SAILDet: Wavelet-Preserved Lightweight One-Stage Detector for Tiny Objects in Remote Sensing

Current convolutional neural network (CNN)-based tiny object detectors in remote sensing commonly face a resolution transform bottleneck, characterized by irreversible feature information loss during downsampling and reconstruction distortions during upsampling. To address this issue, we propose a lightweight one-stage detector, small-object-aware intelligent lightweight detector (SAILDet). Its core principle is to preserve information fidelity at the source rather than compensating for its loss in downstream stages. This is achieved through a paired design that employs Haar wavelet downsampling (HWD) to retain high-frequency details at the source and Content-Aware ReAssembly of FEatures (CARAFE) to perform artifact-free, fine-grained upsampling, thereby establishing a high-fidelity feature processing loop. Experiments on the DOTA dataset demonstrate that, compared to the baseline model, SAILDet reduces GFLOPs and parameters by 11.7% and 13.0%, respectively, while improving mAP@50–95 from 0.263 to 0.266 and mAP@50 from 0.411 to 0.422. In addition, consistent gains are also observed on AI-TOD, reinforcing that directly optimizing the resolution-transform operators is more effective than downstream compensation.

• Publication year

  2026

• Venue

  IEEE Geoscience and Remote Sensing Letters

• Publication date

  Unknown publication date

• Fields of study

  Computer Science, Environmental Science

• Identifiers
  DOI 10.1109/LGRS.2025.3631843
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• DFAS-YOLO: Dual Feature-Aware Sampling for Small-Object Detection in Remote Sensing Images

  2025cited by this paper
• A survey of small object detection based on deep learning in aerial images

  2025cited by this paper
• LEAF-YOLO: Lightweight Edge-Real-Time Small Object Detection on Aerial Imagery

  2025cited by this paper
• A Tiny Object Detection Method Based on Explicit Semantic Guidance for Remote Sensing Images

  2024cited by this paper
• SOD-YOLO: Small-Object-Detection Algorithm Based on Improved YOLOv8 for UAV Images

  2024cited by this paper
• Unsupervised Bidirectional Contrastive Reconstruction and Adaptive Fine-Grained Channel Attention Networks for image dehazing

  2024cited by this paper
• Haar wavelet downsampling: A simple but effective downsampling module for semantic segmentation

  2023cited by this paper
• The Importance of Anti-Aliasing in Tiny Object Detection

  2023cited by this paper
• HIC-YOLOv5: Improved YOLOv5 For Small Object Detection

  2023cited by this paper
• A Super Lightweight and Efficient SAR Image Ship Detector

  2023cited by this paper
• Multi-Vision Network for Accurate and Real-Time Small Object Detection in Optical Remote Sensing Images

  2022cited by this paper
• Detecting tiny objects in aerial images: A normalized Wasserstein distance and a new benchmark

  2022cited by this paper
• Tiny Object Detection in Aerial Images

  2021cited by this paper
• Generalized Focal Loss: Learning Qualified and Distributed Bounding Boxes for Dense Object Detection

  2020cited by this paper
• CARAFE: Content-Aware ReAssembly of FEatures

  2019cited by this paper
• Dynamic Convolution: Attention Over Convolution Kernels

  2019cited by this paper
• MobileNetV2: Inverted Residuals and Linear Bottlenecks

  2018cited by this paper
• Focal Loss for Dense Object Detection

  2017cited by this paper
• ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices

  2017cited by this paper
• DOTA: A Large-Scale Dataset for Object Detection in Aerial Images

  2017cited by this paper
• Deconvolution and Checkerboard Artifacts

  2016cited by this paper
• Xception: Deep Learning with Depthwise Separable Convolutions

  2016cited by this paper
• Feature Pyramid Networks for Object Detection

  2016cited by this paper
• You Only Look Once: Unified, Real-Time Object Detection

  2015cited by this paper

• No citing papers are available for this paper.