MDA-DETR: Enhancing Offending Animal Detection with Multi-Channel Attention and Multi-Scale Feature Aggregation.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Wildlife Management

Background:

• Human-wildlife conflict is a growing global issue, causing significant resource loss and safety risks.
• Accurate detection of offending animals, especially in challenging conditions like obscured or blurry nighttime images, is crucial for mitigation.

Purpose of the Study:

• To develop an advanced deep learning model for accurate detection and identification of offending animals in challenging image conditions.
• To improve upon existing methods for wildlife detection, particularly in scenarios relevant to human-wildlife conflict mitigation.

Main Methods:

• Introduction of the Multi-Channel Coordinated Attention and Multi-Dimension Feature Aggregation (MDA-DETR) model.
• Integration of multi-scale features using a Multi-Channel Coordinated Attention (MCCA) mechanism for location, semantic, and long-range information.
• Utilization of a Multi-Dimension Feature Aggregation Module (DFAM) for cross-scale feature aggregation and VariFocal Loss for enhanced detail focus.

Main Results:

• MDA-DETR demonstrated superior performance on a dataset from Northeast China Tiger and Leopard National Park, featuring six common offending species.
• The model achieved higher mAP50 scores compared to RT-DETR-r18, yolov8n, yolov9-C, DETR, Deformable-detr, and DCA-yolov8.

Conclusions:

• MDA-DETR significantly enhances the accuracy of animal detection in challenging nighttime and obscured images.
• The proposed model offers a promising solution for mitigating human-wildlife conflict through improved wildlife monitoring and identification.