Small Object Detection in Complex Large-Scale Spatial Images by Concatenating SRGAN and Multi-Task WGAN

I. Overview

Small targets occupy very few pixels in large spatial images and can be obscured by complex backgrounds. This work combines generative super-resolution and multi-task learning so the detector receives a representation with more recoverable object detail.

II. Key Contributions

• Connects SRGAN-based enhancement with a multi-task Wasserstein GAN pipeline.
• Targets the loss of detail that makes small spatial objects difficult to distinguish.
• Coordinates image reconstruction and detection-related learning objectives.

III. Methodology

Low-resolution image regions are first enhanced by a super-resolution model. The multi-task generative stage refines representations for the downstream detector, allowing reconstruction and detection cues to contribute to the learned features.

IV. Research Focus

The project investigates whether generative detail recovery can improve small-object visibility without treating super-resolution and detection as completely separate tasks.

@inproceedings{fu2021small,
  title={Small object detection in complex large scale spatial image by concatenating srgan and multi-task wgan},
  author={Fu, Yu and Zheng, Chengyu and Yuan, Liyuan and Chen, Hao and Nie, Jie},
  booktitle={2021 7th International Conference on Big Data Computing and Communications (BigCom)},
  pages={196--203},
  year={2021},
  organization={IEEE}
}