Integration of deep learning and change detection for flood-damaged greenhouse extraction using UAV multispectral imagery

Rapid and accurate extraction of spatial location information of damaged greenhouses is of great significance for disaster loss assessment and post-disaster reconstruction. This paper takes the flood discharge zone in Yangying Village, Wuqing District, Tianjin City during the "23·7" catastrophic basin-wide flood of the Haihe River as the study area. Utilizing multi-temporal multispectral remote sensing data from UAVs, we constructed multidimensional spatial features and employed multiple deep learning networks to compare different image resolutions for greenhouse identification and flood-damaged greenhouse detection. The results indicate that the Blue band, Green band, and NDVI index in the multidimensional spatial features exhibited higher separability for greenhouses. The SegUnet-based classification model with 0.2m resolution achieved the highest accuracy, with an overall accuracy of 99.02% and a Kappa coefficient of 0.92. By detecting dynamic changes in greenhouses across different periods, the spatial distribution of damaged greenhouses was identified with an overall accuracy of 98.87% and a Kappa coefficient of 0.80. The research findings provide a reference for the application of UAV multispectral remote sensing in greenhouse identification, disaster impact assessment, and scientific post-disaster reconstruction.