of Solar Panels using Artificial
Intelligence and Machine Learning
Timeline of The Project
Data Collection
Data Collection
Collected data from multiple locations and satellites, including both India and outside India of both high and low resolutions.
Data Annotation
Data Annotation
Performed Manual annotation of data using Roboflow technology to separate the arrays of solar panels from non-significant regions.
Data Augmentation
Data Augmentation
Data Augmentation technique to augment data and orient it into multiple variations of the original images for data diversity.
Model Training, Testing and Validation
Model Training, Testing and Validation
You Only Look Once (YOLO) Model is trained and tested on the augmented dataset to develop the confidence scores and oriented bounding boxes for the solar panels.
Image Black Filtering
Image Black Filtering
After detection of the solar panels, if any, and development of oriented bounding boxes, the non-solar panel area is blackened out by reducing the three color channels to zero.
HED and Canny Edge Detection Method
HED and Canny Edge Detection Method
Holistic Nested Edge and Canny Edge Detection algorithm is used for defining the closed edges as the quantization of solar panels for high resolution images.
Area Evaluation Method
Area Evaluation Method
Evaluation of solar panel area using subtractive methodology from uni-scale image area and approximating the total energy production.
Balanagar NRSC
Balanagar NRSC
Evaluation of solar panel area in Balanagar NRSC, using the deployed mode.
Shadnagar NRSC
Shadnagar NRSC
Evaluation of solar panel area in Shadnagar NRSC, using the deployed mode.
This paper presents a real-time deep learning model for detecting defects such as cracks, erosion, and dust deposition on photovoltaic panels. The model uses multi-variant neural networks and a higher-order localization technique to accurately identify and localize defects.
M. Arif Wani and T. MujtabaRenewable Energy ResearchersThis study applied Self-Supervised Learning (SSL) for solar panel segmentation to handle corrupted datasets, improving the detection accuracy of solar panels in aerial imagery, even withlimited annotated data. The method reduces annotation costs while maintaining robust performance.
De-Yue Chen et alResearchers in Remote SensingetA proposed lightweight deep learning model for solar panel defect detection uses MobileNet-based architecture. This method performs segmentation and defect localization using less computational power, making it suitable for devices with limited resources.
M. Arif Wani and T. MujtabaResearchers in Photovoltaic SystemsHyperionSolarNet, a deep learning-based approach to detect and map solar panels globally using satellite imagery. The model achieves high accuracy in classifying and segmenting solar panels from aerial views, aiding in renewable energy planning
Jane AustenPride and Pr and udiceInvestigation into the impact of solar irradiation angles and temperature on photovoltaic module performance. The study quantifies how these factors affect power output and efficiency, with the findings emphasizing the importance of panel orientation and temperature management.
Moby-Dick