Cephalopod species identification using integrated analysis of machine learning and deep learning approaches

Area of Science:

• Marine biology
• Computational biology
• Taxonomy

Background:

• Limited taxonomic research on Malaysian cephalopods despite their ecological and commercial importance.
• Cephalopod beaks offer a reliable alternative for species identification due to their durable nature.
• Traditional identification methods are time-consuming.

Purpose of the Study:

• To develop an automated model for identifying cephalopod species using beak images.
• To compare the effectiveness of traditional morphometric features versus deep features for classification.
• To evaluate different machine learning models for cephalopod identification.

Main Methods:

• Collected 174 samples of seven cephalopod species from Peninsular Malaysia.
• Extracted and imaged upper and lower beaks.
• Extracted traditional features (HOG, MSD) and deep features (VGG19, InceptionV3, Resnet50).
• Applied eight machine learning approaches for classification.

Main Results:

• The Artificial Neural Network (ANN) model achieved the highest accuracy (91.14%) using deep features from lower beak images with the VGG19 model.
• Deep features outperformed traditional features in distinguishing species.
• Lower beak images provided more accurate identification than upper beak images.

Conclusions:

• Automated identification using deep features from cephalopod beaks is highly effective.
• Lower beak morphology offers more distinct species-specific characteristics.
• Future research should expand species and sample size for improved model accuracy.