|
Listen to this story
|
In a major step forward for AI in agriculture, agri-tech startup KissanAI today announced the launch of Dhenu, a series of fine-tuned vision language models (VLMs) aimed at revolutionising disease detection in crops. The first model in the series, Dhenu-vision-lora-0.1, is built on Alibaba’s Qwen-VL-Chat and delivers a 2x performance boost in identifying diseases in rice, maize, and wheat.
KissanAI founder Pratik Desai revealed that Dhenu-vision-lora-0.1 was trained on a synthetic dataset of 9,000 images covering 10 common diseases in the three crops. The model leverages low-rank adaptation (LoRA) techniques to enable efficient fine-tuning for agricultural applications.
Here’s the link to the Hugging Face repository: https://huggingface.co/KissanAI/Dhenu-vision-lora-0.1
Conversational Insights for Farmers
Beyond just identifying diseases from leaf images, Dhenu engages in dialogue to provide farmers with key information on:
- Disease symptoms
- Severity assessment
- Treatment and prevention methods
“Dhenu merges the depth of agricultural practices with modern AI capabilities to enrich the farming community with actionable insights,” Desai explained. “It embodies the fusion of tradition and technology.”
Outperforming Industry Benchmarks
In initial evaluations on 500 images, Dhenu-vision-lora-0.1 achieved 36.13% accuracy – double that of the base Qwen-VL-Chat model at 17.95%. While still trailing OpenAI’s GPT-4 (51.59%), KissanAI is confident its tailored approach will yield superior results for the ag sector.
The startup plans to expand Dhenu’s capabilities to over 15 crops and 80 diseases in future releases. With this transformative technology, KissanAI aims to guide farmers toward greater prosperity and sustainability.
