InspirationChallenges we ran into
The biggest problem was that photos taken in a real field have shadows and dirt, while our training photos were perfect. We had to "toughen up" the AI by giving it blurry and messy photos to practice on. We also had to shrink the AI's size so it wouldn't crash older phones or require super-fast internet. Accomplishments that we're proud of We are proud that our AI is 98% accurate. We successfully turned a very complex computer program into a simple app that anyone can use. Most importantly, we made it small enough to fit on a basic smartphone, making it accessible to farmers who need it most. What we learned We learned that clean data is better than a complex program. If you show the AI bad photos, it learns bad habits. We also learned how to bridge the gap between "high-tech" lab science and the "real-world" needs of a farm. It taught us that technology is most powerful when it solves a simple, human problem. What's next for AgriVision AI Next, we want to add weather alerts to warn farmers before a disease even starts. We also want the app to speak in local languages so it's easier to use. Finally, we hope to connect the AI to drones so large farms can be scanned from the sky in just a few minutes.I saw that many farmers lose their crops because they can't tell if a plant is sick until it's too late. While experts are hard to find and expensive, almost every farmer has a smartphone. I wanted to build a "Digital Doctor" that helps farmers identify plant diseases instantly, saving their food and their money. What it does AgriVision AI is an easy-to-use tool for farmers. You simply take a photo of a plant leaf, and the AI scans it. Within two seconds, it tells you if the plant is healthy or sick, naming the specific disease (like "Late Blight") and how sure it is about the result. It’s like having a scientist in your pocket. How we built it We used Python to write the code and a dataset of over 50,000 leaf images to teach the AI. We built a "brain" called a CNN (Convolutional Neural Network) that learns to recognize patterns like spots and colors. To make it work on a phone, we used a lightweight version called MobileNetV2 and built the final website using Streamlit.
Log in or sign up for Devpost to join the conversation.