Image recognition and classification deep learning models: A journey, problems, and the ways ahead
Transactions on Computing Science · 2026
DOI: 10.63808/tcs.v2i1.291 ↗Kurz gefasst
Fachartikel in der Zeitschrift Transactions on Computing Science, 2026. Verfasst von Yang Liu. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Digital imaging tech paired with so much visual material must mean that the field researching image recognition and classification was far ahead of all other artificial intelligence research, development and application fields. Deep learning, Convolutional neural networks shook it up pretty good, reached human levels and above. Recognize an object, identify a face, get another viewpoint on medical imagery, assist autos in driving themselves, all of these. In this paper I have given a comprehensive analysis from multiple aspects on deep learning-based vision understanding model. It takes us from the most basic type of network right up to today’s complex CNNs and even up to the revolutionary vision transformers. The study carefully examines some important technical regulations and improved methods of training and the strictness of the requirements for performances. And also, there’s digging into deep, long-term issues like endless demand for labeled info, seriously big computing and environment cost from training, difficult to understand “black box” situation with the model, concerns about bad people trying to trick the model on purpose, and heavy ethical problems with being unfair, not treating everybody equally, and watching over who you share your information with. and I see so many case studies of diagnostic imaging in healthcare and e-commerce on visual search from our research today we’ve seen so many models just being thrown to do incredibly transformative things but incredibly constrained as well. Finally, the paper concludes with a roadmap of research suggestions for what to tackle next, some exciting new frontiers in data-efficient learning (FE and self-supervised), automated NAS, explainable AI (XAI), building more energy efficient ‘green AI’ models that are sustainable for long-term use, sound ethical governance. Discussion expands to include the emerging area of multimodal foundation models that combine vision and language and other sensory modes. This area both creates new possibilities and new problems. Putting together all of these results suggests that we need continuous, full-spectrum innovation in all of these areas: novel algorithms, large and sustainable infrastructure, and principled ethics if we really want to break through the problems we currently face and use deep learning to transform every part of society.
Zitieren
APA
Liu, Y. (2026). Image recognition and classification deep learning models: A journey, problems, and the ways ahead. Transactions on Computing Science. https://doi.org/10.63808/tcs.v2i1.291
Vancouver
Liu Y. Image recognition and classification deep learning models: A journey, problems, and the ways ahead. Transactions on Computing Science. 2026. doi:10.63808/tcs.v2i1.291
BibTeX
@article{Liu2026,
author = {Liu, Yang},
title = {Image recognition and classification deep learning models: A journey, problems, and the ways ahead},
journal = {Transactions on Computing Science},
year = {2026},
doi = {10.63808/tcs.v2i1.291}
}
EUVIMED