EUVIMEDEUVIMED entdecken
Open Access · CC BY-SA 4.0

Vegetation Classification in Land Cover Mapping Using SSD with MobileNetV2 Architecture

Yuslena Sari, Yudi Firmanul Arifin, Muhammad Tommy Maulidyanto, Nurul Fathanah Mustamin, Firman Aziz, Firda Aulia

JOIV : International Journal on Informatics Visualization · 31.05.2026 · 10 (3) · 980

DOI: 10.62527/joiv.10.3.3588 ↗

Kurz gefasst

Fachartikel in der Zeitschrift JOIV : International Journal on Informatics Visualization, online erschienen am 31.05.2026. Verfasst von Yuslena Sari und 5 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-SA 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate land cover classification is essential for environmental monitoring and land use planning, but challenges remain in detecting diverse terrain features, especially using lightweight models suitable for edge devices. This study investigates the performance of the SSD MobileNetV2 architecture for classifying vegetation types and other land-cover objects across varied environmental conditions. The dataset consists of 24 videos, each lasting 1 to 4 minutes, collected from multiple locations with differing visual and terrain characteristics. These videos were processed into image frames and used to train the model with a batch size of 32, 40,000 steps, and an 8:1:1 split for training, validation, and testing. Evaluation results showed that SSD MobileNetV2 achieved an average precision (AP) of 91.6% for high vegetation, 76.8% for medium vegetation, and 90.4% for houses. Lower performance was observed in detecting rivers (62.9%) and roads (58.7%). The model’s overall mean average precision (mAP) reached 76.06%. These results demonstrate that SSD MobileNetV2 is highly effective at identifying dominant land-cover types and offers fast inference with low computational cost, making it practical for real-time deployment. Future endeavors should focus on diversifying the training data, particularly for less common classes, by employing targeted data augmentation techniques and refining the model’s architecture or feature-extraction layers. Such positive advancements aim to increase the model's accuracy and robustness, thereby improving its reliability across a broad spectrum of land-cover types, including the most diverse and intricate.

Zitieren

APA

Sari, Y., Arifin, Y. F., Maulidyanto, M. T., Mustamin, N. F., Aziz, F., & Aulia, F. (2026). Vegetation Classification in Land Cover Mapping Using SSD with MobileNetV2 Architecture. JOIV : International Journal on Informatics Visualization, 10(3), 980. https://doi.org/10.62527/joiv.10.3.3588

Vancouver

Sari Y, Arifin YF, Maulidyanto MT, Mustamin NF, Aziz F, Aulia F. Vegetation Classification in Land Cover Mapping Using SSD with MobileNetV2 Architecture. JOIV : International Journal on Informatics Visualization. 2026;10(3):980. doi:10.62527/joiv.10.3.3588

BibTeX
@article{Sari2026,
  author = {Sari, Yuslena and Arifin, Yudi Firmanul and Maulidyanto, Muhammad Tommy and Mustamin, Nurul Fathanah and Aziz, Firman and Aulia, Firda},
  title = {Vegetation Classification in Land Cover Mapping Using SSD with MobileNetV2 Architecture},
  journal = {JOIV : International Journal on Informatics Visualization},
  year = {2026},
  volume = {10},
  number = {3},
  pages = {980},
  doi = {10.62527/joiv.10.3.3588}
}
Kostenlos starten →