Combining Faster Region Convolutional Neural Network and Self Organizing Map for Segmentation and Image Extraction of Toraja Buffalo

Abstract

This study aims to develop a hybrid approach for segmenting and extracting Toraja buffalo images using Faster Region Convolutional Neural Network (Faster R-CNN) and Self Organizing Map (SOM). The objective is to accurately classify two buffalo types—Saleko and Bonga—by combining deep learning and unsupervised clustering. The research utilizes a dataset of Toraja buffalo images collected in various lighting conditions. Faster R-CNN is employed to detect regions of interest, followed by SOM to segment and cluster buffalo features. Evaluation using mean Average Precision (mAP) and confusion matrix shows that the model achieves a mAP50 of 0.986 and overall classification accuracy of 97.7%. Compared to conventional methods, the proposed approach demonstrates improved segmentation quality and robustness in diverse conditions. The novelty of this work lies in the effective integration of Faster R-CNN and SOM for livestock identification in a culturally significant context. (Abstract)

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint