About This Project

I think we have the technology to develop an artificial intelligence model that can differentiate between Dinoflagellates by simply examining their images. I will generate a database of 64 Dinoflagellate species, with 2000 images of each species. These images will feed an artificial neural network to train it to classify each species. My aim is to develop an image database of Dinoflagellate species, as well as a trained neural network that classifies them with greater than 99% accuracy.

Ask the Scientists

What is the context of this research?

Dinoflagellates are photosynthesizing organisms with 2,294 species. They are important to our environment because most living organisms rely heavily on them.

Machine learning is part of the larger artificial intelligence field. The field is rapidly growing and one of its most useful structures is the artificial neural network. However, machine learning requires large data to produce a useful neural network.

I hypothesize that by creating a data set of 64 Dinoflagellate species each with 2,000 images, I can feed this data set to an artificial neural network to produce a Dinoflagellate image classifier with above 99% accuracy. I will train our neural network then test it by showing it images of Dinoflagellates and checking to see if the AI correctly identifies the species.

What is the significance of this project?

If I successfully produce a neural network that can correctly identify Dinoflagellates from an image this solution could save time and money by replacing the traditional method of sending live or stained samples to specific database vendors where humans identify the species. A new modern tool can automate this task by not only giving scientists, like marine and environmental biologists, the ability to identify Dinoflagellates from samples by simply taking a microscope photo, but also identify different species in a mixed sample and count the different species.

Additionally, this proof of concept that a neural network is capable of classifying organism with minimal visual information can be expanded to automate other types of labor intensive cell identification in the lab and the field.

What are the goals of the project?

The ultimate goal of this project is to develop a database of images of 64 different Dinoflagellate species (2000 images per species) then feed this database into an artificial neural network to train it.

The hypothesis is, using this database I can develop an artificial neural network that classifies Dinoflagellates by species, and if given an image with mixed cells it would still identify each species with 99% accuracy.

To test it, I will run the database on an object detection neural network which will train on a training set (usually 75% of the images divided randomly). The training outcome is then tested on a testing set (usually 25% of the images), and this gives us the accuracy of the training (how much of the images are correctly classified).