E!12759, WITURBISA, Chalmers
Diarienummer | |
Koordinator | Chalmers Tekniska Högskola AB - Institutionen för Elektroteknik |
Bidrag från Vinnova | 999 889 kronor |
Projektets löptid | juni 2019 - maj 2023 |
Status | Avslutat |
Utlysning | Eurostars – för forskande små och medelstora företag |
Viktiga resultat som projektet gav
The project has contributed to mitigating collisions between bird and wind turbine on a species level, especially in terms of the ability to identify birds in wind farms automatically
Långsiktiga effekter som förväntas
+The classification of birds required images with bird at short range to perform optimally. +It was clearly demonstrated that classification of bird species is possible using deep learning models. +With higher resolution cameras, the method will provide significant contribution to classify birds in real-time for mitigating collisions between birds and wind turbines.
Upplägg och genomförande
+ A new bird detection model has been developed to deal with false-negatives by using a gigantic detector-YOLOv7E6E at resolution-1280 and false-positive by class filtering. + A significant amount of data has been collected, labeled, processed, and involved in training a more powerful classifier (ResNet18) to perform bird species classification. + Since the distribution of the collected data is long-tailed, new techniques were involved to balance the data during training and eventually achieved better results. + The overall classification accuracy is around 92%.