Mobile-Based Nondestructive Identification of
Cacao's Maturity Using Object Detection

This project was developed as part of the author's Computer Science thesis and introduces a mobile-based solution to identify the maturity of UF18 cacao pods nondestructively using the YOLOv5 object detection algorithm. After training, the model was exported to TensorFlow Lite to be incorporated into an Android-based mobile app.

Features:

Nondestructive Real-time detection and identification of Cacao's Maturity via mobile app
Maturity categorization: Ripe, Unripe, and Infected

Development Environment:

Google Colab
- GPU: Tesla V100-SXM2-16GB
- High-RAM with CUDA 11.8
YOLOv5 Algorithm
PyTorch 2.0.1+cu118
Tensorflow v2.12.0
Android Studio Giraffe 2022.3.1
Comet ML for Visualization

UF 18 Cacao Dataset

Download the Dataset here: UF18 Cacao Maturity Computer Vision Project

Dataset's Health and Class Balance

Repository Contents:

YOLOv5 custom training notebook tutorial integrated with Google Colab.
Mobile app Android source code modified from this GitHub Repository.

Metrics

View the training metrics via Comet ML here: Cacao Maturity Detection Experiment

Detection Sample