ManalagiCare
Overview
Mobile application for apple fruit disease classification in the orchard: an Apache Cordova app sends each photo through a Laravel backend to a Flask prediction server running an InceptionV3 model.
- Year
- 2025
- Status
- Ongoing
- Role
- Researcher and Mobile Developer
- Kind
- Mobile Apple Disease Detection
- Domain
- AI & Vision
- Terminal command
/manalagicare
What it does
- 5-class apple fruit disease classification with InceptionV3 (97.15% accuracy in the ICIMCIS 2025 paper)
- Server-side inference: Cordova app → Laravel backend → Flask prediction server
- Image classification, video identification, a location map and disease information in one app
- Backed by published research at ITIS 2025 / ICIMCIS 2025
Apple disease classification in the orchard
ManalagiCare is the mobile app from the ICIMCIS 2025 paper. A farmer photographs an apple and an InceptionV3 model sorts it into one of five classes: Black Pox, Black Rot, Codling Moth, Green Fruitworm or Healthy. The model runs on a prediction server, not on the phone.
My role spans research and mobile development: data curation, model training, visualisation and writing on the paper, and integrating the trained model into the app. The project remains ongoing.
Read more: ICIMCIS 2025 · the app's paper · Research and development experience
From a photo to a diagnosis
Three system layers, as described in the ICIMCIS 2025 paper. Inference runs on the prediction server.
Apache Cordova app
The user logs in, the app reads the device location, and the fruit photo is uploaded as base64.
Laravel backend
Receives the image and forwards it to the prediction server.
Flask prediction server
The InceptionV3 model predicts the disease class and returns it.
Stored and shown
Laravel saves the result in MySQL and the app shows it with symptoms, treatment and a heatmap.
What the app gives a farmer
The main page offers four features. A classification result comes with the disease's symptoms, treatment steps and prevention advice, plus a heatmap of the parts of the photo the model focused on.
- Image classification
- Video identification
- Location map
- Disease information
Read more: ICIMCIS 2025 · the app's paper
The research behind the app
The ICIMCIS 2025 paper trained four CNNs on 3,360 photos collected in Poncokusumo orchards. InceptionV3 was the most accurate (97.15%; precision, recall and F1 of 0.97) but also the slowest to train and the largest to store. I am the second of three authors.
The ITIS 2025 paper, “Improving Apple Fruit Disease Classification using InceptionV3 and Adam Optimizer,” laid the groundwork on fruit rather than leaves; I am the first of three authors. The ICAIDES 2025 paper then compares compact networks that classify on the phone itself.
These figures are paper results, not measured accuracy or performance of the app.
Read more: ICIMCIS 2025 · the app's paper · ITIS 2025 · InceptionV3 for apple fruit · ICAIDES 2025 · lightweight on-device models
References
- ICIMCIS 2025 · the app's paper →
Dataset, model comparison, app architecture and features, explained in full.
- ITIS 2025 · InceptionV3 for apple fruit →
Five-class fruit classifier, 97.15% test accuracy, confusion matrix.
- ICAIDES 2025 · lightweight on-device models →
Compact networks for classification on the phone; MobileNetV3 at 95%.
- Research and development experience →
Deep-learning internship, model training and app integration.
Stack
- Frontend & mobile
- Apache Cordova
- Backend & APIs
- LaravelFlask
- Data
- MySQL
- AI & vision
- TensorFlowInceptionV3