Spatial–Sequence Interaction with Wavelet–Energy Fusion for Remote Sensing Tiny Object Detection

Tiny objects in aerial images occupy few pixels, and their features can be suppressed by downsampling and background texture before query selection. We develop an RT-DETR-based detector that couples spatial interaction, paired-sequence exchange, and wavelet–energy fusion. A Multidirectional Spatial Interaction Block gates aligned feature streams before established two-dimensional selective scanning. A Bidirectional Sequence Interaction Block interleaves corresponding tokens to exchange local and contextual responses. A Wavelet–Energy Fusion Block uses signed wavelet details and energy-guided context to construct decoder memory. A scale-conditioned localization objective combines bounded box-distance regression with normalized size weighting. The detector achieves 35.6%, 28.2%, and 21.4% AP on VisDrone, AI-TOD-v2, and TinyPerson, respectively, exceeding RT-DETR by 4.8, 7.2, and 6.3 percentage points. Architecture-only ablations account for 4.0, 5.5, and 5.0 points of these gains. Common-size inference latency increases from 12.5 to 14.1 ms with 640-by-640-pixel inputs. The results support this combination of feature interaction and localization supervision for horizontal tiny-object detection in aerial and long-distance imagery.

Authors

Institutions

Publication Details

Journal
Sensors
Published
2026-11-16
DOI
https://doi.org/10.3390/s26227231
Primary Topic
Advanced Neural Network Applications
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Spatial–Sequence Interaction with Wavelet–Energy Fusion for Remote Sensing Tiny Object Detection

Yao Feng-qin, Xuan-Ming Lei
Sensors
Advanced Neural Network Applications
article

Spatial–Sequence Interaction with Wavelet–Energy Fusion for Remote Sensing Tiny Object Detection

Yao Feng-qin, Xuan-Ming Lei
article en

Abstract

Tiny objects in aerial images occupy few pixels, and their features can be suppressed by downsampling and background texture before query selection. We develop an RT-DETR-based detector that couples spatial interaction, paired-sequence exchange, and wavelet–energy fusion. A Multidirectional Spatial Interaction Block gates aligned feature streams before established two-dimensional selective scanning. A Bidirectional Sequence Interaction Block interleaves corresponding tokens to exchange local and contextual responses. A Wavelet–Energy Fusion Block uses signed wavelet details and energy-guided context to construct decoder memory. A scale-conditioned localization objective combines bounded box-distance regression with normalized size weighting. The detector achieves 35.6%, 28.2%, and 21.4% AP on VisDrone, AI-TOD-v2, and TinyPerson, respectively, exceeding RT-DETR by 4.8, 7.2, and 6.3 percentage points. Architecture-only ablations account for 4.0, 5.5, and 5.0 points of these gains. Common-size inference latency increases from 12.5 to 14.1 ms with 640-by-640-pixel inputs. The results support this combination of feature interaction and localization supervision for horizontal tiny-object detection in aerial and long-distance imagery.

SensorsVol. 26(22)
Zhengzhou University (CN)
Advanced Neural Network Applications
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon