Crop disease diagnosis is a labor-intensive and challenging process that can be tackled with machine learning. The potential to reduce produce costs by improving crop diagnostics is a lofty goal to work towards.
HerbVector is an image-recognition software platform to be used in conjunction with drone technology to produce diagnostic maps of crop fields. Images of crops are taken using a drone, which produce a map of: crops, any diseases they appear to have, and how to treat them.
We used machine learning with cross-entropy to develop an image-recognition model for identifying diseases in crops.
We have a functional model which can correctly identify Pierce's Disease, Downy Mildew, and Phylloxera that affect grape vines while differentiating the diseased ones from thehealthy plants!
Challenges that we ran into:
The main challenge was getting adequate data sets to identify the diseases. We had to build these from scratch with google images. We had some trouble with memory leaks in the server, causing it to become overloaded.
What's next for CropVector:
We plan to deploy the trained model in Raspberry Pi and mount it on a drone.Once a prototype drone is functional, we will begin data collection for crop fields.Later, we will develop a diagnostic map for crop fields.