Centinela
ML project applying transfer learning to detect Anastrepha fraterculus plagues and notify the farmer.
Centinela uses a transfer-learning model to spot Anastrepha fraterculus infestations in crop images and alerts the farmer as soon as a plague is detected.
Why did I build it?
I identified a huge problem in my country: the fruit fly causes massive economic losses for farmers and the government.
Challenges
Talking about lab tests was a huge W for my project, using public and Dryad dataset images the model reached almost a perfect score, but yeah I knew that was a problem, perfection isn't a good signal. I tried with homemade images, and the result was expected, the model failed miserably in the tests. I engineered traps specially for the anastrepha fraterculus in my zone to gather my own dataset, so I can train my model with real-world trap images to fill that gap.
Learning
I finally get a good performance of the model by using a dataset specialized on the real traps the monitoring station actually use.