ADVANCEMENTS IN SEMANTIC SEGMENTATION USING DEEP LEARNING TECHNIQUES FOR IMAGE ANALYSIS

Document Type : Original Article

Author

Biomedical Engineering Research Centre, University of Anbar, Anbar, Iraq

Abstract

In this study, the progression in semantic segmentation has been explored using deep learning architectures such as SegNet, FCN-AlexNet, and U-Net with EfficientNet-B3 backbone. It assesses their performance on a range of datasets including UAV imagery, Cityscapes and ADE20K, as well as comparing to the accuracy of U-Net (89.7% MPA) and generalization. Problems such as computational sophistication, class inequality, and real-time processing limitations are investigated, highlighting trade-offs between acceleration and exactness. While identifying gaps for domain adaptation, and adversarial robustness, the paper discusses optimization strategies as attention mechanisms or self-supervised learning. Practical deployment would follow in future directions of the deployment with lightweight models, multimodal fusion and explainable AI. The results emphasized that for segmentation tasks, encoder−decoder designs are beneficial for their utility in autonomous vehicles as well as for medical imaging. Semantic segmentation is one of the important tasks in Computer Vision where each pixel of an image is labelled to accurately segment an object.

Keywords

Files

Share

How to cite

Statistics