# A Dataset with Multibeam Forward-Looking Sonar for Underwater Object Detection

**Authors:** Kaibing Xie, Jian Yang, Kang Qiu  
**arXiv:** [2212.00352](https://arxiv.org/abs/2212.00352)  
**Date:** 2022-12-01  
**Summary page:** https://www.summarizepaper.com/en/paper/2212.00352v2/

## Key points

- Multibeam forward-looking sonar (MFLS) is crucial for underwater detection
- Challenges in researching underwater object detection with MFLS
- Lack of available datasets for this specific field
- Traditional processing of sonar images at the pixel level and their transformation to sector representation poses a disadvantage for AI research
- A team of researchers from Peng Cheng Laboratory in Shenzhen, China has developed the underwater acoustic target detection (UATD) dataset
- UATD dataset consists of over 9000 MFLS images captured using Tritech Gemini 1200ik sonar
- UATD dataset provides raw data of sonar images along with annotations for 10 categories of target objects
- Data collected from lakes and shallow water environments
- UATD dataset validated by applying it to state-of-the-art detectors and conducting benchmark tests
- Refined dataset aims to overcome limitations posed by lack of available datasets and improve research in AI-related areas for underwater object detection using MFLS technology
- UATD dataset contributes significantly to the field by providing a comprehensive dataset that can be utilized by researchers working on underwater object detection with MFLS

## Summary

The use of multibeam forward-looking sonar (MFLS) is crucial for underwater detection, but there are several challenges in researching underwater object detection with MFLS. One major challenge is the lack of available datasets for this specific field. Additionally, the traditional processing of sonar images at the pixel level and their transformation to sector representation, which aligns with human visual habits, poses a disadvantage for research in artificial intelligence (AI) areas. To address these challenges, a team of researchers led by Kaibing Xie, Jian Yang, and Kang Qiu from Peng Cheng Laboratory in Shenzhen, China has developed a novel dataset called the underwater acoustic target detection (UATD) dataset. This dataset consists of over 9000 MFLS images that were captured using Tritech Gemini 1200ik sonar. The UATD dataset provides raw data of sonar images along with annotations for 10 categories of target objects such as cubes, cylinders, and tires. The data was collected from both lakes and shallow water environments. To validate the practicality of the UATD dataset, the researchers applied it to state-of-the-art detectors and conducted benchmark tests to evaluate its accuracy and efficiency in detecting underwater objects. This refined dataset aims to overcome the limitations posed by the lack of available datasets and improve research in AI-related areas for underwater object detection using MFLS technology. Overall, this research contributes significantly to the field by providing a comprehensive dataset that can be utilized by researchers working on underwater object detection with MFLS. The UATD dataset not only addresses the scarcity of available data but also offers valuable insights into improving AI algorithms for more accurate and efficient detection capabilities in underwater environments.