| Full text | |
| Author(s): |
Moreira, Larissa Ferreira Rodrigues
;
Moreira, Rodrigo
;
Travencolo, Bruno Augusto Nassif
;
Backes, Andre Ricardo
Total Authors: 4
|
| Document type: | Journal article |
| Source: | Neurocomputing; v. 623, p. 22-pg., 2025-01-17. |
| Abstract | |
Deep learning models are widely employed to solve complex problems in different areas, particularly for image classification, because of their high performance in pattern recognition tasks. The demand for image classification extends beyond traditional computing environments and often requires deployment of embedded and low-cost devices in real-world scenarios to meet low-latency applications and user requirements. Embedding deep learning in low-cost devices is challenging due to their constrained resources, whereas deep learning models require many resources. In the literature, there are different approaches to make this embedding viable, such as reducing model complexity or improve system efficiency. Understanding these particularities is essential for proposing new low-cost model-embedding methods. Hence, this paper presents a systematic review of deep learning models for image classification using embedded devices. This review covers studies published between 2013 and 2023 and indexed in the ACM Digital Library, IEEE Xplore, PubMed, and Scopus. Our analysis included 111 studies, and we categorized eligible papers based on various attributes from the deep learning models. The key contributions of this study include identifying prevalent trends, challenges, and advancements in the field as well as summarizing techniques that enable the deployment of high-performance models on resource-constrained devices. Our findings are expected to significantly benefit this area by pointing to the perspectives and challenges inherent in the use of low-cost devices for deep learning image applications. (AU) | |
| FAPESP's process: | 18/23097-3 - SFI2: slicing future internet infrastructures |
| Grantee: | Tereza Cristina Melo de Brito Carvalho |
| Support Opportunities: | Research Projects - Thematic Grants |