AMSP-UOD: When Vortex Convolution and Stochastic Perturbation Meet Underwater Object Detection

TL;DR

AMSP-UOD tackles underwater object detection under non-ideal imaging conditions by introducing AMSP-VConv to disrupt noise distributions and reduce parameters, and FAD-CSP to fuse global and local features, complemented by NMS-Similar post-processing for dense scenes. Grounded in the imaging model $I=H(J,B,t)+N$, the approach demonstrates improved noise immunity and robustness on URPC and RUOD datasets while maintaining efficiency. Key contributions include the AMSP-VConv backbone, the Global Feature-Aware Representation with RepBottleneck in FAD-CSP, and the NMS-Similar post-processing, each validated through extensive ablations. The results indicate strong practical potential for real-world underwater exploration and monitoring applications, especially in cluttered environments with small targets like waterweeds and fish clusters.

Abstract

In this paper, we present a novel Amplitude-Modulated Stochastic Perturbation and Vortex Convolutional Network, AMSP-UOD, designed for underwater object detection. AMSP-UOD specifically addresses the impact of non-ideal imaging factors on detection accuracy in complex underwater environments. To mitigate the influence of noise on object detection performance, we propose AMSP Vortex Convolution (AMSP-VConv) to disrupt the noise distribution, enhance feature extraction capabilities, effectively reduce parameters, and improve network robustness. We design the Feature Association Decoupling Cross Stage Partial (FAD-CSP) module, which strengthens the association of long and short range features, improving the network performance in complex underwater environments. Additionally, our sophisticated post-processing method, based on Non-Maximum Suppression (NMS) with aspect-ratio similarity thresholds, optimizes detection in dense scenes, such as waterweed and schools of fish, improving object detection accuracy. Extensive experiments on the URPC and RUOD datasets demonstrate that our method outperforms existing state-of-the-art methods in terms of accuracy and noise immunity. AMSP-UOD proposes an innovative solution with the potential for real-world applications. Our code is available at https://github.com/zhoujingchun03/AMSP-UOD.

Figures (6)

• Figure 1: AMSP-UOD network architecture: AMSP-VConv for underwater noise elimination; FAD-PAN for information analysis, FAD-CSP for semantic feature decoupling; NMS-Similar for merging traditional and Soft-NMS for efficient dense scene detection.
• Figure 2: AMSP Vortex Convolution. (a) The AMSP-VConv structure, (b) an expanded diagram of VConv, featuring the uniquely designed Shared Conv with BN, complemented by the SiLU activation function.
• Figure 3: FAD-CSP structure. The FAD-CSP module is built on a cross-stage network, comprising an efficient Global Feature-Aware (GFA) and a local decoupling-focused RepBottleneck. The essence of FAD-CSP lies in creating an efficient decoupling network through the interaction of long and short distance features.
• Figure 4: (a) The represents the RepBottleneck structure with $n$=3, and (b) the detailed design of Bottleneck, a residual structure composed of pointwise convolution and depthwise convolution.
• Figure 5: Visualization of object detection results of different object detection methods on URPC (Zhanjiang). (a) YOLOv3 redmon2018yolov3, (b) YOLOv5s yolov5, (c) YOLOv6sli2022yolov6, (d) YOLOv7-tiny wang2023yolov7, (e) Faster R-CNN girshick2015fast, (f) Cascade R-CNN cai2018cascade, (g) RetinaNet lin2017focal, (h) FCOS fcos9010746, (i) ATSS zhang2020bridging, (j) TOOD feng2021tood, (k) PAA kim2020probabilistic, (l) Ours-Standard, (m) Ours-AMSP-VConv, (n) Ours-AMSP-VConv + NMS Similar.