Enhancing U-Net performance for high-resolution land cover classification using a dynamic epoch-centric optimizer (DECO).

Farhangi, Mahdi ORCID: https://orcid.org/0009-0001-5545-8604; Milan, Asghar ORCID: https://orcid.org/0000-0003-0187-6074; Shokri, Danesh et Homayouni, Saeid ORCID: https://orcid.org/0000-0002-0214-5356 (2025). Enhancing U-Net performance for high-resolution land cover classification using a dynamic epoch-centric optimizer (DECO). Remote Sensing Applications: Society and Environment , vol. 39 . p. 101668. DOI: 10.1016/j.rsase.2025.101668.

Prévisualisation

PDF - Version soumise Télécharger (3MB) | Prévisualisation

Résumé

In recent years, deep learning models—particularly U-Net—have garnered significant attention for applications such as high-resolution land cover mapping. A key challenge in improving these models' performance lies in the proper selection and tuning of optimizers: each algorithm (e.g., Adam, Nadam) offers distinct strengths and weaknesses, and reliance on a single optimizer may not yield optimal results across all training stages. Here, we introduce DECO, a novel hybrid optimizer that dynamically switches among multiple optimizers across epochs to enhance overall convergence and stability. U-Net trained with DECO on aerial imagery of buildings, forests, roads, and water in the Minski region of Warsaw, Poland, achieved 96.13 % overall accuracy, a Kappa coefficient of 91.49 %, an F1 score of 96.08 %, and a Jaccard index of 64.53 %. To assess generalizability, the model was further evaluated on a test region in the Malopolskie province, yielding 86.74 % accuracy, 73.75 % Kappa, 87.29 % F1, and 55.02 % Jaccard. Moreover, to demonstrate DECO's broader applicability, we implemented it on the DeepLab v3+ architecture, observing likewise improvements in validation accuracy and training stability. These findings substantiate that dynamic, epoch-centric optimizer switching can substantially boost the precision and robustness of deep learning models for high-resolution land cover classification.

Type de document:	Article
Mots-clés libres:	dynamic epoch-centric optimizer (DECO); deep learning optimization; optimizer switching; u-net; deeplab v3+; high-resolution aerial imagery; convolutional neural networks
Centre:	Centre Eau Terre Environnement
Date de dépôt:	08 sept. 2025 19:26
Dernière modification:	08 sept. 2025 19:26
URI:	https://espace.inrs.ca/id/eprint/16597

Gestion Actions (Identification requise)

Modifier la notice