AugMoCrack: Augmented morphological attention network for weakly supervised crack detection

Most crack-detection methods adopt a pixel-level segmentation-based approach, which requires considerable time and complexity to detect the pixel area of crack. Unlike in this paper, authors proposed an AugMoCrack network, bounding box-level crack detection approach for weakly supervised achieved by augmenting training data with Poisson blending, as well high-frequency discrete cosine transform-based features. The detects box position object from morphological perspective, such neighbour connectivity within pixels crack-area fitting. Based on also new attention loss functions considering border. Specifically, trained network using two datasets verified its performance based 591 validation images concrete dataset 672 Crack500 dataset. Compared baseline architecture, authors’ increases approximately 4.5% points 2.5% mean average precision (mAP) datasets, respectively. outperforms previous state-of-the-art learning environment where are insufficient.