Enhancing clariid catfish species classification: A deep learning framework utilizing cranial morphology and explainable AI

Enhancing clariid catfish species classification: A deep learning framework utilizing cranial morphology and explainable AI

The accurate classification of fish species is vital for biodiversity monitoring and aquaculture management. In this study, a deep learning-based framework is proposed to classify bighead catfish (Clarias macrocephalus), North African catfish (Clarias gariepinus), and their F1 hybrids using cranial...

Full description

Saved in:
Bibliographic Details
Main Authors: Kriengsak Treeprapin, Rakdee Bandatang, Thitipong Panthum, Worapong Singchat, Jiraboon Prasanpan, Kornsorn Srikulnath, Suchin Trirongjitmoah
Format: Article
Language:English
Published: Elsevier 2025-12-01
Series:Smart Agricultural Technology
Subjects:
Deep learning
Catfish classification
Cranial segmentation
Class imbalance
Explainable AI
Online Access:http://www.sciencedirect.com/science/article/pii/S2772375525003971
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The accurate classification of fish species is vital for biodiversity monitoring and aquaculture management. In this study, a deep learning-based framework is proposed to classify bighead catfish (Clarias macrocephalus), North African catfish (Clarias gariepinus), and their F1 hybrids using cranial morphological features. Eliminating the need for full-body analysis, this approach prioritizes head-based classification through YOLOv11 cranial segmentation, deep learning models (e.g., ResNet18), and dataset balancing techniques (e.g., augmentation) to enhance accuracy. The highest accuracy (99.517%) and F1-score (99.551%) were achieved using ResNet18, which demonstrated superior feature extraction capabilities and resilience to class imbalances. Classification reliability was ensured by genetic validation, while gradient-weighted class activation mapping (Grad-CAM) visualizations confirmed the significance of cranial features, particularly in the occipital region. The proposed framework demonstrates improved classification accuracy, offering a scalable and reliable alternative to conventional taxonomy. Future work could explore hybrid convolutional neural network (CNN) and vision transformers (ViT) models to optimize performance and efficiency.
ISSN:2772-3755

Similar Items