Research And Application Of Deep Learning-based Semantic Segmentation For Photovoltaic Rooftops In Remote Sensing

Remote sensing semantic segmentation is a field that combines remote sensing technology with deep learning.It has become a research hotspot in recent years,with the main goal of automatically classifying pixels or regions in remote sensing images into meaningful land cover categories.Among them,convolutional neural networks,one of the deep learning algorithms,have shown excellent performance in semantic segmentation tasks.Remote sensing semantic segmentation technology has been widely used in urban planning,land use and land cover mapping,natural resource management,environmental monitoring,and other fields.This paper focuses on the demand for remote sensing building roof identification in the photovoltaic industry.We address the challenges of a lack of high-quality datasets and deep learning models that can estimate the photovoltaic capacity of roofs.We propose an application and solution for using remote sensing image semantic segmentation technology for the photovoltaic roof identification task in the photovoltaic industry.We establish a remote sensing map dataset for photovoltaic roof identification tasks and propose a gated decoder that integrates self-attention mechanisms to improve the existing Deeplabv3+ model.Our work includes:(1)proposing an application for using remote sensing semantic segmentation technology for photovoltaic roof identification tasks to solve industry challenges such as excessive workload and data distortion caused by traditional methods that rely on unmanned aerial vehicle photography and manual surveys.The solution includes image preprocessing,dataset establishment,photovoltaic capacity statistics,and model training to provide valuable resource statistical guidance data and analysis methods for the photovoltaic industry.(2)establishing a remote sensing map dataset suitable for photovoltaic roof segmentation tasks to address the inadequacies of existing datasets and publicly available remote sensing map data in photovoltaic roof identification tasks.The dataset contains three classes: color steel tile roofs,common roofs(including flat concrete roofs and tiled roofs),and unavailable roofs.(3)proposing a gated decoder that integrates self-attention mechanisms to improve the decoding module of the Deeplabv3+ model,effectively extracting high-level semantic information while preserving detailed information about low-level features.This approach improves the accuracy and generalization ability of the model and has a significant effect on improving the division of building roof edges and solving identification gaps within large objects.(4)designing a photovoltaic resource intelligent analysis system by integrating existing remote sensing image data sources,effective semantic segmentation models,and readily available internet development technologies.The system is mainly used for intelligent analysis of photovoltaic roof resources,integrating theoretical research and practical application.