Snake-DETR: a lightweight and efficient model for fine-grained snake detection in complex natural environments

Area of Science:

• Ecology and Conservation Biology
• Computer Vision and Machine Learning

Background:

• Biodiversity is declining rapidly, with snakes crucial for ecological balance facing extinction.
• Accurate snake detection is vital for ecological monitoring but challenging due to camouflage and elusive behavior.
• Existing methods struggle with data loss and feature extraction in complex natural environments.

Purpose of the Study:

• To develop an enhanced snake detection model, Snake-DETR, for improved ecological monitoring.
• To address challenges in detecting camouflaged and elusive snakes in complex natural settings.
• To create a model suitable for real-time processing and edge device deployment.

Main Methods:

• Developed Snake-DETR based on RT-DETR, incorporating an Enhanced Generalized Efficient Layer Aggregation Network with Context Anchor Attention for improved feature extraction.
• Introduced an Enhanced Feature Extraction Backbone Network to preserve spatial and semantic information.
• Utilized lightweight Group-Shuffle Convolution for encoder optimization and Powerful-IoU loss for enhanced regression accuracy.

Main Results:

• Snake-DETR achieved high accuracy (97.66%), recall (93.92%), and mAP (95.23% @0.5).
• The model demonstrated significant reductions in computational load (47.2%) and parameters (52.2%) compared to benchmarks.
• Real-time processing capability reached 43.5 frames per second, suitable for practical applications.

Conclusions:

• Snake-DETR offers a robust solution for fine-grained snake object detection in complex environments.
• The model's efficiency and real-time performance make it ideal for wild snake fauna monitoring and edge device deployment.
• Snake-DETR provides crucial technical support for ecological research and biodiversity conservation efforts.