Semi-supervised and unsupervised method to increased database labels in the case of classes imbalances

The project aims to improve the amount of labelled samples in a semi-automatic or automatic manner using AI to impove a CNN performance. We will test various state-of-the-art AI methods, in the context of forest inventory, and select the most effective ones.

The benefits will be significant because labelling is an important but tedious task, in many cases, when working with natural forests, some tree species will not occur as often as others (hence creating a shortage in some classes), also there can be co-species to many other species and they are difficult to identify clearly.

Anirudha Jitani
Faculty Supervisor: 
Blake Richards
Partner University: