Ramping up AI-driven crop disease monitoring

AI crop — Harvesting olives on the Greek island of Zakynthos. The ‘Vision Transformer’ (ViT) is currently being developed privately and is at the proof-of-concept stage with an olive-farming client. Photo: Mark Pasveer

Progress, benefits and potential risks: an exclusive interview with Osledy Bazó, founder of the AI firm AlamedaDev in Spain. “It is fundamentally a tool that supports, rather than replaces, the skilled labour in agriculture.”

AI vision systems are revolutionising crop management in several ways. In the field of disease detection, vision systems along with new leaf health sensor systems will soon enable AI to detect signs of infection before they are apparent to the human eye. AI crop disease detection systems can also alert the farmer at a point when a set ‘infection threshold’ is met, automatically recommending the right crop fungicide or other product to be sprayed when the AI determines weather conditions are optimal.

For further insights, we interviewed Osledy Bazó, Founder and CTO of AI firm AlamedaDev in Barcelona.

What can AI disease detection do for crop farmers?

“This feature primarily enhances the efficiency of disease detection in crops by assisting in the rapid and automated classification. But this is not just about recognising problems that the human eye might miss. It is about speeding up the labourious task of identifying and documenting various aspects of crop health issues. The AI system records the detection date, type of disease or pest, potential impact on the crop and the spread rate. This data is crucial for prompt actions to prevent further contamination or spread to adjacent crops.

Moreover, the AI technology contributes significantly to reducing manual scouting time. By automating the detection and recording processes, it allows growers to focus on other critical tasks, saving valuable time and resources. The rapid detection capabilities ensure that diseases are managed before they become more severe, facilitating quicker responses such as the application of pesticides or fungicides.”

Can these capabilities be tied in directly with spraying systems so that a farm’s AI completely oversees crop disease management? 

“Yes, this is a potential development for future phases. An advancement like this would enable AI to directly control drones to apply treatments precisely where and when needed, based on the severity thresholds established.”

Tell us about what the first season would be like for a grower using an AI crop disease detection system like yours. How will it work in terms of scale, learning curve, capital costs, etc.? 

“The experience would be very customised to a grower’s specific operational processes and needs. The financial viability and scale of the technology deployment would largely depend on the specific crops and the extent of the area under cultivation.

Our role includes helping clients identify and prioritise their needs based on realistic assessments, rather than pursuing overly ambitious or broad implementations from the start.

Time commitment for training and proficiency with the system can vary significantly based on the scope of deployment and the existing technical proficiency within the client’s team. Our approach is to streamline this process, ensuring that clients can efficiently manage the tech with minimal disruption to their existing operations.”

Training for personnel is another critical component, aimed at achieving a high level of accuracy with the AI system before full-scale implementation

At this point, what else should growers know about this technology?

“The initial steps toward integrating an AI disease detection system involve establishing clear priorities and operational parameters tailored to the client’s specific agricultural context. Developing the technology to suit particular crops — recognising the differences, for instance, between olives, sunflowers, and tulips — is crucial. This customisation extends to training the AI model with real crop data to ensure it can accurately detect and classify disease conditions.

Training for personnel is another critical component, aimed at achieving a high level of accuracy with the AI system before full-scale implementation. This ensures that the technology integrates effectively into existing operations and meets the necessary standards for effective disease management.

Prospective users must consider these aspects along with the anticipated costs for ongoing maintenance and potential upgrades. An example of a potential upgrade could be implementing decision support features that recommend specific actions when a disease is detected. Each element contributes to a comprehensive approach to integrating AI into their agricultural practices, ensuring readiness and efficacy from the outset.”

That brings us to the risks to be in using AI for crop disease detection. Will it be reliable enough in the near future that farmers can depend on it?

“While AI significantly increases efficiency and accuracy in identifying diseases, it is fundamentally a tool that supports, rather than replaces, the skilled labour in agriculture. It is crucial that farmers and agricultural workers continue to engage actively with the processes, verifying that the AI’s findings are integrated thoughtfully into broader farm-management practices.”

When will this type of technology be available?

“Our ViT technology is a private feature and not available for commercial sale. However, we are open to collaborating with other companies to implement a proof-of-concept in their processes.

Our initial focus is on olive crops, for which there is substantial data available, making it a prime candidate for early application. This focus aligns with our strategy to utilise crops that provide a robust dataset for training and refining our models.

As for the broader deployment and commercialisation, it is important to note that there is still a significant gap in awareness within the agricultural sector about the capabilities and benefits of AI. However, efforts are being made to encourage the adoption of new systems to enhance production processes and cost efficiency.”

Could you give us some details?

“In Europe, significant efforts are being made to promote the adoption of disruptive technologies across various sectors, including agriculture. In Spain, there is a strong push to encourage companies to integrate AI into their operations through innovative public procurement. This initiative is part of a broader strategy to enhance production processes and cost efficiency, not just in businesses but also in agriculture.

These efforts are aligned with larger European Union initiatives, such as the Next Generation EU programme. This programme aims to foster a more resilient and digital Europe, where advanced technologies play a crucial role in various industries. Through these actions, both national and EU-level institutions are actively working to support the development and integration of AI, ensuring that the agriculture sector can benefit from these technological advancements.

AlamedaDev’s progress

This AI firm works in many fields. In ag, it has developed a crop monitoring and analysis system called ‘Vision Transformer’ (ViT). ViT is currently being developed privately and is at the proof-of-concept (POC) stage with an olive-farming client in Andalusia.

The ViT AI model was trained on a vast dataset of crop leaf images, each labeled with specific disease markers. As with any computer vision system in agriculture, it can also be applied to the identification of pests and weeds, assessment of crop quality, yield prediction, soil health and moisture monitoring, post-harvest analysis, etc.

During this POC phase, field workers helped train ViT by sending it images as they manually inspected crops. “The project is progressing, focusing on validating the technology’s effectiveness through smartphone-captured images before expanding to more sophisticated data capture methods like drones,” says Osledy Bazó, AlamedaDev Founder and CTO. “It is currently developed as a proprietary feature within this client’s private enterprise system. The implementation across additional farms would require custom adaptation to the specific processes and needs of each prospective client, acknowledging the unique characteristics of their operations.”