Table 5.

The advantages and limitations of DL methodologies in RCM.

DL Methodologies	Advantages	Limitations	Accuracy
Classification	Better than conventional ML approaches in terms of performance.	Demands training on large volumes of data. Very high-resolution images subjected to stitched patches with distresses. Thus, the results are discontinuous and have ambiguous structural semantics.	Ranges from 90–97%
Segmentation	Performs pixel-level classifications. Pixel-wise class assignment allows an in-depth analysis of an image. Helps in determining the morphology of the distress.	Demands training on large volumes of data. Requires post-processing algorithms to extract exact and smooth shapes from pixelated outlines. Results are prone to noises. Most of the studies seldom focus on studying the physical characteristics associated with the defects, such as width and length.	Ranges from 70–99%. Higher accuracies observed in single-class segmentation problems.
Detection	High accuracies in pavement distress detection. Provide classification as well as localization of defects. Allow the mapping of defects. With technologies, such as depth measurement systems using LiDAR, and laser, and point clouds, the measurement of physical characteristics of pavement distress is possible.	Demands training on large volumes of data. Physical characteristics of pavement distresses remain a gap, when limited to 2D data evaluation.	Ranges from 70–97%. Higher accuracies observed in single-class object detection, when compared to multi-classification and detection.