CSIRO is a renowned institution at the forefront of harnessing the transformative power of artificial intelligence (AI) across diverse scientific domains. As a pivotal player in the realm of research, CSIRO has played a central role in exploring and responsibly integrating AI into its scientific endeavours.

One of the standout applications of AI by CSIRO has been in addressing the inherent biological limitations that often hinder scientific exploration. Notably, the human eye’s limited capability to discern objects smaller than approximately 0.2mm has posed considerable challenges in scientific research.

In response to this persistent limitation, CSIRO is developing cutting-edge AI models that excel in quantifying and identifying objects of interest with unparalleled accuracy. These advancements underscore CSIRO’s unwavering commitment to leveraging AI for the sake of scientific discovery.

Within the sphere of agriculture, CSIRO has actively collaborated with industry experts to create AI models specifically designed to quantify the number of trichomes on the surface of cotton leaves. Trichome density plays a pivotal role in influencing insect resistance, fibre yield, and the overall value of cotton varieties. Through meticulous research and development efforts, CSIRO has not only succeeded in automating the assessment of trichome density but has also introduced novel AI-augmented scoring methods that significantly enhance the reliability and accuracy of this assessment.

An example of this innovation is the HairNet2 model. It represents a quantum leap in the automated assessment of trichome density. Rather than merely replicating human assessments, HairNet2 introduces a groundbreaking AI-augmented scoring approach. This approach focuses on estimating the area of the leaf-covered by trichomes by meticulously identifying every trichome on the leaf’s surface. While this task is not entirely insurmountable for a human, it is incredibly labour-intensive and time-consuming. HairNet2’s development process involved training the model using a database of around 1000 images, each annotated meticulously by human experts to identify every single trichome.

This arduous annotation process paved the way for the creation of an AI tool that can not only replicate human assessments but, more importantly, automate trichome density scoring beyond human capabilities. As a testament to its practicality, these newly developed models are now in the process of being integrated into a web interface accessible to breeders, offering them a streamlined and precise tool to assess trichome density during the forthcoming cotton season.

In the realm of environmental science, CSIRO has extended its AI applications to address the complex challenge of identifying harmful algal blooms. These blooms consist of large populations of toxic algae, posing significant risks to both human and animal health. Identifying and quantifying these harmful algal blooms traditionally involved extensive manual labour with the use of microscopes and counting chambers.

This approach, in addition to being labour-intensive, carries the risk of adverse health effects for scientists who spend extended periods at the microscope. In recognition of these challenges, CSIRO embarked on a mission to harness the capabilities of machine learning models to automate the detection of harmful algae in images.

To achieve this, CSIRO’s team has systematically captured images and annotated various algae strains from the Australian National Algae Culture Collection. This painstaking effort is supported by a range of AI tools aimed at expediting the annotation process.

The collaborative endeavour has resulted in the creation of a substantial annotated dataset comprising 15 different algae strains. Early testing of AI models utilising this dataset has showcased a remarkable ability to detect target strains with a high degree of accuracy. The implications of this development are profound, as it promises faster and more accurate detection of toxic algae, with far-reaching economic, environmental, and social impacts.

Improved and expedited harmful algae detection through AI can serve as an early warning system for water managers, offering insights into when and where blooms might occur. The implications ripple across multiple sectors, including environmental protection, coastal community well-being, consumer safety, and the sustainability of Australian fisheries and aquaculture businesses. The fusion of AI and environmental science has the potential to yield far-reaching benefits for both ecosystems and communities.

Moreover, CSIRO recognises that the potential of AI extends beyond the boundaries of research and scientific exploration. Many organisations across various industries are currently grappling with the possibilities and challenges that AI brings. CSIRO emphasises that the entry barrier for implementing AI for object detection has significantly lowered over the years, making it accessible to organisations with modest datasets. Dr Moshiur Farazi, an expert in computer vision, underlines the importance of asking the right questions and preparing data effectively to unlock the full potential of AI in addressing complex challenges.