Exportar Publicação
A publicação pode ser exportada nos seguintes formatos: referência da APA (American Psychological Association), referência do IEEE (Institute of Electrical and Electronics Engineers), BibTeX e RIS.
Fogaça, J., Brandão, T. & Ferreira, J. (2023). Deep learning-based graffiti detection: A study using Images from the streets of Lisbon. Applied Sciences. 13 (4)
J. Fogaça et al., "Deep learning-based graffiti detection: A study using Images from the streets of Lisbon", in Applied Sciences, vol. 13, no. 4, 2023
@article{fogaça2023_1724524992851, author = "Fogaça, J. and Brandão, T. and Ferreira, J.", title = "Deep learning-based graffiti detection: A study using Images from the streets of Lisbon", journal = "Applied Sciences", year = "2023", volume = "13", number = "4", doi = "10.3390/app13042249", url = "https://www.mdpi.com/2076-3417/13/4/2249/htm" }
TY - JOUR TI - Deep learning-based graffiti detection: A study using Images from the streets of Lisbon T2 - Applied Sciences VL - 13 IS - 4 AU - Fogaça, J. AU - Brandão, T. AU - Ferreira, J. PY - 2023 SN - 2076-3417 DO - 10.3390/app13042249 UR - https://www.mdpi.com/2076-3417/13/4/2249/htm AB - This research work comes from a real problem from Lisbon City Council that was interested in developing a system that automatically detects in real-time illegal graffiti present throughout the city of Lisbon by using cars equipped with cameras. This system would allow a more efficient and faster identification and clean-up of the illegal graffiti constantly being produced, with a georeferenced position. We contribute also a city graffiti database to share among the scientific community. Images were provided and collected from different sources that included illegal graffiti, images with graffiti considered street art, and images without graffiti. A pipeline was then developed that, first, classifies the image with one of the following labels: illegal graffiti, street art, or no graffiti. Then, if it is illegal graffiti, another model was trained to detect the coordinates of graffiti on an image. Pre-processing, data augmentation, and transfer learning techniques were used to train the models. Regarding the classification model, an overall accuracy of 81.4% and F1-scores of 86%, 81%, and 66% were obtained for the classes of street art, illegal graffiti, and image without graffiti, respectively. As for the graffiti detection model, an Intersection over Union (IoU) of 70.3% was obtained for the test set. ER -