Original title: W-ESDETR: An Efficient Scale-Aware Transformer Detection Network for Inland Waterway Surveillance
Zhongzhen Yan, Yuan Cao, Yi Yu, Jie Gao +1 · September 2026
Spotting boats on inland rivers using video cameras is tough. Messy shorelines and bad weather make it hard for computer vision to find ships quickly and accurately. To fix this, researchers built W-ESDETR, a fast object detection network based on RT-DETRv2. The system filters out background clutter using a selective-kernel module. It also uses wavelet downsampling to keep fine surface textures, which helps tell similar-looking ships apart. Finally, it uses a specialized bounding-box loss to pinpoint vessels even when their outlines look blurry. In tests, the model ran at 173.4 FPSFrames per second, the number of images the model can process each second; higher values mean faster processing.. It scored 99.18% mAP@50A standard score for object detection that measures both classification and localization accuracy, where 50% overlap is required between predicted and true bounding boxes. on the SeaShips dataset and 95.2% mAP@50 on a new inland dataset, using 8.8% fewer parameters than the baseline.
You are trying to spot a rare bird in a dense forest during light rain. The leaves and branches keep getting in the way, rain blurs the bird's shape, and another bird species looks almost identical. In this paper, the ship detection system deals with the same problem: complex shoreline backgrounds, bad weather, and visually similar boats. It uses special processing steps to ignore background clutter and keep fine textures clear, just like using a camera filter and sharpening tool.
Accurate, real-time ship detection is critical for avoiding collisions and monitoring traffic on inland waterways. Current systems often struggle with complex backgrounds and bad weather, leading to delays or mistakes. This network achieves near-perfect accuracy at high speed, making it practical for live surveillance without heavy computing hardware.
The evaluation is limited to two datasets (SeaShips and a self-constructed one), so performance on other waterways or under conditions like fog, night, or different camera angles is not explicitly tested. The self-constructed dataset details are not provided, limiting reproducibility. The computational requirements for edge deployment are not discussed.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
W-ESDETR: An Efficient Scale-Aware Transformer Detection Network for Inland Waterway Surveillance — Zhongzhen Yan, Yuan Cao, Yi Yu et al. — Electronics — doi:10.3390/electronics15184322
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.