For small and marginal farmers, a crop disease outbreak can quickly turn into a devastating financial loss. Limited access to agricultural experts, delayed diagnosis and the difficulty of identifying diseases at an early stage can leave farmers with few options to protect their crops. Artificial intelligence and computer vision are increasingly emerging as powerful tools to address these challenges by enabling faster, more accessible and data-driven crop disease detection.

A patented AI-driven innovation developed by researchers at VNR Vignana Jyothi Institute of Engineering & Technology (VNRVJIET), Hyderabad, is seeking to address this challenge through a convolutional neural network (CNN)-based plant disease detection system. Developed for crops including tomatoes, potatoes and peppers, the technology is designed to identify common diseases from plant images and provide actionable recommendations on pesticide use.

The research is also deeply personal for Dr Vijaya Saraswati R, Department of Computer Science and Engineering, VNRVJIET, who grew up in a farming family and witnessed first-hand the impact of crop disease on farmers’ livelihoods. The system achieved 96 per cent accuracy in classifying 10 common diseases affecting nightshade crops and has been designed for deployment on affordable devices with limited computing requirements.

In this interview with AI Spectrum, Dr Vijaya Saraswati R discusses the personal experiences that inspired the research, the technical challenges involved in developing a robust AI model for real-world agricultural conditions, the role of interdisciplinary collaboration, and the technology’s potential to support precision agriculture. She also outlines plans to develop a mobile application, expand the system to additional crops, integrate drone-based diagnostics and explore edge and multimodal AI for early disease prediction.

Could you tell us about the inspiration behind this research and how your personal experiences influenced the development of this solution?

The background of this research is very close to my heart. As someone who grew up in a farmer’s family, I have seen how vulnerable small cultivators are. I remember one instance when I was about 13 years old, where my father used all three out of the total five acres of land that we own to cultivate tomatoes. Our entire crop was damaged due to an extremely rapid infection since we could not predict that in advance. The financial loss that my family incurred due to that unfortunate incident stuck with me.

As a child, I questioned why something could not be done to protect the farmers from this kind of problem. After almost two decades, now as someone with a PhD in advanced computing, I finally realize I had the power to solve this childhood dilemma through this patented technology.

What were the key technical challenges in developing a robust CNN model for accurate disease detection, and how did you address issues such as varying field conditions and image quality?

While developing a model that is accurate in a lab setting may be one thing, creating something reliable in practice is another issue altogether. For instance, while taking pictures of crops, farmers may deal with various light levels, shades, distortions from wind, soil interference in the background and various cameras’ quality. To create a truly reliable system, our interdisciplinary team paid considerable attention to developing special pre-processing algorithms. We developed new approaches to image scaling, advanced data augmentation and noise reduction. Instead of relying only on perfect images available publicly, we used huge amounts of images, about 20,000 of tomato, potato and pepper plants that were collected in the field and augmented to reflect adverse conditions. Thanks to this, the approach to extracting features by neural networks helped us to solve common problems associated with computer vision.

How does your patented system improve upon existing AI-based plant disease detection solutions in terms of accuracy, scalability, usability, and deployment in real-world farming environments?

While there are various applications that can help diagnose plant diseases, our solution incorporates a completely new and different approach to image capture and data processing, and this has helped us acquire a patent.

Accuracy: By fine-tuning our CNN architecture to detect micro structural abnormalities in plant leaves, we have managed to achieve incredible accuracy of 96% in classifying 10 of the most common diseases that occur in nightshades (including tomatoes, potatoes, and peppers).

Actionable Usability: While other systems just diagnose the problem, leaving farmers to guess what remedy to use, our system makes one very important leap ahead by automatically providing recommendations about the right kind and amount of pesticides to use. This ensures that farmers can make quick decisions, avoiding the common problem of over-chemicalization.

Scalability & Deployment: Lightweight nature of our image processing engine means that our solution demands very little computing power. Thus, we can deploy the technology smoothly on affordable mobile devices without using costly server infrastructure.

Precision agriculture is becoming increasingly important for improving food security and sustainability. How do you see AI-powered computer vision technologies transforming farming practices in India, particularly for small and marginal farmers?

In India, where most of the farmers own small/marginal lands, conventional consultations of agricultural experts tend to be either out of reach, costly, or delayed. With the outbreak of a disease, a lag of only 48 hours may result in losing one’s harvest altogether.

With the help of artificial intelligence in agriculture via computer vision, we make agricultural expertise available for everyone. In effect, we put a scientist in the pocket of each and every marginal farmer. From reactive actions to proactive approaches in terms of crop disease detection, we can ensure that local problems will not turn into widespread epidemics. Moreover, by making recommendations on hyper-targeted interventions instead of widespread use of chemicals, we save our budget, protect the soil, and prevent pollution of our food ecosystem.

The project involved researchers from multiple engineering disciplines. How did this interdisciplinary collaboration contribute to the development of the patented technology, and what role does such collaboration play in advancing AI innovation?

AI is not capable of solving practical problems on its own. The success of this technology is owed to the collective and cooperative effort from a number of departments in VNRVJIET, including CSE, IT, EEE, and ECE.

Each department contributed something that was essential for the completion of the task, CSE and IT came up with the deep learning algorithm and data modelling pipelines, while ECE improved image acquisition methods and signal processing, whereas EEE provided some important suggestions regarding hardware-software interaction in future use. That is precisely what drives innovation in AI and turns theoretical technology into patented marvels. Innovation occurs at the junction of different disciplines where intelligent software meets domain hardware.

Looking ahead, what are the next steps for this technology? Are there plans to expand the system to support additional crops, integrate it with drones or mobile applications, or leverage emerging AI technologies such as multimodal AI or edge intelligence for real-time agricultural decision-making?

• The current priority is the commercialization and transition into the field. Although we managed to get our patent, the commercial side of our innovation has not been entirely realized yet — however, we are working on it at the moment.
• In order to move from patent to marketable product, we are currently preparing to develop a lightweight version of our tool in the form of mobile application which will be easy for users to install and use as an app for smartphones.

Together with our commercialization activities, we are preparing our next technological steps:

• Crop Extension: Extending our database to include more types of commercial and regional cash crops beyond the current ones like tomatoes, potatoes, and peppers.
• Aerial Diagnostics: Combining our algorithm with drones to provide farmers with the possibility to scan their fields from above and thus diagnose the whole field without analyzing single plants separately.
• Edge & Multimodal AI: Improving neural networks to work locally on devices (Edge Intelligence) without internet access. Also, we are aiming at implementing Multimodal AI that will combine our leaf scanner's computer vision with local weather and soil forecast to predict the disease outbreaks in advance.