Abstract:This study presents a novel method for the nighttime detection of waterborne individuals using an enhanced YOLOv5s algorithm tailored for infrared thermal imaging. To address the unique challenges of nighttime water rescue operations, we have constructed a specialized dataset comprising 5736 thermal images collected from diverse aquatic environments. This dataset was further expanded through synthetic image generation using CycleGAN and a newly developed color gamut transformation technique, which significantly improves the data variance and model training effectiveness. Furthermore, we integrated the Convolutional Block Attention Module (CBAM) at the end of the last encoder's feedforward network. This integration maximizes the utilization of channel and spatial information to capture more intricate details in the feature maps. To decrease the computational demands of the network while maintaining model accuracy, Ghost convolution was employed, thereby boosting the inference speed as much as possible. Additionally, we applied hyperparameter evolution to refine the training parameters. The improved algorithm achieved an average detection accuracy of 85.49% on our proprietary dataset, significantly outperforming its predecessor, with a prediction speed of 23.51 FPS. The experimental outcomes demonstrate the proposed solution's high recognition capabilities and robustness, fulfilling the demands of intelligent lifesaving missions.

What problem does this paper attempt to address?

The paper aims to address the problem of effectively detecting individuals in dark water environments, such as night-time water rescue operations. Current methods heavily rely on infrared thermal imaging technology, but its application in water scenarios is limited. The paper proposes an improved YOLOv5s model specifically optimized for infrared thermal imaging. The research team constructed a dedicated dataset consisting of 5736 thermal images from different water environments and increased data diversity through CycleGAN and a new color space transformation technique. They also integrated Convolutional Block Attention Module (CBAM) into the model to enhance the capture capability of feature map details and utilized Ghost Convolution to reduce computational requirements and improve inference speed. Additionally, training parameters were optimized through hyperparameter evolution. The improved algorithm achieved an average detection accuracy of 85.49% and a prediction speed of 23.51 FPS on the proprietary dataset, significantly outperforming the original model. The paper emphasizes the challenges in real water environments, such as small object occlusion, the impact of seasonal and climate changes on thermal imaging, low contrast when temperatures are close, and hardware limitations for model deployment. To address these issues, they employed image enhancement techniques, created a thermal image dataset covering various scenarios, developed thermal imaging color space transformation data augmentation techniques, and designed a lightweight thermal imaging water surface object detection network (IWT-YOLO). This network is capable of accurately identifying boats, intelligent rescue robots, humans, and humans in different states such as drowning and swimming. In conclusion, the proposed solution in this paper improves the ability of target detection in dark water environments, contributing to intelligent rescue missions, especially for water surface rescue robots like "Dolphin I".

Personnel Detection in Dark Aquatic Environments Based on Infrared Thermal Imaging Technology and an Improved YOLOv5s Model

YoloXT: A Object Detection Algorithm for Marine Benthos

MDM-YOLO: Research on Object Detection Algorithm Based on Improved YOLOv4 for Marine Organisms.

Underwater Object Detection Based on Enhanced YOLO

An Enhanced Target Detection Algorithm for Maritime Search and Rescue Based on Aerial Images

YOLO‐UOD: An underwater small object detector via improved efficient layer aggregation network

Research on Thermal Infrared Imaging and Positioning Technology for Emergency Rescue at Sea in Bad Weather

A Deep Learning-Driven High-Performance Underwater Sonar Image Target Recognition Model

Lightweight enhanced YOLOv8n underwater object detection network for low light environments

MS-YOLO: A Lightweight and High-Precision YOLO Model for Drowning Detection

Attention-Based Lightweight YOLOv8 Underwater Target Recognition Algorithm

An Improved YOLOv5-Based Underwater Object-Detection Framework

Underwater Object Detection Algorithm Based on an Improved YOLOv8

Underwater target detection algorithm based on improved YOLOv4 with SemiDSConv and FIoU loss function

RSE-YOLOv8: An Algorithm for Underwater Biological Target Detection

Underwater Object Detection Network with Dilated Deformable Convolutions and Dual-Branch Occlusion Attention Mechanism

Thermal imaging pedestrian detection algorithm based on attention guidance and local cross-level network

Underwater object detection algorithm based on attention mechanism and cross-stage partial fast spatial pyramidal pooling

A detection method of the rescue targets in the marine casualty based on improved YOLOv5s

An Improved YOLOv5 Algorithm for Drowning Detection in the Indoor Swimming Pool

Underwater Target Detection Algorithm Based on Improved YOLOv5