Future crops

[CSIRO logo appears on screen.]

[Timelapse shot of a busy city intersection with traffic] Voiceover: Busy roads are filled with visual and auditory cues that we need to perceive, analyse and understand to make safe decisions.

[Computer graphic of a computer processing data] Artificial intelligence and computer vision have been harnessed to engineer self-driving cars.

[Computer graphic of cars sensing each other as they travel down a road] These can autonomously make sense of their surroundings by detecting and identifying objects to guide decisions.

[Drone shot of a large cotton field] Breeding better crop varieties also requires many observations to be made to select the top of the crop.

[Shot of a cotton breeder inspecting cotton in a field] And a lot of these are still done manually, by a number of different people, which presents a number of limitations.

[Scene of a CSIRO researcher looking down a microscope, followed by a data scientist coding at a computer] At CSIRO, we have combined our expertise in crop breeding, microimaging, and computer vision to build HairNet, a machine learning-powered solution to score leaf hairiness.

[Shot of a glasshouse with cotton, followed by a person in a cotton field] In Cotton, leaf hairiness is a key indicator of fibre yield, its economical value and the ability of the plant to resist certain insects.

[Shot of a person looking at the hairs on a cotton leaf, and two people in a glasshouse looking at cotton leaves] Each year, breeders visually inspect and score leaf hairiness to select the best lines. But different people may give the same plant a slightly different score.

[Shot of cotton plants in a glasshouse] And this test relies on the reflection of sunlight on the surface of the leaf so it can only be done on cloudless days.

[Scene of a cotton leaf being put under an imaging system, with the lead appearing magnified on a computer screen] We are using a simple imaging system to generate images of leaves. These images are then fed to our HairNet model which can predict leaf hairiness with 95% accuracy, without operator bias and regardless of the weather!

[Sunset scene of three cotton harvesters working in a field] This new tool will give cotton breeders robust measurements on any day, which is important in the middle of a busy breeding season.

[Shot of cotton shirts blowing in the wind on a clothesline, followed by a drone shot of a cotton field] Beyond Cotton, a similar approach could be used to score a range of other important characteristics such as how well plants can utilise water, or resist disease.

[Shot of a farmer standing in the field at sunset] With tools like HairNet we are building a future where artificial intelligence assists humans make better decisions to breed the future crops we need.