@article{ART003370682},
author={Ju-Hwan Han and Gye-Young Kim},
title={Rethinking Data Augmentation for CCTV-Based Rainfall Intensity Classification},
journal={Journal of The Korea Society of Computer and Information},
issn={1598-849X},
year={2026},
volume={31},
number={8},
pages={51-61}
TY - JOUR
AU - Ju-Hwan Han
AU - Gye-Young Kim
TI - Rethinking Data Augmentation for CCTV-Based Rainfall Intensity Classification
JO - Journal of The Korea Society of Computer and Information
PY - 2026
VL - 31
IS - 8
PB - The Korean Society Of Computer And Information
SP - 51
EP - 61
SN - 1598-849X
AB - As the frequency and intensity of localized heavy rainfall increase due to climate change, research aimed at utilizing the dense network of CCTV cameras installed throughout cities as supplementary rainfall observation data is gaining attention. However, a common constraint in this field is the lack of labeled video data, and the effectiveness of standard techniques—such as transfer learning, data augmentation, and multi-class classification—under small-scale, imbalanced conditions has not been sufficiently validated. This study conducts four series of controlled experiments using a small-scale CCTV video dataset (6,805 clips, 3-class) collected from a single site, and cross-validates the results through augmentation decomposition, baseline, and multi-seed experiments. Under the single-site CCTV rainy-weather video conditions examined in this study, removing the standard augmentation combination (horizontal flip + color jitter) consistently improved Macro-F1 across five random seeds, with a mean improvement of +0.215 (0.428→0.643; paired t-test, t(4) = 8.90, p = 0.0009, two-sided). These results suggest that augmentation strategies validated in general action recognition need to be re-examined before being applied to the rainfall video domain.
KW - localized heavy rainfall;CCTV;rainfall intensity classification;transfer learning;data augmentation;ablation study;video recognition
DO -
UR -
ER -
Ju-Hwan Han and Gye-Young Kim. (2026). Rethinking Data Augmentation for CCTV-Based Rainfall Intensity Classification. Journal of The Korea Society of Computer and Information, 31(8), 51-61.
Ju-Hwan Han and Gye-Young Kim. 2026, "Rethinking Data Augmentation for CCTV-Based Rainfall Intensity Classification", Journal of The Korea Society of Computer and Information, vol.31, no.8 pp.51-61.
Ju-Hwan Han, Gye-Young Kim "Rethinking Data Augmentation for CCTV-Based Rainfall Intensity Classification" Journal of The Korea Society of Computer and Information 31.8 pp.51-61 (2026) : 51.
Ju-Hwan Han, Gye-Young Kim. Rethinking Data Augmentation for CCTV-Based Rainfall Intensity Classification. 2026; 31(8), 51-61.
Ju-Hwan Han and Gye-Young Kim. "Rethinking Data Augmentation for CCTV-Based Rainfall Intensity Classification" Journal of The Korea Society of Computer and Information 31, no.8 (2026) : 51-61.
Ju-Hwan Han; Gye-Young Kim. Rethinking Data Augmentation for CCTV-Based Rainfall Intensity Classification. Journal of The Korea Society of Computer and Information, 31(8), 51-61.
Ju-Hwan Han; Gye-Young Kim. Rethinking Data Augmentation for CCTV-Based Rainfall Intensity Classification. Journal of The Korea Society of Computer and Information. 2026; 31(8) 51-61.
Ju-Hwan Han, Gye-Young Kim. Rethinking Data Augmentation for CCTV-Based Rainfall Intensity Classification. 2026; 31(8), 51-61.
Ju-Hwan Han and Gye-Young Kim. "Rethinking Data Augmentation for CCTV-Based Rainfall Intensity Classification" Journal of The Korea Society of Computer and Information 31, no.8 (2026) : 51-61.
