Image Processing-Based Dust Detection for Solar Panels

This study focuses on dust detection in solar panels by utilizing image processing techniques. Dust detection helps in forecasting the maintenance needs and ensures system reliability. Inefficient maintenance practices of solar panels lead to increased downtime and decreased energy production. Implementation of dust detection holds importance in ensuring consistent power output. In the existing methodologies, Artificial Neural Network algorithms are implemented for dust detection. Artificial Neural Network algorithms show a considerably lower accuracy in image processing due to overfitting and complexity of image data. In the proposed system, a comparative analysis is performed by implementing Multi-Layer Perceptron technique and Dense Net algorithm. Using Multi-Layer Perceptron, a Sequential model is built with one flatten layer and two dense layers. It shows an accuracy of 88%. The Dense Net algorithm was also implemented on the processed image dataset and it shows an accuracy of 98%. This helps in increasing the accuracy and minimizing error. The Thonny IDE is employed for the implementation of the models, with validation conducted using hardware components such as Raspberry Pi, solar panels, and camera modules. Real-time images are captured to assess the performance. The utilization of the Convolution Neural Network algorithm, specifically Dense Net, for dust detection has led to enhanced system accuracy and efficiency.