, is built using the Mask R‐CNN model. Model performances are then discussed and compared to the manual air‐void enhancement technique. Finally, the selected open‐source strategy is exposed. Overall, the model shows a good precision (mAP = 0.6452), and the predicted air void percentage agrees with experimental measurements highlighting the model's potential to assess concrete durability in the future.

Paste, aggregate, or air? That is the question

“…Cropping images to highlight specific areas from the cross-hair mimics labelling techniques seen in other research that employs pixel-wise and instance segmentation, by effectively minimising background noise by ensuring the focal phase occupies a larger portion of the image. 5,6 While the coarse aggregate's substantial size led to consistent and accurate identification across various image dimensions, both fine aggregate and nonaggregate phases (including voids and paste) showed improved results with reduced image sizes. By correlating the resized images to a corresponding point count density, we can depict the link between a basic CNN model's anticipated accuracy and the volume of images necessary for attaining such precision, as demonstrated in Figure 13.…”

“…Since the publication of Procedure C, multiple research teams have applied machine learning and computer vision techniques for air void analysis 4–8 . However, standards governing current construction quality assurance practices have yet to catch up with these advances.…”

“…Since the publication of Procedure C, multiple research teams have applied machine learning and computer vision techniques for air void analysis. [4][5][6][7][8] However, standards governing current construction quality assurance practices have yet to catch up with these advances. As such, ASTM C457 Procedure C was specified by the owners for the project examined here: the ongoing construction of the Gordie Howe International Bridge.…”

Modular deep learning segmentation algorithm for concrete microscopic images

“…However, thanks to the implemented strategy associating the detection on full-size and reduced-size image, the detection of air voids with various sizes with minimal diameters of around some pixels is possible (around 10-20 μm) and a final air void content of 7.76 % is predicted which is on par with the industriallymeasured air content. Further analysis can be performed, and the algorithm's superior performance compared to manual contrast enhancement methods has been highlighted in another study [50]. The median distance-to-air-void has been extracted from the cumulative curves as illustrated in Fig.…”

Laboratory Dangerous Operation Behavior Detection System Based on Deep Learning Algorithm

“…Conv module of original YOLOv5 backbone network 3×3 adopts the design of direct connection of convolution kernel and activation function, which is often effective for feature extraction of pedestrian detection in non-dense scenes, but it is often difficult to extract features effectively for dense occlusion phenomenon. Therefore, RepVGG module is introduced and modified, and the original activation function ReLU is replaced by SiLU activation function (Smit et al, 2022;Hilloulin et al, 2022). RepVGG modules all adopt a four-layer multi-branch structure, which is composed of 3×3 convolution, 1×1 volume integral branch and residual difference branch of identity, in which the first layer is a down-sampling layer with a step size of 2.…”