Applied AI · Agriculture
Feggyfy - AI Crop Disease Detection
End-to-end Android app for early crop disease detection in Kenteng, Boyolali, with an on-device MobileNetV2 model validated by local farming experts.
- >70%model accuracyMobileNetV2 model with TensorFlow Lite, validated by local farming experts on real-world field images.


Context
Farmers in Kenteng, Boyolali, lost crops because diseases were diagnosed too late. They needed a way to recognize a disease early, in the field, with only a phone.
Contribution
I built the Android application end to end: camera capture, on-device inference, results, recommendations, and scan history.
Technical approach
- On-device detection. A MobileNetV2 CNN trained with transfer learning runs with TensorFlow Lite, so analysis works on the phone. CameraX and MediaPipe provide the live camera pipeline.
- Guidance, not only a label. Each detection comes with treatment recommendations for the farmer.
- History in the cloud. Past scans are stored in Firebase so farmers can track a field over time.
- Validated with local experts. Predictions were checked by local farming experts on real-world field images.
Outcome
The model reached over 70% model accuracy, validated by local farming experts on real-world field images, giving farmers in Kenteng, Boyolali, an earlier signal than waiting for a manual diagnosis.