Successful Application of Deep Learning to Biosecurity Surveillance: A Systematic Approach Using the Asian House...

AI-based solutions offer the potential for rapid taxonomic identification of species of biosecurity concern, enhanced global accessibility and time-saving in contrast to traditional taxonomic identification by humans. This study provides a systematic approach to the application of deep learning for biosecurity surveillance, using the Asian House Gecko (AHG), Hemidactylus frenatus, Schlegel, 1836, as a case study. An effective triage tool for rapid initial identification of this invasive species was developed using machine learning, achieving high accuracy. This demonstrates the efficacy of deep learning for identifying complex morphological characteristics. The AI model used the AHG's head as a key identifying feature, highlighting the importance of specific morphological features for effective identification of target species. A structured approach for the use of machine learning was developed, which included the collation of source images, cataloguing, tagging, naming and storing images, validating and uploading images, labelling images, creating, training and deploying the model, testing model accuracy and retraining the model. This procedure allows for more rapid application of the methodology in biosecurity surveillance. The structured methodology developed can be applied to similar AI-based projects. Outcomes of this research have the potential to reduce the time delays associated with taxonomic identification of invasive species, allowing follow-up action to occur sooner. Reducing time delays is critical to implementing effective biosecurity measures.