FLORA: An Intelligent Plant Disease Diagnosis Platform Driven by a Three-Layer AI Architecture

FLORA is a hybrid AI agricultural platform that combines computer vision and large language models. It enables real-time plant disease diagnosis and professional agricultural advice through its three-layer architecture of CNN, Gemini, and Gemma. It addresses the issues of traditional manual diagnosis—relying on expert experience, being time-consuming, and geographically limited—helping farmers detect crop diseases early and reduce economic losses.

Global agriculture loses 20% to 40% of its yield annually due to pests and diseases (data from the Food and Agriculture Organization of the United Nations). Farmers in developing countries lack professional technical support and often detect problems only after diseases have spread. FLORA needs to provide a hybrid local+cloud solution to address geographical and network constraints.

FLORA's three-layer pipeline and failover mechanism:

1. CNN Image Recognition: Extracts features like leaf lesions and supports user-customized Keras models;
2. Gemini Diagnosis Validation: A cloud-based multimodal model that integrates images and agricultural knowledge to ensure accuracy;
3. Gemma Friendly Explanation: Runs locally on LM Studio, supports automatic detection of Arabic/English, and falls back to Gemini on timeout.

Tech Stack: Frontend uses pure HTML/CSS/JS, backend uses Node.js Express, and CNN runs locally on Python Flask. Highlights:

• Real-time streaming response (typewriter effect);
• Elastic failover: Gemma timeout → Gemini, Gemini failure → CNN results, adapting to unstable networks.

FLORA Application Scenarios:

1. Self-diagnosis for smallholder farmers: Get professional advice by taking photos with a mobile phone;
2. Agricultural extension: Quick screening of large-area crop health;
3. Education and training: Assists in plant pathology teaching;
4. Scientific research: Rapid collection and annotation of disease samples.

Deployment Process: Configure MongoDB and API keys to start locally; run the frontend using VS Code Live Server. The interface supports Arabic RTL alignment, and LM Studio simplifies local Gemma deployment, catering to global user needs.

Future Expansion Directions: Integrate more crop models, combine meteorological data to provide preventive advice, and establish a farmer community. For developers, its modular architecture, error handling, and multilingual support are excellent reference implementations.