RGM-YOLO

Introduction

Underwater object detection remains a challenging task due to the presence of noise, lighting variations, and occlusions in underwater images. To address these challenges, we proposes an improved underwater object detection model based on YOLOv9, integrating advanced attention mechanisms and a dilated large kernel algorithm. Specifically, the model incorporates a Residual Attention Block (RAB) to enhance local feature extraction and denoising capabilities. Additionally, a content-guided hybrid multi-attention fusion module is designed to improve contextual awareness and target focus. Finally, a dilated large kernel network, GLSKNet, is employed to dynamically adjust the receptive field, making the model more suitable for detecting underwater objects of varying sizes, particularly small and blurred targets.

Experimental results on the RUOD and DUO datasets demonstrate that our model outperforms several state-of-the-art models, achieving impressive mAPs (mean Average Precision) of 88.8% and 89.7% on the RUOD and DUO datasets, respectively. These findings underscore the effectiveness of attention mechanisms and the dilated large kernel algorithm in enhancing underwater object detection performance.

Application Scenarios:

• Real-time object detection on underwater exploration devices.
• Underwater object recognition and environmental monitoring.
• Target tracking and detection in oceanographic research.

Key Algorithm Description and Implementation

• Residual Attention Block: Enhances the model's denoising capability and optimizes feature quality.The Residual Attention Block (RAB) combines residual connections with spatial attention mechanisms to enhance the model's feature extraction capabilities and denoising performance. It captures deep-level features through multiple residual stacking, while the spatial attention mechanism guides the model to focus on target regions and suppress background noise. This effectively optimizes feature quality and improves model performance in complex scenarios.The code implementation can be found in DRANet.py.
• Expansion Large-kernel Network: Improves algorithm flexibility by expanding the receptive field.It enhances feature extraction capabilities by combining multi-scale convolutions and global context attention. It first captures multi-scale features and then uses the LSKblock with large kernel attention mechanisms to focus on important regions. This design improves flexibility, highlights key features, and suppresses noise, resulting in better performance in complex tasks.The code implementation can be found in GLSKNet.py.
• Content-Guided Attention-Based Hybrid Fusion Module: Fuses low-level and high-level features, focusing on more valuable information.It combines channel attention, spatial attention, and pixel-level attention mechanisms to effectively fuse multi-scale contextual information. First, spatial attention highlights important regions, while channel attention captures inter-channel dependencies. Then, pixel attention adaptively weights the fused features at the pixel level. The final fusion integrates the weighted contributions of the input features, ensuring that the output retains both global context and fine-grained details, making it highly suitable for complex tasks requiring precise feature representation.The code implementation can be found in DEANet.py.

Requirements

Dependencies

The dependency list can be found in the requirements.txt file. The main dependencies include but are not limited to:

• Python>=3.8.0
• pytorch>=2.0.1
• matplotlib>=3.2.2
• numpy>=1.18.5
• opencv-python>=4.1.1
• Pillow>=7.1.2
• PyYAML>=5.3.1
• requests>=2.23.0
• scipy>=1.4.1
• thop>=0.1.1
• torch>=1.7.0
• torchvision>=0.8.1
• tqdm>=4.64.0
• pandas>=1.1.4
• seaborn>=0.11.0
• albumentations>=1.0.3
• pycocotools>=2.0

Easily install the Dependencies with the following command:

pip install -r requirements.txt

It was ran in following environment:

• NVIDIA 3090 GPU
• Intel(R) Core(TM) i7-14700K
• Ubuntu 22.04.4 LTS
• Python

Preparing Data

1. To build dataset, you'll also need following datasets.

• RUOD
• DUO

2. Structure of the generated data should be：

├── datasets
    ├──Images
    │  ├── train
    │  │   ├── 000001.jpg
    │  │   ├── 000002.jpg
    │  │   └── ...
    │  ├── val
    │  │   ├── 000001.jpg
    │  │   ├── 000002.jpg
    │  │   └── ...
    │  ├── test
    │  │   ├── 000001.jpg
    │  │   ├── 000002.jpg
    │  │   └── ...
    ├──Labels
    │  ├── train
    │  │   ├── 000001.jpg
    │  │   ├── 000002.jpg
    │  │   └── ...
    │  │
    │  ├── val
    │  │   ├── 000001.jpg
    │  │   ├── 000002.jpg
    │  │   └── ...
    │  └── test
    │      ├── 000001.jpg
    │      ├── 000002.jpg
    │      └── ...
    │
    └──data.yaml

data.yaml can be written as follows:

# Path
train: ./datasets/images/train  
val: ./datasets/images/val 
test: ./datasets/images/test 

# Classes
names:
  0: holothurian
  1: echinus
  2: scallop
  3: starfish
  4: fish
  5: corals
  6: diver
  7: cuttlefish
  8: turtle
  9: jellyfish

Pre-trained Model

You can get the pre-trained model based on the YOLOv9 from here.

Getting Started:

Usage Example

• Training

python train_dual.py --workers 8 --device 0 --batch 8 --data data/pp_fall/voc.yaml --img 640 --cfg models/detect/yolov9-c.yaml --weights '' --name yolov9-c --hyp hyp.scratch-high.yaml --min-items 0 --epochs 500 --close-mosaic 15 --evolve

• Validation

python val_dual.py --data data/pp_fall/voc.yaml --img 640 --batch 32 --conf 0.001 --iou 0.7 --device 0 --weights './yolov9-c.pt' --save-json --name yolov9_c_640_val

• Evaluation

python detect_dual.py --source './data/images/horses.jpg' --img 640 --device 0 --weights './yolov9-c-converted.pt' --name yolov9_c_c_640_detect