This AI algorithm counts flowers on trees to predict crop yields months in advance

Scientists have trained an AI algorithm to count the number of flowers on fruit trees using only smartphone images. The system could predict the size of a harvest months in advance — saving farmers time, money, and water.  

Researchers from Chile, Spain, and the UK’s National Robotarium developed the algorithm. The Robotarium is a centre for robotics and AI based in Edinburgh, Scotland. The team looks to address a global problem.

“In countries all across the world, farmers often rely on manual methods to estimate their yields, which can have a significant error margin,” said lead researcher Fernando Auat Cheein, an associate professor at the National Robotarium.  

Agriculture uses around 70% of the world’s freshwater and wastes almost half of it. Around 50% of the fruit and vegetables grown for human consumption are also lost. 

The flower-counting AI could reduce this waste. It could also boost farmers’ yields by improving their allocation of water and fertiliser. 

“By focusing their efforts on areas of the farm that are expected to yield the most fruit, farmers can optimise resources, reduce their environmental footprint, and maximise both the quantity and quality of their harvest,” said Cheenin.

The team has already tested the AI at an orchard in Spain. It predicted flower counts there with 90% accuracy. Manual counts typically range from 50% to 70%, for comparison. The system recognises the unique patterns, shapes and colours of flowers even when they overlap or are partially obscured. 

In September, the researchers will validate the AI’s predictions against the real peach harvest. If the approach proves effective, they believe it could be adapted for other important crops such as apples, pears, and cherries. 

“The principles behind this technology could be applied to a wide range of fruit crops worldwide, including those grown in the UK,” said Cheein. The system arrives in rapidly changing times for agriculture.

Farming is one of the world’s oldest industries, but also one of the least efficient. Increasingly, however, farmers are adopting technologies like AI, drones, and robots in a bid to make their jobs easier, greener, and more profitable.  

For instance, German startup Constellr has developed a satellite-based crop monitoring system that helps farmers identify crop damage days or even weeks in advance. Another startup, Belgium’s Robovision, has created a platform for managing AI-based vision systems on farms.