A lightweight network model designed for alligator gar detection.

Area of Science:

• * Computer Vision
• * Wildlife Monitoring
• * Machine Learning

Background:

• * Existing real-time detection algorithms face limitations in turbid waters, poor lighting, and object obstruction.
• * Monitoring aquatic species like alligator gar requires efficient and accurate detection systems.

Purpose of the Study:

• * To develop a lightweight, real-time detection network model (ARD-Net) for alligator gar monitoring.
• * To enhance detection efficiency and accuracy in challenging underwater conditions.

Main Methods:

• * Developed ARD-Net, a lightweight network model focusing on reduced computation and enhanced feature extraction.
• * Integrated a cross-domain grid matching strategy for accelerated network convergence.
• * Employed an involution operator and dual-channel attention mechanism for a refined feature extractor and multi-scale detection module.

Main Results:

• * ARD-Net achieved comparable detection accuracy to YOLOv5 but with a smaller model size and 1.48x faster detection speed.
• * Demonstrated superior detection efficiency and model size advantages over the State-of-the-Art YOLOv8 method.
• * Exhibited strong real-time performance in alligator gar detection tasks.

Conclusions:

• * ARD-Net offers an effective solution for real-time alligator gar detection in adverse aquatic environments.
• * The model's lightweight design and enhanced features contribute to improved efficiency and speed.
• * ARD-Net represents a significant advancement for wildlife monitoring using computer vision technologies.