Wild Bird Species Identification Using ResNeXt50 Feature Extraction and Random Forest Classification

Abstract

Accurate recognition of wild bird species from audio signals is essential for ecological monitoring and biodiversity conservation. This paper presents a hybrid ensemble-based system for bird species identification, aiming to improve recognition performance and robustness in real-world environments. The proposed approach utilizes a pre-trained ResNeXt50 deep learning model for feature extraction, followed by a Random Forest classifier for final classification. By combining deep feature learning with ensemble learning, the system achieves improved accuracy, robustness, and generalization across diverse field conditions. Furthermore, a Flask-based web interface with secure authentication is developed to facilitate system testing and performance analysis. Experimental results demonstrate that the proposed hybrid ensemble model outperforms standalone lightweight CNN models and is effective for real-time bird species recognition