Centralized Copy-Paste: Enhanced Data Augmentation Strategy for Wildland Fire Semantic Segmentation

Collecting and annotating images for training segmentation models is often cost prohibitive, and in wildland fire science this challenge is intensified by the limited availability of reliable public datasets with labeled ground truth. This paper presents the Centralized Copy-Paste Data Augmentation (CCPDA) method, designed to support the training of deep-learning multiclass segmentation models, with particular emphasis on improving segmentation performance for the fire class. CCPDA consists of three main steps: (i) identifying fire clusters in the source image, (ii) applying a centralization technique to isolate the core fire region, and (iii) pasting the refined fire clusters onto a target image. This method increases dataset diversity while preserving the essential characteristics of the fire class. The effectiveness of this augmentation technique is evaluated through numerical analysis and comparisons with alternative augmentation strategies using a weighted-sum multi-objective optimization approach. The results show improved performance metrics for the fire class, which holds greater operational importance than other classes (fuel, ash, or background). Overall, the numerical assessment validates the efficacy of the CCPDA method in alleviating the difficulties associated with small, manually labeled training datasets. It also illustrates that CCPDA outperforms other augmentation strategies in the application scenario considered, particularly in enhancing fire class segmentation performance.