Segmentation of crack disaster images based on feature extraction enhancement and multi-scale fusion

The automatic detection and early warning of surface cracks have become critical for ensuring mine safety. Accurately segmenting mine crack images is a difficult task due to the intricate nature of their content. Segmentation precision can be hindered by the presence of complex environmental backgrounds, such as varying crack sizes, illumination, slope, and vegetation, all of which can interfere with the accuracy of the detection process. In this paper, we present a deep learning-based method for mine crack segmentation, which involves data preprocessing, classification of crack and non-crack images, and crack segmentation. In the segmentation network, we innovatively use the mid-level feature with rich semantic information from the specific layer in the encoder and propose a unique multi-scale attentional feature fusion method in the decoder to combine the extracted the mid-level feature with the low-level and high-level features obtained from the DeeplabV3+. This fusion method mentioned above can preserve the semantic information in the image while minimizing computational complexity, thus ensuring both efficient and accurate segmentation of cracks. We have also created a unique mine crack dataset. The experimental outcomes on this exemplary dataset suggest that our methodology achieves an 82.67% F1-score and an 70.47% IoU, surpassing other state-of-the-art methods in segmentation accuracy.