Content Details | NUML Online Research Repository

Title RjNet: Convolutional Neural Network for Detecting Dust on Solar Panel

Author(s) Abdul Rasheed

Abstract Electricity production from fossil fuels causes increasing greenhouse gas emissions in the environment. This climate impact can be considerably reduced by utilizing power with renewable resources, particularly solar energy. Due to this, electricity production from photovoltaic (PV) systems has increased during the recent few decades. However, several factors, most notably the accumulation of dust on the panels, have resulted in a significant reduction in PV energy output. To detect dust and minimize power loss, many techniques are being researched, including thermal imaging, image processing, Internet of Things sensors, machine learning, and deep learning, highlighting various downsides, including high maintenance costs and inconsistent accuracy. In this study, we used a dataset from Kaggle and built another dataset of solar panels from Pakistan. We carefully incorporated a variety of lighting conditions to make the dataset more comprehensive, allowing the model to perform well in a variety of realworld scenarios. These two merged datasets were then tested using the current state-of-theart classification methods (SOTA). Afterward, a new convolutional neural network (CNN) architecture, RjNet, is presented exclusively for detecting dust on solar panels. The suggested RjNet model outperforms other SOTA algorithms, achieving 99.218% accuracy with only 2.14 million trainable parameters. Hence, future research should concentrate on diversifying the dataset for multi-class classification by including images from various global regions and climates, using automated data collection methods such as drones, and incorporating environmental factors while addressing class imbalances to improve model robustness.

Type Thesis/Dissertation MS

Faculty Engineering and Computer Science

Department Mathematics

Language English

Publication Date 2024-12-13

Subject Mathematics

Publisher

Contributor(s)

Format

Identifier

Source

Relation

Coverage

Rights