基于深度学习的屋顶光伏影像自动识别方法及系统
The UNetFormer photovoltaic perception model, developed through deep learning, solves the challenges of identifying changes in illumination and complex environments in rooftop photovoltaic image recognition. It achieves efficient and accurate photovoltaic area detection and representation, generating high-quality photovoltaic vector patterns.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JIANGSU ELECTRIC POWER RES INST
- Filing Date
- 2026-04-27
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies struggle to effectively handle variations in lighting, shadows, differences in roof materials, and insufficient geometric regularity of photovoltaic arrays in rooftop photovoltaic image recognition. This results in low recognition accuracy and reliance on manual intervention, making it difficult to achieve high-quality automatic recognition under complex lighting conditions and building environments.
We employ a deep learning-based photovoltaic perception UNetFormer model, which extracts local texture features and periodic patterns through a backbone convolutional branch and a photovoltaic array structure branch. Combined with a reflection and shadow decoupled illumination robust coding module, we utilize a multi-layer self-attention mechanism and a feedforward network to capture global semantic information, perform cross-attention fusion and skip connections, and optimize the segmentation results to generate high-quality photovoltaic vector patterns.
It improves the efficiency and accuracy of automatic recognition of rooftop photovoltaic images, enhances the robustness of the model under complex lighting conditions, optimizes the boundary processing of segmentation results, and is suitable for automatic recognition in complex lighting and building environments.
Smart Images
Figure CN122090289B_ABST