CASolarNet: Channel Attention EfficientNet-based Model for Solar Panel Dust Detection
Hussein M. A. Mohammed, Abdulfattah E. Ba Alawi · 2024
The accumulation of dust on solar panels dramatically reduces the efficiency of solar energy systems, an issue that is often overlooked in both urban and rural settings. This work proposes a novel solution to this widespread problem of Solar Dust Detection. Utilizing EfficientNet, which is known for its balanced approach to model depth, width, and resolution, our proposed model accurately identifies dust on solar panels. The suggested model (CASolarNet) employs Channel Attention, EfficientNet, and focal loss to enhance the performance of detecting dust on solar cells. By dynamically modifying feature weights, channel attention techniques enhance dust detection by emphasizing the locations of dust in an image. This capability is crucial for maintaining optimal energy production and extending the lifespan of solar cells. Our model has been extensively tested and validated, resulting in a high accuracy rate of more than 98%. The obtained performance demonstrates the model's effectiveness in various environmental conditions, establishing a new standard in solar maintenance technology. The use of the proposed CASolarNet model has several advantages, including increased energy efficiency, lower maintenance costs, and greater sustainability of solar power systems.