Artificial intelligence could soon work with humans to grow the right strawberry.
Researchers at Western University have developed an AI system that goals to revolutionize the cultivation of certainly one of the world's hottest fruits – with potential ripple effects throughout the agricultural sector.
And no, this has nothing to do with OpenAI's O1 model, formerly codenamed “Project Strawberry”.
The studypublished within the journal Foods, represents a remarkable advance in agricultural technology.
Using advanced machine learning techniques, the team has developed a system that detects strawberry ripeness and diseases with almost 99% accuracy – all through easy camera monitoring.
“We wanted to scale back the dimensions of those AI models to make them actionable for farmers and native production,” said Joshua Pearce, John M. Thompson Chair in Information Technology and Innovation at Western Engineering and Ivey Business School.
“We not only wanted to extend the accuracy, which is over 98 percent, but in addition reduce the dimensions of the models.”
What sets this research apart is its give attention to accessibility. Unlike many high-tech solutions for agriculture which can be aimed toward large-scale operations, Pearce and his colleague Soodeh Nikan designed their system for small and medium-sized farms.
The team’s methodology combined revolutionary AI techniques with practical agricultural knowledge:
- They began by collecting different images of strawberries, including healthy fruits and people affected by various diseases.
- These images were then processed and augmented to create a sturdy training dataset.
- The researchers fine-tuned three different AI models – Vision Transformer, MobileNetV2 and ResNet18 – each of which brings unique strengths to the duty.
- To be sure that the AI can handle the variability of the actual world, they integrated techniques akin to class weighting and artificial image generation.
- Perhaps most significantly, they incorporated “attention mechanisms” into the models that allowed the AI to give attention to essentially the most relevant parts of every image.
The system is characterised by two predominant tasks:
- Maturity detection: It can accurately classify strawberries as ripe or unripe and helps farmers optimize the harvest timing.
- Disease identification: The AI can detect and discover seven various kinds of strawberry diseases: angular leaf spot, anthracnose fruit rot, blossom rot, gray mold, leaf spot, powdery mildew on fruits and powdery mildew on leaves.
The results speak for themselves. With accuracy rates of around 98%, the system far surpasses previous attempts at automatic strawberry monitoring.
However, the impact of this research goes far beyond simply improving strawberry yields.
The potential to scale back food waste can also be obvious. According to the Food and Agriculture Organization of the United Nations approx. 14% of the food produced is lost between harvest and sale.
Technologies like this AI system could help solve this problem by optimizing harvest timing and reducing losses resulting from disease or overripeness.
“Reducing waste and lowering food costs is clearly an enormous issue lately. Like everyone else, I'm at all times surprised when I am going to the supermarket and see the costs of fresh fruit and vegetables,” Nikan said.
“When choosing projects, I normally search for something that’s safety critical or represents a societal need. With my experience in other applications, I jumped at the possibility to use my knowledge and expertise in food safety.”
In the longer term, the team is already planning to check its system outdoors, possibly using drones for more comprehensive field monitoring.
They are also investigating the usage of AI-generated synthetic images to further reduce the info requirements for training effective models.
“Instead of taking pictures of hundreds of thousands of strawberries, which is a less efficient and expensive approach, we use synthetic images and open source software to create hundreds of thousands of images ourselves using relatively little computer power. So we are actually in a position to make highly detailed observations about maturity and disease for very specific plants,” Nikan said.
Pearce added: “The software is totally free and open source, and farmers of every kind can download it totally free after which customise it to suit their needs. They might want the AI system to email them or ping their phone when it detects a disease, and even send them an image of a specific plant ready for harvest. The software is totally open to make it your individual.”
