Sparse‐sensing and superpixel‐based segmentation model for concrete cracks

• DOI: 10.1111/mice.12903
• Publication Date: 2022-08-12T06:38:36Z
• Authors: Xiongyao Xie; Jielong Cai; Haozheng Wang; Qiang Wang; Jieying Xu; Yingxin Zhou; Biao Zhou
• Keywords: Overfitting; Memory footprint

Abstract Efficient image‐recognition algorithms to classify the pixels accurately are required for computer‐vision‐based inspection of concrete defects. This study proposes a deep learning‐based model called sparse‐sensing and superpixel‐based segmentation (SSSeg) accurate efficient crack segmentation. The employed sparse‐sensing‐based encoder decoder was compared with six state‐of‐the‐art models. An input pipeline 1231 diverse images specially designed train evaluate results indicated that SSSeg achieved good balance between recognition correctness completeness outperformed other models in both accuracy efficiency. also exhibited resistance interference surface roughness, dirty stains, moisture. increased depth receptive field units guaranteed representability; meanwhile, structured sparse characteristics protected network from overfitting. lightweight omitted skip connections, which greatly reduced computation memory footprint enlarged size inference.