TSNET: A solid waste instance segmentation model in China based on a Two-Step detection strategy and satellite remote sensing images

作者：Jiaqi Yu, Pan Mao, Wenfu Wu, Qingtao Wang, Shao Xiang, Jiahua Teng, Yifei Wang · 发表于：International Journal of Applied Earth Observation and Geoinformation · 年份：2025 · DOI：10.1016/j.jag.2025.104366 · 被引用次数：8 · 研究领域：Municipal Solid Waste Management、Remote-Sensing Image Classification、Healthcare and Environmental Waste Management

Due to the concealment and randomness of solid waste disposal sites, the time and manpower costs for manual on-site inspections or remote sensing visual interpretation are significantly increased. Therefore, there is an urgent demand for the development of instance segmentation models for solid waste based on remote sensing imagery. However, there is currently no instance segmentation model specifically designed for solid waste. Hence, this study utilized high-resolution satellite remote sensing imagery to create a four-band sample dataset named Four-Band Solid Waste Dataset, which covers common types of solid waste in China. Utilizing this dataset, we propose a network structure called the Two-Step Detection Network (TSNET). In the initial detection step, we design an Segmentation with Feature Enhancement-YOLO (SFE-YOLO) instance segmentation model, prioritizing the instance segmentation of three categories: industrial waste, tailings ponds, and the other solid waste. In the second detection phase, we utilize the ResNet network model to reclassify the other solid waste identified in the first step, thereby distinguishing construction waste, household garbage, and mixed garbage. This approach addresses the imbalance in solid waste samples and achieves instance segmentation for various types of solid waste. Experimental validation conducted on the Four-Band Solid Waste Dataset we curated shows that our proposed TSNET model, employing the two-step detection strategy, achieves a...