当前位置：
X-MOL 学术
›
Remote Sens. Environ.
›
论文详情
Our official English website, www.x-mol.net, welcomes your
feedback! (Note: you will need to create a separate account there.)
CascadeDumpNet: Enhancing open dumpsite detection through deep learning and AutoML integrated dual-stage approach using high-resolution satellite imagery
Remote Sensing of Environment ( IF 11.1 ) Pub Date : 2024-08-10 , DOI: 10.1016/j.rse.2024.114349 Siwei Zhang , Jun Ma
Remote Sensing of Environment ( IF 11.1 ) Pub Date : 2024-08-10 , DOI: 10.1016/j.rse.2024.114349 Siwei Zhang , Jun Ma
Managing unregulated dumpsites in urban environments is a pressing issue, but determining their locations presents persistent challenges due to their irregular distribution. Recently, researchers have begun employing aerial view remote sensing imagery coupled with deep learning detection to identify open dumpsites. However, due to interference from scenes with similar visual features in the satellite imagery, the false alarm rate remains high. A model named CascadeDumpNet is proposed by this study, which integrates deep learning and automated machine learning for open dumpsite detection, effectively combining the visual characteristics of open dumpsites with the features of the surrounding environment to eliminate erroneous detection and improve detection precision. Notably, the model is equipped with a novel Contextual Feature Synthesis (CFS) module that was specifically designed to enhance object detection in bird-view remote sensing imagery. This module is adept at leveraging remote sensing contextual information, thereby significantly refining the detection process by considering the broader environmental features of dumpsites. The performance of the model was compared with six advanced object detection architectures to demonstrate its superiority. High-resolution multispectral satellite imagery from the Pléiades satellite, featuring a spatial resolution of 0.5 m, was utilized to analyze the dumpsite distribution in Shenzhen, China. The model was applied to this densely urbanized area, demonstrating its effectiveness in detecting open dumpsites within such environments. Additionally, the transferability of the model was verified through successful applications in two other major Chinese cities, Shanghai and Guangzhou. Furthermore, the distribution pattern of dumpsites was analyzed, which revealed that the density of dumpsites is predominantly concentrated in highly urbanized areas, which are characterized by high population densities, and a strong correlation was observed between the locations of these dumpsites and proximities to forests, elevated highways, and industrial zones. Overall, the development of the CascadeDumpNet model provides new insights into urban waste management, offering a novel, efficient, and precise approach to the detection and analysis of open dumpsites.
更新日期：2024-08-10
京公网安备 11010802027423号