WingsSL: Integrated Approach for the Detection and Conservation of Birds in Sri Lanka

Birds are a vital part of the ecosystem and excellent indicators of its conditions because of their responsiveness to environmental changes and ease of observation. In the presence of a perpetual tropical climate favorable to bird life, Sri Lanka is well known for its abundant avian diversity, both resident and migratory species. However, insufficient research on local bird diversity and abundance limits conservation efforts. This study focused on an integrated approach with four components to bird detection and conservation in Sri Lanka, addressing two main challenges: identifying birds along with analyzing their behavior, and understanding variations in bird populations due to environmental factors. First, the bird image identification model was developed, utilizing a transfer learning Convolutional Neural Network(CNN), trained on a comprehensive dataset of Sri Lankan bird images, to recognize birds in various postures, angles, and environmental situations. This model achieved a test accuracy rate of $\mathbf{98.04\%}$. Secondly, a CNN based bird song classification model was developed to differentiate species through their vocalizations and was achieved a test accuracy of 87.78%. A Long Short-Term Memory(LSTM) model categorized bird calls into mate attraction, alarm signaling, and territory defense which achieved a test accuracy of $\mathbf{94\%}$. Finally, the system forecast bird population trends using a K-Nearest Neighbors(KNN) algorithm, trained on historical population records, habitat details, and climate data, enabling it to predict future populations based on similar past data points. Integrating these functionalities, provided valuable insights for bird conservation efforts in Sri Lanka, facilitating informed decision-making and effective conservation strategies.