Multidefect Detection Tool for Large-Scale PV Plants: Segmentation and Classification

This project focuses on developing a Multidefect Detection Tool tailored for Large-Scale Photovoltaic (PV) Plants, with a primary emphasis on segmentation and classification tasks. The input dataset comprises infrared (IR) images capturing the plant's modules. Ground truth data is incorporated through the image labeler app to facilitate supervised learning. 

The Faster R-CNN architecture is employed, integrating ground truth data, layers, and training options to effectively identify defective modules. For semantic segmentation, U-Net layers are created and subsequently combined with the Faster R-CNN architecture to enhance the detection capabilities. The system's failure classification is addressed, categorizing abnormalities based on their nature. Evaluation metrics, including Intersection over Union (IoU), accuracy, precision, and recall, are calculated to assess the model's performance. 

Through this comprehensive approach, the tool not only identifies defects within PV modules but also classifies the type of abnormality, providing a robust solution for quality assurance and maintenance in large-scale PV plants. The integration of Faster R-CNN and U-Net layers enhances the tool's versatility, making it adept at both detection and detailed segmentation tasks in the context of PV plant monitoring.

Index Terms—Class of abnormality, Faster RCNN failure mode, image classification, image segmentation, large-scale photovoltaic (PV) plant, thermographic inspection.