In 2021, BlueTrend established Taiwan's Ocean Citizen Scientist marine life database platform, collecting photos of marine life from across Taiwan. In 2022, the team consolidated user feedback and carried out the platform's first major overhaul — redesigning the database interface to be more intuitive, streamlining the upload process, and introducing a community identification feature to put the spirit of citizen science into practice. In 2023, the revamped database officially launched, and with the support of various events, the database had accumulated over ten thousand photos by year's end.
To ensure the accuracy of all information, every identification record is reviewed and confirmed by a biologist. The surge in data submissions also significantly increased the manpower required for photo identification. Against this backdrop, the idea of integrating AI identification emerged — with the hope that AI could help ease the burden on human identifiers. In 2023, the team participated in the Taiwan Mobile Foundation's Tech The Dreamers programme and secured funding to begin planning a technology system for AI-powered marine life photo identification. (Truth be told, we secretly sent Blu off to training camp!)
The Challenge of Obtaining Marine Life Data
AI image recognition is already a highly mature technology, yet resources for identifying marine life remain far scarcer than those for land animals. The primary reason is that marine research is considerably more difficult than research conducted in terrestrial environments. Due to the technical and cost constraints of ocean exploration, relevant biological data is comparatively limited — and this directly affects the building of databases needed to train AI models.
The team encountered the same challenge when training "AI Blu." Effective machine learning and AI development depend on vast amounts of data to train a model; each species requires at least 500–1,000 photos to achieve meaningful training results. However, marine life photos are severely lacking in quantity, and photo identification also requires very distinct features to work accurately — making the development of an AI marine life photo identification system all the more difficult. At present, "AI Blu's" knowledge is still at a relatively early stage, so expanding the collection of marine life photos is the most important priority right now.
Underwater photography demands far greater technical skill and cost than photography on land.
Introducing the AI Ocean Identification System
Many thanks to the Taiwan Mobile Foundation for their financial support — Blu has finally enrolled in marine life school and begun official training!
When selecting species to start with, the team naturally chose nudibranchs first, since the camera rolls of every macro photography enthusiast are bound to be filled with sharp, vivid nudibranch shots! Even so, the data-collection process was no easy feat. Blu has currently learned to identify over 20 species of nudibranch, with 100–300 training photos per species — and those 20-odd species alone required more than 3,000 photos. (The Editor is quietly wishing everyone would upload more photos!)
In addition, Blu has also been learning about echinoderms, gastropods and bivalves, molluscs, and the sea turtles most commonly spotted around Xiaoliuqiu. The selection criteria focus mainly on marine life that is photogenic and serves as a good indicator species. Blu has now learned to identify 80 species of marine life! The goal is to reach 150 species by the end of 2024, at which point we also hope to start expanding into fish identification.
Did you spot the AI-exclusive version of Blu?
Decoding the AI Identification Logic
For those of you who have already had AI Blu identify something for you — has anyone wondered what the percentage scores it gives you actually mean? Here's an explanation of how the AI identification logic works.
We group submissions by broad category — that is, the broad category you fill in during the upload process. From there, labels are assigned within each group, photos are fed to each label, and AI Blu begins learning. So when you're uploading, please make sure you don't select the wrong category!
Next comes the computation. Because Blu learns from a large number of different photos rather than repeating training on the same image, it searches within the broad category you've selected and returns the closest match it can find. Since the training database is not yet robust enough for AI Blu to become a true identification master, the road ahead is still long. To help with this, the team asked our engineers to add a button on the front end so that, if AI Blu makes an incorrect identification, users can step in and correct it.
AI Blu will show you the likelihood of a species identification — the final call is yours!
Keeping the Database at Maximum Accuracy
If you've made it this far through all the technical content, give yourself a well-deserved round of applause — three claps, please!
Many of you might be thinking: "I don't know these species to begin with — how would I know if AI Blu got it wrong?" Rest assured, before any data is exported or used, the team will always have professional taxonomists review and verify the records. We also conduct periodic checks of the database to look for any photos that may have been filed in the wrong place.
If you think AI Blu has made a wrong identification, go ahead and enter the correct species to override it!
Thank you so much to all our users for your support throughout this journey — your contributions have allowed the database to advance so quickly. We hope everyone will keep uploading photos; the Editor will be feeding them straight into AI Blu's training dataset so it can keep hitting the books and improving its skills!
Once again, a heartfelt thank you to the Taiwan Mobile Foundation's Tech The Dreamers programme for enabling the team to build AI Blu and bring the database to the next level.
AI Ocean — where technology and citizen science meet marine research, and together we promise Taiwan's oceans a biodiverse future.
Related Links
Further Reading
