Low-cost image-based coffee plant variety identification system using mobile photography under controlled environment
- Pathum Thani, Thailand : Asian Institute of Technology, 2025
- 87 leaves : ill. + 1online resource
- Thesis ; no. AS-25-06 .
- Asian Institute of Technology. Thesis ; no. AS-25-06 .
A Thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Agricultural Systems and Engineering
Thesis (M. Eng.) - Asian Institute of Technology, 2025
For crop management, quality assurance, and yield optimisation, accurate coffee plant variety identification is essential, especially in smallholder farming systems with limited access to expert knowledge. This study presents a low-cost, image-based system for identifying six coffee varieties, namely SL-28, Java, Geisha, Catimor, Syrina, and Bourbon, using smartphone-captured leaf images in small-scale Polyhouse environments. A lightweight convolutional neural network (MobileNetV2) is used to classify the images after they have been pre-processed using RGB normalisation, resizing to 224224 pixels, and data augmentation methods like flipping, rotation, zoom, and brightness adjustments. The dataset is then split using an 80/20 organized train-test approach, and performance is evaluated via the accuracy, precision, recall, F1-score, and confusion matrix metrics. This trained model is converted to TensorFlow Lite (TFLite) for low-cost edge-device compatibility. Although full Android deployment was not completed, the evaluated TFLite model produces both a predicted class and the corresponding leaf image (e.g., 2Predicted Class: Geisha3) and achieves a validation accuracy of 90%, demonstrating that compression preserves performance. The system remains affordable, portable, and practical as a prototype for resource-limited farming, while challenges such as misclassification of morphologically similar varieties under inconsistent lighting persist. By addressing the underexplored area of variety-level coffee identification and leveraging accessible mobile-based imaging, this work provides a scalable AI-driven foundation for enhancing precision agriculture in smallholder coffee farms.