Deep Learning Edge Detection for Image Segmentation: Advances and Challenges

Abstrak

This study offers a thorough Systematic Literature Review (SLR) of current advancements in deep learning-based edge detection techniques for picture segmentation. The study is motivated by the shortcomings of conventional edge recognition methods in processing complicated images, especially when there is significant noise, low contrast, and a variety of texture variations. Deep learning techniques are becoming more and more popular because to the growing need for precise picture segmentation in a variety of industries, including autonomous driving and medical imaging. This study examines 32 carefully chosen scientific papers from reliable sources using the PRISMA 2020 technique. The results show a substantial departure from traditional approaches in favor of Transformer-based models, encoder-decoder models like U-Net, and Convolutional Neural Network (CNN)-based architectures that increase edge detection accuracy and consistency. Additionally, it has been demonstrated that combining attention processes with multi-scale feature extraction improves object border accuracy. Nonetheless, issues including the need for sizable labeled datasets, computational complexity, and restricted generalization capacity continue to be major worries. Future trends toward the creation of more effective, flexible, and real-time models are also identified by this study. It is anticipated that the results will be used as a guide for creating more reliable and useful edge detection techniques.

Identifier

• OpenAlex: https://openalex.org/W7167507540
• DOI: https://doi.org/10.59934/jaiea.v5i3.2456