YOLOv7-ConvTrans: Hybrid Vision Transformer with YOLOv7 for Underwater Object Detection

Underwater object detection plays a significant role in marine exploration activities such as ecological monitoring, conservation of undersea ecosystems, and underwater robotics. In contrast to detection in the atmosphere on land, underwater detection is challenging due to color distortion and attenuation, which render the detection process more difficult and imprecise. To address these issues, this paper proposes an improved detection framework, YOLOv7-ConvTrans. The YOLOv7 architecture is optimized by integrating a convolution vision transformer in the neck region to enhance the detection accuracy of small and occluded objects. The proposed method reduces redundancy, computational cost, and recognition time to extract significant feature maps from the imagery. In the experimental assessment on the URPC2019 dataset, a mAP of 80.39 and on the Brackish dataset, 92.61, indicate high detection accuracy. Finally, our proposed approach provides a valuable framework for designing underwater vision systems that can be utilized in various scientific and exploratory domains.