EcoDetect-YOLOv2: A High-Performance Model for Multi-Scale Waste Detection in Complex Surveillance Environments

Area of Science:

• Computer Vision
• Artificial Intelligence
• Environmental Monitoring

Background:

• Conventional waste monitoring relies on manual methods, which are inefficient and prone to errors.
• Existing object detection models struggle with real-world surveillance data due to cluttered backgrounds, scale variations, and small object sizes.
• There is a need for robust, efficient, and scalable solutions for automated waste detection in complex environments.

Purpose of the Study:

• To develop a lightweight and high-efficiency object detection model for real-world waste surveillance.
• To enhance the detection of small and multi-class waste objects in intricate environments.
• To improve the robustness and generalizability of waste detection models against environmental noise and scale variations.

Main Methods:

• Introduced EcoDetect-YOLOv2, a model based on YOLOv8s architecture, incorporating a P2 detection layer for small objects.
• Integrated an efficient multi-scale attention (EMA) mechanism and a Dynamic Upsampling Module (Dysample) for improved feature representation.
• Replaced conventional convolution layers with Ghost Convolution (GhostConv) and proposed GhostResBottleneck and ResGhostCSP modules to reduce computational overhead.
• Utilized the Intricate Environment Waste Exposure Detection (IEWED) dataset, featuring complex, real-world scenes for training and evaluation.

Main Results:

• EcoDetect-YOLOv2 demonstrated superior performance over the baseline YOLOv8s on the IEWED dataset.
• Achieved improvements of 1.0% in precision, 4.6% in recall, 4.8% in mAP50, and 3.1% in mAP50:95.
• Reduced the parameter count by 19.3% while maintaining or improving detection accuracy.
• Showcased enhanced sensitivity to small objects and robustness against cluttered backgrounds and scale variations.

Conclusions:

• EcoDetect-YOLOv2 is an effective and efficient model for real-time, multi-object waste detection in complex environments.
• The model offers a scalable solution for automated urban waste management and digital governance.
• The proposed architectural modifications enhance detection capabilities and reduce computational load, making it suitable for practical deployment.