I grew up on a farm, my parent's main challenge while farming was crop pests and disease. They would spray the whole farm with certain fungicides, sometimes without explicit knowledge of what is affecting the farm. When I started this project, the goal was to help farmers like my dad detect the correct type of disease affecting the crops. Though the project is in the early stages, I believe having it as an open-source will lead to collaboration from different engineers and make this a working product.
What it does
Given a crop leaf image, the model can detect if it's from a healthy plant or an infested plant with a particular type of crop disease.
How I built it
- Working on image processing and transformations.
- Using PyTorch pre-trained model densenet 201 and customizing it to this project.
- Create a function in which you pass the image path, then perform prediction on the image.
Challenges I ran into
Two years ago, I did not have a lot of experience in Image processing. It's a bit difficult to work on the images for this particular case.
Accomplishments that I'm proud of
- Being able to correctly predict the classes of the images that the model has learned. ## What I learned
- PyTorch is very resourceful in research, and currently, one can host any PyTorch model in production.
What's next for Build a Simple Crop Disease Detection Model with PyTorch
Collect hundreds of thousands of crop disease images to increase the efficiency of the model.
Create a hardware spraying machine that can have different fungicides, probably powered by a drone and camera that kind of patrols the farm, takes an image, and feed it to the model. Then the model can detect the specific crop disease and spray instantly with the correct chemical. Having such a machine for farmers will reduce the wastage of spraying one fungicide on the whole farm.