🤖 AI

Cognitive Edge Device (CED) for Real-Time Environmental Monitoring in Aquatic Ecosystems

This paper introduces the Cognitive Edge Device (CED) computing platform and two new underwater datasets to facilitate the real-time detection of invasive crayfish and plastic debris using YOLO-based object detection models.

Original authors: Dennis Monari, Farhad Fassihi Tash, Jordan J. Bird, Ahmad Lotfi, Isibor Kennedy Ihianle, Salisu Wada Yahaya, Isibor Kennedy Ihianle, Md Mahmudul Hasan, Pedro Sousa, Pedro Machado

Published 2026-02-10

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Dennis Monari, Farhad Fassihi Tash, Jordan J. Bird, Ahmad Lotfi, Isibor Kennedy Ihianle, Salisu Wada Yahaya, Isibor Kennedy Ihianle, Md Mahmudul Hasan, Pedro Sousa, Pedro Machado

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The "Underwater Security Guard": Protecting Our Rivers and Lakes

Imagine you are a lifeguard at a massive, crowded swimming pool. Your job is to watch for two things: people who shouldn't be there (uninvited guests) and trash floating in the water. Now, imagine that the pool is actually a vast, murky river, the "guests" are invasive creatures that destroy everything they touch, and the "trash" is plastic that chokes the life out of the water.

Doing this job manually is exhausting. You’d need thousands of people staring into the dark, cloudy water 24/7. This is exactly the problem scientists face in our aquatic ecosystems.

This paper introduces a solution called the Cognitive Edge Device (CED). Think of it as a "Smart Underwater Security Guard" that never sleeps, never gets bored, and can "see" through the gloom.

The Villains of the Story

The researchers are fighting two main enemies:

The Signal Crayfish: Think of these as the "bullies" of the river. They move in, take over the homes of the local, friendly crayfish, and spread a "plague" (a disease) that wipes out the native population.
Plastic Pollution: This is the "silent killer"—bits of trash that clutter the habitat and harm the animals living there.

The Hero: The CED (The Smart Security Guard)

Instead of sending humans down in scuba gear to count every crayfish and piece of plastic, the researchers built a high-tech gadget.

How it works (The Brains):
The device uses something called "Edge Computing."

The Old Way (The Slow Mail): Usually, a camera would take a picture, send it through a long cable to a big computer far away, wait for the computer to think, and then send an answer back. By then, the crayfish has already moved!
The CED Way (The Instant Reflex): The "brain" is attached directly to the camera. It’s like having a reflex. If you touch a hot stove, your hand moves instantly because the signal doesn't have to travel to your brain and back; the reflex happens right at the site. The CED processes the video right there on the spot, making decisions in a fraction of a second.

The Training: Teaching the Guard to See

To make this "guard" smart, the researchers had to teach it. They used AI (Artificial Intelligence), specifically a system called YOLO (which stands for "You Only Look Once").

Think of YOLO like a professional athlete with incredible peripheral vision. Instead of looking at a photo piece by piece, YOLO looks at the whole image once and instantly shouts, "Crayfish at 2 o'clock! Plastic at 6 o'clock!"

The researchers fed the AI thousands of photos of crayfish and plastic in messy, dark, and blurry underwater conditions so the AI wouldn't get confused by shadows or floating dirt.

The Winner: The "Goldilocks" Model

The researchers tested four different versions of this AI "brain."

Some were super smart but too slow and hungry (they used too much battery).
Some were fast but a bit clumsy (they missed things).
They eventually found the "Goldilocks" model (YOLOv8n). It was just right: it was incredibly fast, didn't eat up much battery life, and was tough enough to keep working even if the water was blurry or the lighting was bad.

Why This Matters

By deploying these "Smart Security Guards" in rivers and lakes, we can get a real-time "health report" of our water. We can see exactly where the bullies (invasive species) are spreading and where the trash is piling up. This gives humans the data we need to step in and save our native wildlife before it's too late.

Technical Summary: Cognitive Edge Device (CED) for Real-Time Environmental Monitoring in Aquatic Ecosystems

1. Problem Statement

Aquatic ecosystems are facing dual threats from invasive species and plastic pollution. Specifically, the invasive signal crayfish (Pacifastacus leniusculus) is devastating native white-clawed crayfish populations in Britain by spreading lethal fungal diseases and destroying habitats through extensive burrowing. Simultaneously, plastic debris degrades water quality and ecosystem health.

Traditional monitoring methods (manual sampling, visual surveys, and expert analysis) are insufficient because they are:

Resource-intensive and slow: Requiring significant manual labor and fieldwork.
Scalability-limited: Difficult to implement over large spatial or temporal scales.
High-latency: Centralized processing of remote data delays timely interventions.
Environmentally challenging: Underwater conditions (low visibility, turbidity, light distortion, and motion blur) make standard computer vision difficult.

2. Methodology

The authors propose the Cognitive Edge Device (CED), a computing platform that integrates Artificial Intelligence (AI), the Internet of Things (IoT), and Edge Computing to enable real-time, on-the-fly detection.

Hardware Architecture: The prototype utilizes an NVIDIA Jetson Orin Nano (NJON) equipped with an Intel RealSense D435i depth sensor. The NJON provides a high-performance GPU (512-core) and Tensor Cores to allow for local, high-speed inference, reducing the need for continuous high-bandwidth data streaming to the cloud.
Software & Model Architecture: The study evaluates four variants of the You Only Look Once (YOLO) object detection framework: YOLOv5n, YOLOv5s, YOLOv8n, and YOLOv8s. These were selected for their ability to balance accuracy with the computational constraints of edge devices.
Data Pipeline:
- Datasets: The researchers utilized and released two custom underwater datasets: one for signal crayfish (2,486 images) and one for aquatic plastic debris (1,220 images).
- Pre-processing & Augmentation: To combat underwater visibility issues, the authors employed background subtraction, image scaling (416x416), vertical/horizontal flipping, random cropping, mosaic augmentation, and color/hue/saturation adjustments to simulate diverse lighting and turbidity.
Evaluation Metrics: Performance was measured using Mean Average Precision (mAP@0.5), Precision, Recall, and F1-score. For hardware efficiency, the authors measured inference time (seconds) and energy consumption (Joules) on both CPU and GPU.

3. Key Contributions

Novel Hardware/Software Integration: The CED platform provides an end-to-end workflow (Data Capture $\rightarrow$ Detection/Tracking $\rightarrow$ Data Archiving) specifically optimized for underwater edge deployment.
Publicly Available Datasets: The release of two annotated, high-quality underwater datasets (crayfish and plastic) provides a benchmark for the research community.
Edge-Optimized Detection: The study identifies the optimal trade-off between detection accuracy and power efficiency for resource-constrained underwater autonomous systems.
Advanced Pre-processing: The inclusion of background subtraction and robust data augmentation specifically tailored for underwater optical distortions.

4. Results

Detection Accuracy:
- YOLOv5s achieved the highest overall accuracy with an mAP@0.5 of 0.90 and a precision of 0.93.
- YOLOv8s demonstrated the best recall (0.84) and the lowest validation box loss (0.95).
Edge Efficiency (on NJON GPU):
- YOLOv8n was identified as the most efficient model for real-time deployment. It recorded the fastest inference time (0.29s) and the lowest energy consumption (0.09J).
- In contrast, YOLOv5n was significantly slower (0.48s) and more power-hungry (2.43J on CPU).
Robustness: YOLOv8n showed superior stability and robustness when subjected to Gaussian noise and motion blur (simulating moving fish or camera instability) compared to the YOLOv5 variants.
Hardware Comparison: GPU execution consistently outperformed CPU execution in both speed and energy efficiency across all models.

5. Significance

The CED platform represents a significant advancement in automated biodiversity monitoring. By moving intelligence to the "edge" (the device itself), the system enables:

Real-time Intervention: Immediate detection allows for faster responses to invasive species spread or pollution events.
Cost-Effective Long-term Monitoring: The low power consumption of the YOLOv8n/NJON combination allows for extended deployment in remote areas with minimal maintenance.
Scalable Conservation: The platform provides actionable, high-resolution data (abundance, species richness, and presence) that can assist policymakers and conservationists in making informed decisions to protect endangered native species and mitigate environmental damage.