insert-agriculture

• 

Inspiration

Agriculture is the backbone of many economies, especially in rural regions. However, plant diseases often go undetected until they cause serious damage, leading to reduced yields and financial losses for farmers. Many farmers lack access to agricultural experts or resources to diagnose diseases early. We wanted to build a solution that leverages modern technology—image processing and AI—to help detect plant diseases quickly, accurately, and affordably, right from a smartphone.

What it does

Our project is an intelligent system that detects diseases in plant leaves through image analysis. The user simply takes a picture of the leaf, and the system identifies whether the leaf is healthy or infected. If a disease is present, the system names the disease and provides basic recommendations for treatment or prevention. The platform can support multiple crops and is designed to be user-friendly for both farmers and agricultural researchers.

How we built it

Data Collection & Preprocessing:

We used the Plant Village dataset, which contains thousands of images of healthy and diseased leaves from various crops.

Images were resized, normalized, and augmented to improve model generalization.

Model Development:

We used Convolutional Neural Networks (CNNs) for feature extraction and classification.

The model was trained using TensorFlow /K eras for multiclass classification across several crop diseases.

User Interface:

A basic web app was built using Flask and HTML/CSS where users can upload images and view results.

(Optional) We plan to expand to an Android app using Flutter or React Native.

Deployment:

The model was converted to a lightweight format and hosted on a local server (or optionally on cloud services like AWS/GCP).

Challenges we ran into

Data Quality & Imbalance: Some diseases had fewer images, which made the model biased toward more common classes.

Image Noise: Real-world images taken from mobile phones often included background noise (e.g., soil, hands, other plants).

Model Overfitting: Preventing the model from memorizing training images while still maintaining accuracy on unseen data.

Latency: Optimizing the system for real-time prediction, especially if deployed on a mobile device with limited processing power.

Accomplishments that we're proud of

Achieved over 90% classification accuracy on test data using a custom CNN model.

Built a complete pipeline from image input to disease prediction and user recommendation.

Designed a user-friendly prototype accessible to non-technical users, including farmers and field workers.

Gained valuable insights into applying AI for a real-world, socially impactful problem.

What we learned

Practical experience with computer vision techniques in agriculture.

How to preprocess and balance real-world datasets effectively.

The importance of explainability and user experience in AI-based tools for non-technical users.

Deployment and performance tuning challenges when moving from a local model to a production-ready application.

What's next for insert-agriculture

Mobile App Development: Creating a cross-platform mobile app for offline disease detection.

Support for More Crops: Expanding the dataset and model to include more plant species and diseases.

Localization: Translating disease names and suggestions into local languages for better accessibility.

Farmer Feedback System: Allowing users to submit feedback or new disease images to improve model accuracy.

Integration with IoT Sensors: Combine with humidity, temperature, and soil sensors for more holistic disease prediction.

Built With

• css
• express.js
• git
• github
• gitlab
• html
• javascript
• mangodb
• nextjs
• node.js
• react