Despite the remarkable achievements in object detection, the model’s accuracy
and efficiency still require further improvement under challenging underwater
condizioni, such as low image quality and limited computational resources. A
address this, we propose an Ultra-Light Real-Time Underwater Object Detection
struttura, You Sense Only Once Beneath (YSOOB). Nello specifico, we utilize a
Multi-Spectrum Wavelet Encoder (MSWE) to perform frequency-domain encoding on
the input image, minimizing the semantic loss caused by underwater optical
color distortion. Inoltre, we revisit the unique characteristics of
even-sized and transposed convolutions, allowing the model to dynamically
select and enhance key information during the resampling process, thereby
improving its generalization ability. Finalmente, we eliminate model redundancy
through a simple yet effective channel compression and reconstructed large
kernel convolution (RLKC) to achieve model lightweight. Di conseguenza, forms a
high-performance underwater object detector YSOOB with only 1.2 million
parametri. Extensive experimental results demonstrate that, with the fewest
parametri, YSOOB achieves mAP50 of 83.1% E 82.9% on the URPC2020 and DUO
set di dati, rispettivamente, comparable to the current SOTA detectors. The inference
speed reaches 781.3 FPS and 57.8 FPS on the T4 GPU (TensorRT FP16) and the edge
computing device Jetson Xavier NX (TensorRT FP16), surpassing YOLOv12-N by
28.1% E 22.5%, rispettivamente.

Questo articolo esplora i giri e le loro implicazioni.

Scarica PDF: