TY - GEN
T1 - Advanced Scale-aware Attention for Enhanced Tiny Object Detection in Remote Sensing
AU - Qiao, Tingting
AU - Chen, He
AU - Chai, Bingqian
AU - Liu, Wenchao
AU - Qi, Guijie
AU - Xie, Yizhuang
N1 - Publisher Copyright:
© 2024 IEEE.
PY - 2024
Y1 - 2024
N2 - Recent advancements in remote sensing object detection have progressed, but distinguishing objects from complex backgrounds remains difficult, especially for tiny objects. To address this, we introduce the Advanced Scale-Aware Attention (ASA) module, which retains key channel information while minimizing computational cost. This module reorganizes channels into the batch dimension and splits them into multiple sub-features. The ASA module contains three branches: the first encodes weights based on scale and channel relationships. The second and third branches use different convolution kernel sizes to capture features across various receptive fields. These branches collaborate to enhance pixel-level relationships, improving foreground-background contrast. On the AI-TOD dataset, our method improved AP50 by 8.6% over state-of-the-art approaches.
AB - Recent advancements in remote sensing object detection have progressed, but distinguishing objects from complex backgrounds remains difficult, especially for tiny objects. To address this, we introduce the Advanced Scale-Aware Attention (ASA) module, which retains key channel information while minimizing computational cost. This module reorganizes channels into the batch dimension and splits them into multiple sub-features. The ASA module contains three branches: the first encodes weights based on scale and channel relationships. The second and third branches use different convolution kernel sizes to capture features across various receptive fields. These branches collaborate to enhance pixel-level relationships, improving foreground-background contrast. On the AI-TOD dataset, our method improved AP50 by 8.6% over state-of-the-art approaches.
KW - Remote Sensing
KW - Scale-Aware Attention
KW - Tiny Object Detection
UR - http://www.scopus.com/inward/record.url?scp=86000019623&partnerID=8YFLogxK
U2 - 10.1109/ICSIDP62679.2024.10868712
DO - 10.1109/ICSIDP62679.2024.10868712
M3 - Conference contribution
AN - SCOPUS:86000019623
T3 - IEEE International Conference on Signal, Information and Data Processing, ICSIDP 2024
BT - IEEE International Conference on Signal, Information and Data Processing, ICSIDP 2024
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2nd IEEE International Conference on Signal, Information and Data Processing, ICSIDP 2024
Y2 - 22 November 2024 through 24 November 2024
ER -