We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
FCA-YOLOv8n: Frequency-Augmented Lightweight Network for River Plastic Waste Detection
Summary
Researchers built a lightweight AI system that drones can use to spot plastic trash floating in rivers—even small, clear, or shiny pieces that are normally hard to see—while running efficiently enough for real-world use. This matters because catching plastic waste in rivers before it breaks down into microplastics could help reduce the amount that ends up in our water, food, and eventually our bodies, where microplastics have raised health concerns. While this study focuses on detection technology rather than health outcomes directly, better tracking tools are a key step toward cleaning up waterways more effectively.
To address the detection challenges of riverine plastic debris from UAV perspectives—characterized by small size, high reflectivity, and semi-transparency—this paper proposes the Frequency-Spatial Collaborative Detection Network, FCA-YOLOv8n. The network reconstructs the detection head by introducing a high-resolution P2 layer and removing the redundant P5 layer, which compresses the parameter count from $ ext{3. 0 M}$ in the original YOLOv8n to $ ext{1. 2 M}$ (a $ ext{6 0 \%}$ reduction). Furthermore, it incorporates three key modules: a Frequency-selective Spatial Fusion Module (Fre-SPPF) to suppress water wave noise, a Multi-form Adaptive Attention Module (DyLG-C2f) to capture multi-scale features, and a Semi-transparent Object Fine Reconstruction Upsampling Module (FAF-Dysample) to restore faint contours. Experiments on our self-built dataset demonstrate that FCA-YOLOv8n achieves a real-time speed of 32.0 FPS, with mAP @ 0.5 reaching 94.53% and mAP @ $0.5: 0.95$ reaching 67.17 %. Compared to the baseline YOLOv8n, our model improves $ ext{mAP} {@} 0.5$ and $ ext{mAP} {@} 0.5: 0.95$ by 2.93 % and 2.80 %, respectively. This approach significantly mitigates missed and false detections in complex aquatic environments, offering an efficient and precise solution for resource-constrained UAV devices.