Fault detection in solar thermal images using Convolutional neural network and hybrid optimization approach

Identification of abnormalities or irregularities in solar thermal images is necessary for fault identification since these features may point to problems like hotspots, shade, dirt, cracks, or defective cells. This paper recommends an optimization-tuned Convolutional Neural Network (CNN) classifier to classify faults, which may be caused by hotspots or cracks in solar thermal images with the abstraction of the substantial features from the input thermal images. Initially, the input image is pre-processed using a filtering approach to remove the artifacts and noise present in the image. Followed by which, the features, such as local binary pattern, Gray-Level Co-Occurrence Matrix (GLCM) feature, Local Directional Texture Pattern, Median Binary Pattern, Scale-Invariant Feature Transform (SIFT), Speed Up Robust Features (SURF) and Gradient Local ternary pattern are extracted and concatenated to form a feature vector that acts as the input to the Convolutional Neural Network (CNN) classifier to perform fault detection. The suggested Aquila Inherited Dwarf Mongoose algorithm (AQI-DM) leverages the unique characteristics of the Dwarf Mongoose and the Aquila search agents. The suggested AQI-DM-based CNN classifier's accuracy will be 0.9207 for a training rate of 60%, 0.91549 for 70%, and 0.89142 for 80%, respectively.