Automated Coral Spawn Monitoring for Reef Restoration: The Coral Spawn and Larvae Imaging Camera System (CSLICS)

This paper introduces the Coral Spawn and Larvae Imaging Camera System (CSLICS), an automated, low-cost computer vision solution that significantly reduces labor-intensive manual counting while accurately monitoring coral spawn and larvae to enhance reef restoration efforts.

The Gist

Imagine the Great Barrier Reef as a bustling, underwater city that is slowly crumbling due to climate change. To save it, scientists are trying to "grow" new coral in giant tanks, like a massive underwater nursery. But there's a huge problem: counting the babies.

Every year, corals release millions of tiny eggs and sperm into the water. To grow new reefs, scientists must catch these eggs, mix them with sperm, and watch them grow into larvae. The problem is that these tiny babies are fragile, and to know if they are healthy, scientists currently have to:

1. Stir the giant tank (which can hurt the babies).
2. Scoop out a tiny cup of water.
3. Look at it under a microscope for 20 minutes.
4. Repeat this for dozens of tanks, every single day.

It's like trying to count the number of people in a stadium by asking one person to walk up to every single seat, look at it, and write it down. It takes forever, it's exhausting, and by the time you finish, the situation might have changed.

Enter CSLICS: The "Smart Eye" for Coral Nurseries

The paper introduces a new robot system called CSLICS (Coral Spawn and Larvae Imaging Camera System). Think of it as a high-tech, underwater security camera that never sleeps, never gets tired, and has a super-powered brain.

Here is how it works, broken down into simple parts:

1. The Hardware: A Tiny Robot in a Waterproof Box

Imagine a small, waterproof camera (about the size of a large smartphone) hanging just above or just inside the water of a coral tank.

• The Eyes: It uses a special microscope lens to see tiny coral eggs (which are smaller than a grain of sand).
• The Brain: Inside the box is a small computer (a Raspberry Pi) that processes images instantly.
• The Job: It takes a picture every 10 seconds, 24 hours a day, without ever touching the water or the coral.

2. The Two Modes: "Surface Watch" and "Deep Dive"

The system is smart enough to change its strategy as the coral babies grow, kind of like a parent changing how they watch their child as they grow from a baby to a toddler.

• Mode A: The Surface Watch (The "Floating Party")

  • What happens: Right after fertilization, the coral eggs are light and float on the surface like bubbles in a soda.
  • The Robot's Job: The camera looks down at the surface. It uses AI (Artificial Intelligence) to count the eggs and check if they are "breaking" (dividing into cells), which proves they are alive and fertilized.
  • The Analogy: It's like a security guard at a pool party, counting how many people are floating on the surface and checking if they are waving (a sign of life).

• Mode B: The Deep Dive (The "Swimming Class")

  • What happens: After about 12–24 hours, the babies grow tiny hairs (cilia) and start swimming down into the water. The water also gets bubbly to mix them around.
  • The Robot's Job: The camera lowers itself just below the surface. It stops looking for specific stages and just counts "swimming blobs" that are in focus.
  • The Analogy: Now the babies are in a swimming pool. The camera is looking through the water, counting the swimmers who are clearly visible, ignoring the blurry ones in the background.

3. The "Human-in-the-Loop" Teacher

The AI didn't know how to spot coral eggs at first. It was like a baby learning to recognize a cat.

• The Process: Scientists showed the computer a few pictures and said, "This is an egg, this is a baby." The computer guessed, and if it was wrong, a human corrected it.
• The Result: The computer learned very quickly. It's like a student who studies hard, takes a practice test, gets feedback, and then aced the real exam.

4. The Big Wins: Why This Matters

The paper tested this system during a massive coral spawning event on the Great Barrier Reef. Here is what they found:

• Super Accuracy: The robot counted the coral babies with about 82-83% accuracy. That's good enough to trust for saving reefs!
• The "Early Warning System": In one tank, the robot saw that the fertilization rate was dropping. It flagged this immediately. If humans were doing it manually, they might have waited a day to check, by which time the whole tank of babies could have died. The robot gave them a heads-up to fix the problem.
• Massive Time Savings: This is the biggest number. By using the robot, they saved 5,720 hours of human labor for just one spawning event.
  • The Analogy: Imagine if you had to count 60 tanks manually. It would take a team of people working non-stop for weeks. With CSLICS, it's like pressing a "Start" button and letting the robot do the work while the humans go home to sleep.

The Bottom Line

This paper isn't just about a cool camera; it's about scaling up hope.

To save the Great Barrier Reef, we need to grow millions of corals. We can't do that if we are stuck counting them one by one with a microscope. CSLICS is the tool that turns coral farming from a slow, manual craft into a fast, automated factory. It allows scientists to watch the "babies" 24/7, ensuring they are healthy, and frees up human experts to focus on the bigger picture: saving the reef.

Technical Summary

Here is a detailed technical summary of the paper "Automated Coral Spawn Monitoring for Reef Restoration: The Coral Spawn and Larvae Imaging Camera System (CSLICS)."

1. Problem Statement

Coral aquaculture is a critical strategy for reef restoration, particularly for the Great Barrier Reef (GBR), which faces severe threats from climate change and bleaching. However, the larval rearing phase represents a significant bottleneck in the production pipeline:

• Labor Intensity: Current monitoring relies on manual sampling, where operators stir tanks, extract water samples, and count individual eggs/larvae under a microscope. This process takes ~20 minutes per sample.
• Data Sparsity: Due to the high labor cost, facilities can typically only sample once per day. This frequency is insufficient to detect rapid culture deterioration (which can occur within hours).
• Physical Risk: Manual stirring and sampling can damage fragile coral spawn and disrupt the delicate hydrodynamics required for early development.
• Scalability: As facilities scale to 60+ tanks by 2030, manual monitoring becomes infeasible.

2. Methodology: The CSLICS System

The authors propose the Coral Spawn and Larvae Imaging Camera System (CSLICS), a low-cost, modular, submersible computer vision system designed for continuous, non-invasive monitoring.

A. System Architecture & Hardware

• Components: The system utilizes a Raspberry Pi 4 (8GB) paired with a High Quality Camera (Sony IMX477 sensor) and a Pimoroni macroscopic lens (0.12–1.8x magnification).
• Enclosure: Hardware is housed in Blue Robotics waterproof enclosures (IP68 rated) to withstand saltwater environments.
• Connectivity: Powered and controlled via Power-over-Ethernet (PoE), allowing remote management and scheduled power cycling.
• Cost: Approximately $1,000 USD per unit.
• Deployment: Units are mounted 5 cm above the water surface for surface monitoring and submerged (front element <1 cm) for sub-surface monitoring.

B. Operational Modes

The system operates in two distinct modes corresponding to the biological development of the coral:

1. Surface Operation (t0–t1, ~12–24 hours):
   • Context: Fertilized eggs are positively buoyant and extremely fragile. Aeration is kept low to minimize surface disturbance.
   • Function: The camera images the water surface to detect fertilization success and early embryogenesis stages (egg, cleavage, 2-cell, 4–8 cell, advanced).
2. Sub-surface Operation (t1–t2, ~6–7 days):
   • Context: Once embryos reach the "prawn-chip" stage, aeration is increased to homogenize the tank. The water surface becomes turbulent, making surface imaging impossible.
   • Function: The camera is submerged to image the water column. The focus shifts to counting total larvae density, as specific developmental staging is less critical at this stage.

C. Software & Computer Vision

• Models: Two YOLOv8 object detectors were trained:
  1. Surface Model: A multi-class detector identifying 6 classes: Egg, First Cleavage, Two-Cell, Four-to-Eight Cell, Advanced, and Damaged.
  2. Sub-surface Model: A single-class detector focusing only on in-focus coral larvae.
• Annotation Strategy: A Human-in-the-Loop (HITL) semi-supervised approach was used. A small initial dataset was manually labeled to train a preliminary model, which then pseudo-labeled new images. These were manually reviewed and used to iteratively retrain the model, significantly reducing annotation effort.
• Metrics: The system calculates Fertilization Success ($f$) using the ratio of cleaved/developing cells to total viable eggs (excluding damaged cells) on a per-image basis.

3. Key Contributions

1. First-of-its-Kind System: The first design and implementation of a tank-mounted computer vision system specifically for automated coral spawn monitoring in aquaculture.
2. Species-Specific Detection: Development of detectors for two key Indo-Pacific reef-building species (Acropora kenti and Acropora loripes), capable of measuring fertilization success rates.
3. Scalable Low-Cost Solution: Demonstration that high-frequency monitoring can be achieved using off-the-shelf, low-cost hardware ($1k/unit) rather than expensive industrial equipment (e.g., flow cytometers costing ~$500k).
4. Open Science: Commitment to releasing trained models and code to foster future research.

4. Experimental Results

The system was deployed during mass spawning events on the GBR in 2022 and 2023.

• Surface Detection Performance:

  • Overall F1 Score: 82.4% (excluding the "damaged" class, which had lower performance due to visual variability and fewer training samples).
  • Class Performance: High precision and recall for early stages (e.g., First Cleavage F1: 80.9%, Two-Cell F1: 83.6%).
  • Fertilization Monitoring: The system successfully distinguished between successful and unsuccessful cultures in real-time. Successful cultures showed a steady increase in fertilization rates, while unsuccessful ones remained low or declined. The Root Mean Square Error (RMSE) compared to manual counts was ~0.10–0.13.

• Sub-surface Detection Performance:

  • Overall F1 Score: 83.0%.
  • mAP@0.5: 87.9%.
  • Tank Count Accuracy: CSLICS estimates of total tank density showed strong alignment with manual counts (RMSE = 45,670 corals), sufficient for operational decision-making.

• Labor Savings:

  • By automating sampling at high frequencies (every 10 seconds for images, processed to counts every 5 mins/1 hour), CSLICS saved approximately 5,720 hours of labor per spawning event for a facility with 60 tanks, compared to manual sampling at the same frequency.

5. Significance and Impact

• Reef Restoration Upscaling: CSLICS removes the labor bottleneck, enabling facilities to manage the millions of larvae required for large-scale reef reseeding.
• Culture Health Optimization: Continuous monitoring allows for immediate intervention if culture health deteriorates (e.g., detecting low fertilization rates early), preventing the loss of entire batches.
• Data-Driven Insights: The high-temporal-resolution data enables researchers to correlate environmental factors (aeration, temperature, flow) with larval survival rates, optimizing aquaculture protocols.
• Non-Invasive: The system eliminates the need for repetitive tank stirring, reducing physical stress on the fragile coral spawn.

Future Work: The authors plan to improve the detection of damaged cells and redesign the housing for side-mounted imaging to prevent lens fouling during sub-surface operations.