Thermal Image-Based Hotspot Detection System in Solar Panels by Lilnet-5 and Relu6-F-Rcnn

Abstract

The effects of high temperatures are the hotspots that occur in solar panels. In previous works, the particular hotspot-affected panel detection on a large scale and the damage-percentage analysis are not concentrated. Hence, this article proposes an efficient detection system for solar panel hotspot identification from thermal images. To convert Photovoltaic (PV) power from one voltage to another voltage level, solar PV panels are employed and connected with the DC-DC converter. The Student T Distributed-Osprey Optimization Algorithm (ST-OOA) is utilized for controlling the converter duty cycle. The output obtained from the converter is in a signal waveform that is converted into a snowflake image by a Symmetrized Dot Pattern (SDP) and given to the LeCun Initialized LeNet-5 (LILNet-5) classifier for fault-panels detection. Afterward, the fault panels’ thermal images are preprocessed along with faults and cracks segmentation by utilizing Rectified Linear Unit 6 Activated Faster Region-based Convolutional Neural Network (ReLu6-F-RCNN). Lastly, to identify the damage percentage of solar panels, power loss is calculated. Moreover, the performance analysis exhibited the framework’s robustness in effective solar panel hotspot detection.