Underwater & marine
Fish detection, segmentation and tracking in video, reef survey image quality, clearer underwater images and 3D depth.
Computer Scientist · Computer Vision & Deep Learning
Research Fellow, AIMS@JCU, Townsville, Australia
PhD in Computer Science
Computer vision that works underwater and in the field.
I’m a computer scientist. I build deep learning and computer vision methods, then apply them to marine science, aquaculture and agriculture: counting and tracking fish, checking reef survey images, and helping spraying robots find weeds. The core of my work is the computer science: models that learn from few labels and run in real time on real equipment.
- Peer-reviewed papers
- 20+
- Citations
- 1,000+
- Less herbicide in 25 ha of field trials
- 35%
- Reef images checked by UNIQAT
- 130,845
What I work on
Vision for places where cameras struggle
Murky water, harsh shadows and changing light break most models. My work makes them reliable there, and cheap to train.
Aquaculture
Measuring fish and prawns from photos: body landmarks, lengths and weight, with no handling. On 1,072 farmed barramundi, weight from images was within 4.5%.
Agriculture & field robots
Weed detection for spot-spraying robots, real-time shadow removal, and vision-language models for farming.
Also: automatic measurements in medical ultrasound images.
Things I've built
Open-source tools and side projects
Favabeans
favabeans.app
Calm, free apps and games for everyday life. No accounts, no ads, and your data stays on your device.
Ask my papers
papers.alzayat.com
An AI assistant that answers questions about my research, with links to the exact page in each paper.
UNIQAT
Open-source Python toolkit · with AIMS
Scores the quality of underwater survey images automatically, at 7.9 4K images per second. Published in Methods in Ecology and Evolution.
DeepFish
Open dataset and benchmark code
40,000 underwater images with labels for counting, locating and segmenting fish.
UDnet
Open-source PyTorch code · MIT licence
Makes murky, blue-green underwater photos clear, and learns without clean reference images. Tested on eight public datasets. Published in Expert Systems with Applications.
UNIQAT
Open-source Python toolkit · with AIMS
Scores the quality of underwater survey images automatically, at 7.9 4K images per second. Published in Methods in Ecology and Evolution.
DeepFish
Open dataset and benchmark code
40,000 underwater images with labels for counting, locating and segmenting fish.
UDnet
Open-source PyTorch code · MIT licence
Makes murky, blue-green underwater photos clear, and learns without clean reference images. Tested on eight public datasets. Published in Expert Systems with Applications.
Highlights
Selected work and results
Computers and Electronics in Agriculture · 2025
Robotic spot-spraying cuts herbicide on sugarcane farms
A camera and deep learning decide where to spray, so only weeds get herbicide. In field trials across 25 hectares:
- 35%less herbicide on average (up to 65%)
- 97%as effective as blanket spraying
- 54%less herbicide in farm runoff
Scientific Reports · 2020
DeepFish: a benchmark for fish in real habitats
About 40,000 underwater images from 20 habitats in tropical Australia, with labels for counting, locating and outlining fish.
Pattern Analysis and Applications · 2024
Tracking and outlining fish without human labels
The model teaches itself from how fish move between video frames, so no one has to draw thousands of outlines. Tested on three public underwater datasets.
Computer Vision and Image Understanding · 2026
Real-time 3D depth for underwater robots
Depth and surface shape from a single camera, with 90% fewer parameters and 80% lower training cost than the large models it learns from. Built with the Australian Institute of Marine Science.
Before
After
Expert Systems with Applications · 2025
FieldNet: removing shadows in real time
Shadows from machinery hide weeds from cameras. FieldNet removes them at 66 frames per second, up to 9 times faster than earlier methods. Built with industry partner AutoWeed.
News
Recent
-
2026
Best Application Award at the RSS 2026 workshop Mapping the Reef, for a semantic 3D Gaussian splatting pipeline for multi-year coral reef monitoring.
-
2026
Paper at ECCV 2026: locating every plant that matches a text description (multi-label visual grounding).
-
2026
UNIQAT published in Methods in Ecology and Evolution, and released as open source with AIMS.
-
2026
Real-time underwater depth paper in Computer Vision and Image Understanding.
-
2026
Launched Ask my papers, an AI assistant for my research.
Publications
All papers
20+ papers in journals such as Scientific Reports, Expert Systems with Applications, Computers and Electronics in Agriculture and Methods in Ecology and Evolution, and at ECCV.
Education
- PhD, James Cook University, 2023
Novel deep learning architectures for marine and aquaculture applications - MPhil, James Cook University, 2020
Developing deep learning methods for aquaculture applications
Methods
- Deep learning
- Self- & semi-supervised learning
- Segmentation & tracking
- Depth estimation
- Vision-language models
- Real-time edge deployment
- Python
Let's work together
I'm open to research collaborations and industry projects in applied computer vision, especially for marine, agriculture and field robotics.