A novel technique for implementing hybrid optimization technique for PV thermal images to categorize and localize the faults

Abstract

In order to identify and locate flaws in solar thermal images, this research suggests using an optimization-tuned CNN classifier. The input thermal images are initially pre-processed to remove the noise present in them. After pre-processing, features like LBP, LDP, and LOOP are extracted. The collected features are then combined to produce a feature vector, which is the input to the proposed CNN classifier. Single hotspots, multiple hotspots, and string hotspots are the three types of faults that are supposed to be classified. After the classification process, the defects are located using the VGG-16 model. The weights of the CNN and VGG-16 models are modified using the proposed AqWH algorithm, which includes the distinctive characteristics of the wild horse and the Aquila search agents, to enhance classification and localization accuracy. The suggested possesses accuracy levels of 90% for classification and 96.11% for localization tasks, showing its superiority over conventional methods.