Optimizing Solar Energy Output Through Automated Dust Detection using CNN-SVM

This research proposes a novel CNN-SVM model to enhance the efficiency of identifying and categorizing dust on solar panels. The build-up of dust has an unfavourable impact on the efficiency of these panels; thus, proper and timely cleaning processes are essential. The model that will be designed will enhance henceforth the abilities of CNNs in spatial hierarchical features extraction compared to SVMs’ inefficient high-dimensional us; the aim is to develop a solid and effective solution to address this common issue. The model was trained and evaluated using an extensive range of photos of solar panels. Concerning the image annotation, the images were annotated with different levels of dust severity, such as LIGHT, MODERATE, HEAVY, and CLEAN. The proposed model CNN-SVM also performed very well, especially with an accuracy of $95 \%$, a precision of $94 \%$, and an F1 score of $93.7 \%$. Comparing the results of the measurements with one another and with the results of the traditional image processing algorithms and independent CNN models, it can be summarized that the suggested method of CNN in conjunction with SVM is highly effective for the particular task.