Exploring an Image Classification Project

Jason van Hamond, a first-year Applied Data Science and Artificial Intelligence student, built an image classification system that tells cuttlefish, squid and octopus apart. The aim was overfishing: accurate species identification through a mobile scanning app.

The model reached 90% accuracy. Human participants attempting the same task managed 50%.

How he got there

He started with research into the problem and a stakeholder analysis, to check an AI solution was actually warranted. Training data came from scraping images online, followed by classification reviews to keep bias down. Peer feedback from other students mattered for making the app usable by people with very different levels of expertise.

The model was developed iteratively - starting simple, then progressively adding features such as tentacle shape and body form, testing and correcting as he went.

The final deliverables were the code, research documentation, model analysis and a presentation of the methodology and findings. He is proudest of “the successful implementation of my model and the valuable lessons learned.”