Abstract
To address the problem of imbalanced data of damaged building samples generated after a tornado disaster in Kaiyuan, this paper proposes overlapping sampling and improved loss function to improve the performance of the study models in identifying building damage information. In this paper, UNet-Vgg16, UNet-Resnet50, DeepLabv3plus-Xception and DeepLabv3plus-MobileNetV2 semantic segmentation models are used to extract the damage information of buildings after the Kaiyuan tornado. Overlapping sampling can expand the sample size of positive building samples, while the removal of all negative samples in a dataset can further reduce the impact of data imbalance. It is proposed that Focal loss and Dice loss be combined as the loss function to solve the imbalance problem of the study sample. The mPA, mIoU and mF1-Score of the identification results for the dataset without overlapping sampling using Ce loss of Unet-Vgg16 were 74.9%, 65.35% and 79.55%, respectively. The UNet-Vgg16 model which combines both the Focal loss and Dice loss as the loss function identifies the dataset best with overlapping sampling; it removes all negative samples, with its mPA, mIoU and F1-score reaching 94.1%, 89.25% and 93.85%, respectively. The optimization method proposed in this paper for imbalanced data can improve the recognition performance of the model for building damage information.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.