CIBIP-AI: Development of California Invasive Bird Identification and Prediction Artificial Intelligence System

Invasive species are detrimental to native wildlife and pose a major threat to biodiversity. In California alone, municipalities and conservation authorities are estimated to spend approximately $50.8 million per year on invasive species. Effective management of bio-invasions depends on reliable identifying the invasive threats and predicting areas at risk of invasion. Yet, there is no efficient solution to address the issue. This paper proposes the first California Invasive Bird Identification and Prediction AI System (CIBIP-AI). The CIBIP-AI system utilizes a deep learning solution to automatically identify and predict two California invasive birds, the mute swan and brown-headed cowbird. The CIBIP-AI system uses a hybrid deep learning architecture based on VGG16 and ResNet50 for identification. The CIBIP-AI system can also predict the geographical areas at risk of invasion based on a set of the birds' physiology features and microclimate conditions. The CIBIP-AI database is created for storing data and continuous training. The initial experimental results show that the classification accuracy for the image input reached 96%. The audio input was more challenging to classify, with 73 % accuracy. After applying knowledge-based filtering and common pattern extraction, the accuracy was improved to 77%. The video was split into multiple images and audio and input into the hybrid models. The classification accuracy of the hybrid models reached 97%. The work shows that the proposed CIBIP-AI system provides significant aid to the early detection and prevention of invasive bird species.