The Butterfly House – Lepidoptera Detection with Neural Networks (CNNs)

The Butterfly House – Lepidoptera Detection with Neural Networks (CNNs)

Introduction

In recent years, the UK Butterfly Monitoring scheme have recognised that there is a decline in native Butterfly and Moth populations. This decline has been associated with the detrimental effects of climate change, especially the irregular seasonal period, fluctuating temperature and humidity that many species rely on.

Every year, the England and Wales Butterfly Conservation charity host a “Big Butterfly Count” event where the public is encouraged to log sightings of Butterfly species.

The current method of identification relies on manually checking the appearance of the butterfly against a chart. This method is slow and often ineffective as it relies on the personal judgement of the individual.

Smartphone cameras are becoming ever more powerful, being able to capture high definition images that rarely blur. Another advantage of smartphone cameras is their availability due to many people always carrying them.

Convolutional Neural Networks have proven to perform classification tasks with great accuracy, permitting that they have been provided with enough training data. By using a convolutional neural network, training data captured from the web and test images captured on mobile devices, a more effective solution for classifying species of butterflies can be developed.

This project aims to develop a service dedicated to the classification of Butterfly species in the UK with the hope of improving surveying efficiency. This project is called “The Butterfly House”.

Solution Overview

Project URL

https://thebutterflyhou.se