Trends, Challenges, And Future Directions of Semantic Segmentation Based on Deep Learning

Abstrak

Semantic segmentation is a fundamental task in computer vision that classifies each pixel in an image into a specific category. Advances in deep learning have significantly improved semantic segmentation performance across various applications, including medical imaging, remote sensing, autonomous driving, and industrial inspection. This study aims to analyze the development of methods, architectures, challenges, and future research directions in deep learning-based semantic segmentation. A Systematic Literature Review (SLR) was conducted using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework. Literature was collected from the SCOPUS database using keywords related to deep learning-based semantic segmentation. A total of 5,867 publications were identified, and 30 studies were selected after applying predefined inclusion and exclusion criteria. The review found that Convolutional Neural Networks (CNNs), Vision Transformers, and hybrid architectures are the dominant approaches. Attention mechanisms and multi-scale feature extraction were also identified as effective techniques for improving segmentation performance. Despite these advancements, challenges such as class imbalance, small object segmentation, and the need for large annotated datasets remain unresolved. The findings provide a comprehensive overview of current trends and highlight potential directions for future research in semantic segmentation.

Identifier

• DOI: https://doi.org/10.59934/jaiea.v5i3.2485
• OpenAlex: https://openalex.org/W7167516091