CrackESS: A Self-Prompting Crack Segmentation System for Edge Devices

TL;DR

CrackESS addresses the need for efficient, high-resolution crack segmentation on edge devices for structural health monitoring. It combines a self-prompting pipeline (YOLOv8 for prompts) with a lightweight, PEFT-based EdgeSAM fine-tuning and a Crack Mask Refinement Module to refine high-resolution crack masks with low latency. Through ConvLoRA-based fine-tuning and the DiceFocalLoss objective, CrackESS achieves competitive accuracy while delivering significantly faster inference than existing SAM-based methods, demonstrated on Khanhha, Crack500, and CrackCR datasets and validated on a climbing robot platform. This work advances deployable crack segmentation for autonomous infrastructure inspection, enabling real-time, high-resolution defect analysis on resource-constrained platforms.

Abstract

Structural Health Monitoring (SHM) is a sustainable and essential approach for infrastructure maintenance, enabling the early detection of structural defects. Leveraging computer vision (CV) methods for automated infrastructure monitoring can significantly enhance monitoring efficiency and precision. However, these methods often face challenges in efficiency and accuracy, particularly in complex environments. Recent CNN-based and SAM-based approaches have demonstrated excellent performance in crack segmentation, but their high computational demands limit their applicability on edge devices. This paper introduces CrackESS, a novel system for detecting and segmenting concrete cracks. The approach first utilizes a YOLOv8 model for self-prompting and a LoRA-based fine-tuned SAM model for crack segmentation, followed by refining the segmentation masks through the proposed Crack Mask Refinement Module (CMRM). We conduct experiments on three datasets(Khanhha's dataset, Crack500, CrackCR) and validate CrackESS on a climbing robot system to demonstrate the advantage and effectiveness of our approach.

Figures (5)

• Figure 1: (a) The climbing robot executed an inspection task on a concrete bridge pier. (b) The result of inspection
• Figure 2: The framework of proposed CrackESS.
• Figure 3: The image encoder block shows the architecture of EdgeSAM’s with trainable and frozen blocks. And the Segment Anything block shows the workflow of SAM.
• Figure 4: The climbing robot system using robotic total station and trackable prism.
• Figure 5: CrackESS visualization results of each module in bridge inspection experiment. The green circles with plus signs in the Point Prompts row represent positive point prompts, while the red circles with minus signs indicate negative point prompts.