Object detection with RetinaNet on aerial imagery: the Algarve landscape
This work presents a study of the different existing object detection algorithms and the implementation of a Deep Learning model capable of detecting swimming pools from satellite images. In order to obtain the best results for this particular task, the RetinaNet algorithm was chosen. The model was...
| Autor principal: | |
|---|---|
| Outros Autores: | , , |
| Formato: | conferencePaper |
| Idioma: | eng |
| Publicado em: |
2021
|
| Assuntos: | |
| Texto completo: | http://hdl.handle.net/1822/75255 |
| País: | Portugal |
| Oai: | oai:repositorium.sdum.uminho.pt:1822/75255 |
| Resumo: | This work presents a study of the different existing object detection algorithms and the implementation of a Deep Learning model capable of detecting swimming pools from satellite images. In order to obtain the best results for this particular task, the RetinaNet algorithm was chosen. The model was trained using a customised dataset from Kaggle and tested with a newly developed dataset containing aerial images of the Algarve landscape and a random dataset of images obtained from Google Maps. The performance of the trained model is discussed using several metrics. The model can be used by the authorities to detect illegal swimming pools in any region, especially in the Algarve region due to the high density of pools there. |
|---|