PhD: Integrating deep learning AI software with hardware for next generation acoustic biodiversity monitoring

Monitoring of biodiversity using acoustics has huge potential but is held back by a lack of access to technology and cutting-edge analytical techniques such as deep learning. Several companies provide hardware and software that allows recordings to be analyzed following collection from the field. However, hardware and software are poorly integrated, if at all. Some companies also provide access to analytical tools, but their performance is often mediocre because they cannot be customized sufficiently to achieve highly accurate results. Recent work conducted by QUT, funded by Cotton RDC through the National Landcare Program (NLP1901), has shown that accurate results can be achieved using deep learning algorithms embedded on field deployable hardware. However, the system is specific to the birds and bats chosen by the researchers and so is not easily adaptable to other species and to different environments. The project relies on existing hardware and deep learning algorithms (CNNs) to achieve its results. The aim of this project is to research, design and build custom hardware and software for the analysis of environmental sounds, and to develop and embed next generation artificial intelligence algorithms capable of operating in real time and using very little power. The project will also design and build a framework where other researchers, land managers or growers can train and deploy their own custom (and so highly accurate) acoustic recognizer algorithms and place them onto the new advanced hardware system. Such a system will revolutionize acoustic monitoring and make the hardware and software required to create operational systems widely available.