Deep Learning Based Solar Cell Recognition for Optical Wireless Power Transfer

TL;DR

This paper presents a deep learning approach using Yolov5-Lite for accurate solar cell recognition in optical wireless power transfer systems, achieving high accuracy and fast deployment potential.

Contribution

It introduces a lightweight deep learning model tailored for solar cell detection in complex backgrounds, improving alignment accuracy in OWPT systems.

Findings

F1 score of 91% achieved

mAP of 94.8% achieved

Model is fast and easy to deploy

Abstract

Optical wireless power transfer (OWPT) is a technology that wirelessly transmit light energy from an optical transmitter to an optical receiver, usually a solar cell. In order to achieve the highest transmission efficiency, the solar cell receiver should be accurately aligned with the optical transmitter. Hitherto, only a few works have been existed for solar cell recognition in presence of complex backgrounds. In this paper, we employ a deep learning approach based on Yolov5-Lite for the solar cell recognition purpose, due to its lightweight, fast and easy to deploy on hardware characteristics. Our tests show a high accuracy of the employed deep learning model with the highest F1 score of 91% and mAP of 94.8%. Therefore, this deep learning model is highly promising for use in OWPT systems to precisely align optical transmitters and solar cell receivers.