Overview / Usage

Aquatic macroinvertebrate biomonitoring is one way to assess water quality. The identification of macroinvertebrates by human experts is costly, time consuming and is becoming a scarce resource. Furthermore, human identifications have repeatedly been shown to be less reliable than assumed. The problem that we are trying to solve is to develop deep learning models to reach and surpass human level identification accuracy.

Methodology / Approach

In this project, we compare the performance of different classifiers such as SVM, kNN, random forest, ridge regression, etc. by extracting features from the images of the macroinvertebrates. Afterwards, we explore whether deep learning networks like VGG16, VGG19 and Inception v3 can improve image the classification task. We will use scikit-learn for classifical machine learning algorithms and Keras for deep learning models.

Technologies Used

Intel DevCloud, Python, Anaconda, Keras, Matplotlib, scikit-learn