Super-Resolution applied to thermal images of high-voltage power lines [I-NERGY]
The project SuperPower 2.0 developed thanks to I-NERGY second open call has developed super-resolution algorithms to increment the pixels in a thermal image using Convolutional Neural Networks.
The datasets included consists of two subserts of original thermal photos of high-voltage power lines and the enhanced photos using super-resolution.
Over 5 million kilometers of power lines deliver energy to European societies. To ensure that they are good maintained, they are required by law to be regularly inspected. Currently, the main means to perform such inspection services are: 1) On-foot inspections and 2) Crewed helicopters. Our goal is to replace such means for long-range drones. These aircraft can save up to 94% costs and perform these operations with no human risks and no emissions.
The end-goal of the developed system is to be able to provide end clients (power line owners) with the best inspection possible of the infrastructure so they can detect defects as early as possible to perform maintenance before the defect becomes a real safety concern. This AI asset is one of the parts of the complete data value chain in the inspection process of the powerlines.
This more efficient, enhanced and less pollutant solution, together with the AI assets of FuVeX' parters and the validation of the final end client, will allow to break into the inspection market with a brand-new technology developing the first complete data value chain.
The shared dataset is composed by thermal images of high-voltage power lines enhanced using FuVeX super-resolution tool. These thermal photos have been improved using super-resolution algorithms to automatocally multiply by four the number of pixels in the input images.
As a tool to correctly appreciate the difference between the original and enhanced photos we include the following datasets:
1. Dataset of input images (original resolution).
2. Output images (enhanced using super-resolution).
This project has received funding from the European Union's Horizon 2020 research and innovation programme within the framework of the I-NERGY Project, funded under grant agreement No 101016508