img
img
Boost Loss Functions for Better Change Detection    
Yazarlar (3)
Ozan Peker
Dr. Öğr. Üyesi Fatih UYSAL Dr. Öğr. Üyesi Fatih UYSAL
Kafkas Üniversitesi, Türkiye
Fırat Hardalaç
Gazi Üniversitesi, Türkiye
Devamını Göster
Özet
With developing remote sensing technologies enabling higher efficiency data to be obtained day by day and solutions are sought with artificial intelligence approaches to global problems such as increasing construction, drought, or loss of green space around the world. In this sense, the importance of the change detection approaches, which is used in detecting the differences in the images obtained at different time intervals of the same location, is increasing. In this study, Siamese networks, which are frequently used in the change detection approach, were used and it was focused on increasing the model performance by intervening in the output of the loss function by looking at the similarity score of the image pairs given as input during the training of the model. In the experiments, LEVIR-CD, which is an open data set, was used as the data set, and cross-entropy loss and PolyLoss were used as the loss functions. Experimental results revealed that the proposed approach showed 1%-2% improvement in F-score metric in both loss functions.
Anahtar Kelimeler
cosine similarity | loss functions | remote sensing | Siamese networks
Bildiri Türü Tebliğ/Bildiri
Bildiri Alt Türü Tam Metin Olarak Yayımlanan Tebliğ (Uluslararası Kongre/Sempozyum)
Bildiri Niteliği Alanında Hakemli Uluslararası Kongre/Sempozyum
Bildiri Dili İngilizce
Kongre Adı III. International Informatics and Software Engineering Conference
Kongre Tarihi 15-12-2022 / 16-12-2022
Basıldığı Ülke Türkiye
Basıldığı Şehir Ankara
BM Sürdürülebilir Kalkınma Amaçları
Atıf Sayıları
SCOPUS 2
Boost Loss Functions for Better Change Detection

Paylaş