Aerial Image Irradiance Mapping Without 3D Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing tools for estimating solar distribution require costly three-dimensional data, which are not available everywhere, limiting their global applicability and usability.
Innovation Solution
A data-driven method using two-dimensional imagery and deep-learning models to estimate solar distribution by training a neural network on a database of images with projected irradiance data, enabling direct estimation of irradiance from two-dimensional images without intermediate three-dimensional steps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If three-dimensional data are used to evaluate solar distribution, then measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The patent uses two-dimensional aerial images as simplified copies of the actual area, replacing complex three-dimensional data. The deep learning model learns to map these 2D images directly to solar irradiance distributions, avoiding the need for expensive and complex 3D reconstructions while maintaining acceptable accuracy for solar potential estimation
Solution Approach 2:
The patent replaces the mechanical/optical system of three-dimensional reconstruction with a data-driven deep learning model. Instead of using complex geometric processing and 3D modeling algorithms, the system uses a trained neural network that processes 2D images to predict solar irradiance patterns, significantly simplifying the computational requirements
2Measurement precision
If three-dimensional reconstruction is performed, then solar distribution accuracy is improved, but loss of time and resources increase
Solution Approach 1:
The patent performs preliminary action by training the deep learning model in advance on a large dataset of 2D images with corresponding solar irradiance labels. Once trained, the model can rapidly predict solar distribution for new areas without requiring time-consuming 3D reconstruction processes, as the complex pattern recognition has already been learned during the training phase
Solution Approach 2:
The system uses readily available two-dimensional aerial imagery as input, which can be obtained quickly from satellite or drone sources, replacing the time-intensive process of acquiring and processing three-dimensional spatial data for solar analysis
3Measurement precision
If three-dimensional data are required, then solar potential estimation accuracy is improved, but adaptability decreases
Solution Approach 1:
The patent creates a universal system that can estimate solar potential in any location using only two-dimensional aerial images as input. The deep learning model is designed to be location-agnostic, adapting to different geographical contexts, building types, and environmental conditions without requiring location-specific three-dimensional data, thereby achieving global applicability
Solution Approach 2:
The system uses universally available two-dimensional aerial imagery from satellites or drones as input, which can be obtained for any location on Earth, replacing the requirement for location-specific three-dimensional data that limits global applicability
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention concerns a method for determining the solar distribution in an area, the comprising: - a phase for collecting data to form a training database (B), - a phase for training a model on the basis of the training database (B) to obtain a trained model, the input of the trained model being an image of an area seen from the sky and the output being a global cartography of the irradiance projected on each surface of the area imaged on the input image, - a phase for operating the trained model comprising: o a step of receiving an image of an area seen from the sky, and o a step of determining by the trained model a global cartography of the irradiance projected on each surface of the area imaged on the received image.