Deep-Learning-Based Semantic Change Detection for Urban Greenery and Comprehensive Urban Areas

Urban greenery is important for maintaining ecological balance and enhancing urban ecosystems. However, it is significantly degrading due to human activities and natural disasters, making it essential to monitor both urban greenery and the overall urban environment. Recent advancements in remote sensing and deep learning technologies have led to the development of semantic change detection (SCD) techniques, which offer more detailed analysis than binary change detection. Detecting changes in natural greenery within urban environments using general SCD techniques is challenging due to the similar spectral characteristics of natural and artificial greenery. Therefore, this study proposes a direct SCD approach focusing on urban green spaces and nongreenery-related changes. This approach distinguishes between new and degraded greenery regions and categorizes them into distinct classes alongside nongreenery changes. Key innovations include the integration of atrous spatial pyramid pooling and an updated spatial attention module, enhancing the network's ability to capture objects of varying shapes and sizes within urban settings. The methodology was evaluated using two open-source datasets, SEmantic Change detectiON Dataset (SECOND) and Wuhan urban sematic understanding (WUSU), customized to emphasize urban greenery changes. Results demonstrate that our approach significantly outperforms the existing SCD techniques in accurately detecting and categorizing new and degraded greenery regions alongside overall urban changes. The proposed method achieved superior performance in terms of separated kappa, reaching 17.72% on the SECOND dataset and 36.18% on the WUSU dataset. This study addresses the limitations of current methods in monitoring urban greenery, providing an efficient tool for assessing the impact of urbanization and natural disasters on urban greenery and the broader urban environment.