Snake Species Classification Using Transfer Learning Technique

EasyChair Preprint 7214

Abstract

Transfer learning is a technique that helps to utilize the knowledge of previously trained machine learning models by extending them to solve any related problem. This technique is predominantly used when there is either a scarcity of computational resource or limited availability of labelled data. Categorizing snake at the species level can be instrumental in treatment of snake bites and clinical management.We propose a deep learning model based on transfer learning technique to build a snake species classifier that uses snake photographic images in combination with their geographic location. We have used the Inception Resnet V2 as a feature extractor, extracted the feature vector for each input image and concatenated it with geographic feature information. The concatenated features are classified using a lightweight gradient boost classifier. We have achieved a training accuracy of 71.32%, validation accuracy of 44.16% and a testing accuracy of 42.96%.

Keyphrases: Gradient Boosting, Inception-ResNet, Metadata Inclusion, Snake Species Classification, Transfer Learning

@booklet{EasyChair:7214,
  author    = {Karthik Desingu and Mirunalini Palaniappan and Jitesh Kumar},
  title     = {Snake Species Classification Using Transfer Learning Technique},
  howpublished = {EasyChair Preprint 7214},
  year      = {EasyChair, 2021}}