<![CDATA[Monitoring and identification of bird species is an important aspect of biodiversity conservation, but manual identification methods based on direct observation and expert listening still have limitations in terms of time, cost, and subjectivity. Other challenges arise due to variations in the quality of sound recordings, the presence of environmental noise, and the similarity in vocalization patterns between species that make it difficult to automate the classification process. This study aims to develop an automatic classification system of bird species based on acoustic signals by combining the YouTube Audio Event Network (YAMNet) model and the Random Forest algorithm. YAMNet is utilized to extract spectral log-Mel features that represent the frequency and temporal characteristics of bird sounds, while Random Forest is used as a classifier to determine species based on those features. The dataset used is the Sound of 114 Species of Birds till 2022, which includes species variation, recording duration, and complex acoustic conditions. The results showed that the features produced by YAMNet were able to form separation between species visually through Principal Component Analysis (PCA), although there was still overlap in species with similar vocalization characteristics. Evaluation using the confusion matrix shows that some species can be classified with a high degree of accuracy, while misclassification occurs mainly in species with similar frequency patterns. Receiver Operating Characteristic (ROC) analysis yields Area Under Curve (AUC) values of up to 0.98 in certain species, indicating the model's excellent discriminating ability. These findings suggest that the integration of YAMNet and Random Forest has the potential to be an efficient and reliable solution to support automated bird species identification systems in nature conservation.]]>
Copyrights © 2026
