Insect Image Classification

Abstract

Insects play diverse roles in our environment, with some being harmful by damaging crops, spreading diseases, or causing poisoning, while others, like bees, are crucial for pollination and ecosystem balance. Accurately classifying insects is vital to understanding their impact and managing their effects on human activities and natural ecosystems. This paper explores the use of Convolutional Neural Networks (CNNs) for the classification of insect images. Specifically, we aim to determine whether a CNN can effectively identify whether an insect is harmful or beneficial based solely on its image.

Objective

The primary objective of this research is to develop and validate a Convolutional Neural Network model capable of accurately classifying images of insects into specific categories, such as harmful or beneficial, based on their visual characteristics. This involves:

• Analyzing and processing a large dataset of insect images from the Natural History Museum, London.

• Utilizing the power of CNNs to learn and recognize patterns and features specific to different insect species.

• Evaluating the model's accuracy and effectiveness in classifying insects, which could have significant implications for environmental management, agriculture, and biological research.

Ultimately, this study aims to contribute to the field of automated image classification in entomology, providing a tool that can assist in the rapid and accurate identification of insect species, thereby enhancing our understanding and management of these crucial components of our ecosystem.

Dataset

The dataset used in this study comprises high-resolution images of insect specimens from the British carabids collection at the Natural History Museum, London. It includes a vast array of 63,364 specimens across 291 species. Each species is organized into a specific folder, labeled with the GBIF (Global Biodiversity Information Facility) number corresponding to that species. For instance, the species 'Carabus problematicus' is stored in the folder labeled '4470555'.

Dataset Source: Insect Identification from Habitus Images on Kaggle

Number of Species: 291

Total Specimens: 63,364

The GBIF is a global initiative funded by governments worldwide. It provides open access to data about all life forms on Earth, thereby supporting a wide range of biological and environmental research.

Column Descriptions for Dataset

The dataset is structured into the following columns:

insect_gbif: The GBIF identification number for the insect species.

path_img: The file path where the insect image is stored.

file_name: The name of the image file.

Colab Notebook

PDF