International Journal of Scientific Engineering and Technology

Luis Pina ¹, Leelavathi Rajamanickam ¹, S. C. Ng ¹

Affiliations
1 School of Information Technology, SEGi University, MY

Abstract

To date, photograph identification systems for individual marine turtles are focused on the facial profile, and scute patterns on the top of the head, and neck. However, to the best of knowledge, there seems to be no photograph identification system focused on recognising marine turtle species based on characteristics of the carapace. Studies argued that by including more features, such as characteristics of the shell, the systems could enhance its classification accuracy. However, previous works have failed to address why none of them used characteristics of the shell for identifying marine turtle species. In this research, a comprehensive study of the effectiveness of the features extracted, colour, shape, and texture, from the carapace is conducted. Several experiments are carried out using the data extracted to find out the suitable data dimensionality, and the "best" hyper-parameters to train the neural networks. The expectation of this research is that these features can be used to develop a non-intrusive automated system for pattern recognition of marine turtle species using the characteristics of the carapace.

Keywords

Gray Level Co-Occurrence Matrix, HSV, RGB, Seven Invariant Moment, Neural Network.

Full Text

   

Luis Pina ¹, Leelavathi Rajamanickam ¹, S. C. Ng ¹

Affiliations
1 School of Information Technology, SEGi University, MY

Abstract

Biologists use a manual sea turtle identification key technique to classify marine turtle species according to their scutes patterns. However, limited research to date has focused on developing a system for recognising marine turtle species. Studies in the field of photograph identification system for individual animals have failed to address why none of them developed a system to categorise marine turtle species. The aim of this research is to develop a hybrid model for pattern recognition of marine turtle species based on the stacked generalisation. The hybrid model consists of two major modules:combination unit, which is the combination of the outcomes of neural network model and C4.5 decision tree model, and meta-learning that uses the neural network to aggregate the results from the combination unit and increases the accuracy of the total classification. Several experiments are carried out, where different parameters influencing the overall performance of modules are investigated. The results showed that the trial-error-test could be used to improve the computational cost and mean absolute error of the stacked generalisation when neural networks are used in both combination unit and meta-learning. Therefore, it can be concluded that the hybrid model is an improvement over the traditional manual method for categorization of marine turtle species.

Keywords

C4.5 Decision Tree, Ensemble Classifiers, Marine Turtle Species, Neural Network, Stacked Generalisation.

Full Text