Detection of Faults in Solar Panels with CNN-VGG19 Based on Coati Optimization Algorithm

Solar power has emerged as a crucial sustainable energy source, with solar panel systems playing a vital role in harnessing solar power. Nevertheless, the efficiency and reliability of PV installations can be impaired by defective solar panels, leading to reduced energy output and potential safety hazards. This study proposes an innovative framework for the detection of faulty solar panels using a VGG19 deep neural network architecture and characteristic property using the Coati Optimization Algorithm (COA). The proposed method addresses the problem of finding defective solar panels from drone or satellite images. Using the exceptional feature extraction aspect of the VGG19 architecture, the deep learning model learns automatically to differentiate features of defective panels crack, hot spot, shading effects, and more. Moreover, to enhance the efficiency of feature selection, we present the Coati Optimization Algorithm that is chosen due to its behavior, which is influenced by the methods used by Coatis in their search for water, operating in units. The experimental results demonstrate that the proposed method successfully the task of solar faulty panels detection from aerial images datasets. The effectiveness of the developed framework that combines the VGG19 architecture and COA based feature selection, lies in its ability to produce consistent and precise results in the detection of different types of panel defects. Compared to baseline techniques, our method delivers superior performance in terms of both detection speed and accuracy. Among the results, the VGG19-COA model worked satisfactorily in terms of all the above-mentioned metrics with a tremendous outcome regarding image classification. The sensitivity turned out to be 98.71% and specificity was 98.69%, which can be considered as positive and negative cases accordingly. Further, the attained accuracy of 98.34% means the number related to general precision during classification. While the model confirms its balance in precision and recall in sure-shot classification with a value of 98.68% in precision and an F1-Score of 98.35%, it has as high as 94.03% AUC score, which underlines very strong discriminative power and, therefore, good model reliability in distinguishing between classes. These outcomes underline the effectiveness of the VGG19-COA method in high-accuracy image analysis.