Improving Pavement Crack Segmentation Using Attention Mechanism and Self-gated Activation

Image segmentation is crucial in various applications, from autonomous driving, agriculture, and manufacturing to medical imaging and satellite imaging. It helps a computer vision-based system perceive the overall scene by identifying and delineating different objects or regions. However, pavement crack segmentation remains one of the most intriguing tasks due to several factors, like irregular shapes and sizes, uneven illumination, complex backgrounds, and noise. This study proposes a lightweight pavement crack segmentation model that harnesses attention mechanisms and self-gated activation to improve the results. Exhaustive experimental analyses on two benchmark datasets demonstrate that the model outperforms the existing solutions by achieving an overall mean intersection over union (mIoU) of 69.3%, which is an improvement of 1.85% compared to the baseline. Hence, the model’s per-sample inference time is 239±7 ms.