AI-Driven mobile application in Sinhala for weed identification and herbicide recommendation
Journal of Multidisciplinary & Translational Research · 2026
DOI: 10.4038/jmtr.v11i4.324 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Multidisciplinary & Translational Research, 2026. Verfasst von ARF Shafana und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Weed infestation poses a significant challenge for paddy farmers in Sri Lanka, since it reduces crop yield, degrades quality, and increases production costs. Despite the availability of many applications for weed detection and management, they are quite expensive, not real-time, and limited to English-language users. To address these research gaps, this research proposes a real-time weed identification and herbicide recommendation system for farmers in the Sinhala language. This research presents a mobile application integrated with an image-based deep learning model that detects and classifies the captured weeds, and then recommends suitable herbicides in Sinhala. Deep learning models, including Simple Convolutional Neural Network (SimpleCNN), MobileNetV2, and EfficientNetB0, were trained using a locally captured dataset of common weeds such as Crassula aquatica, Micranthemum, and Scurvy, commonly found in Sri Lankan paddy fields. The primary dataset consisted of 300 images measuring 256×256 pixels for each of the three classes, which were captured, preprocessed, augmented and annotated to avoid overfitting and ensure class balance. For a fair comparison, all three models were trained with a batch size of 32, an initial learning rate of 0.001, 8 epochs, 0.05 label smoothing and with Adam optimizer. Among these, EfficientNetB0 outperformed the others, achieving 97.53% accuracy, 0.97 precision, 0.975 recall and 0.977 F1-score. The final model was converted to. tflite format and integrated into a Flutter-based mobile application. In addition, a secondary dataset on herbicides was obtained from online sources, and its credibility was verified by a local agricultural advisor. The developed mobile application provides users with weed identification results and herbicide recommendations in Sinhala. A qualitative usability evaluation of the mobile application was conducted with eight local farmers with moderate knowledge of smartphones, using a purposive sampling technique. Insights from farmers indicated that this system is a promising, cost-effective alternative for weed detection and management. This study contributes to enhancing sustainable agriculture through AI-assisted weed detection and localized herbicide recommendations.
Zitieren
APA
Shafana, A., Rathnayaka, R. M. N., & Pallegedara, P. K. (2026). AI-Driven mobile application in Sinhala for weed identification and herbicide recommendation. Journal of Multidisciplinary & Translational Research. https://doi.org/10.4038/jmtr.v11i4.324
Vancouver
Shafana A, Rathnayaka RMN, Pallegedara PK. AI-Driven mobile application in Sinhala for weed identification and herbicide recommendation. Journal of Multidisciplinary & Translational Research. 2026. doi:10.4038/jmtr.v11i4.324
BibTeX
@article{Shafana2026,
author = {Shafana, ARF and Rathnayaka, R. M. N. and Pallegedara, P.G.K.K. K.},
title = {AI-Driven mobile application in Sinhala for weed identification and herbicide recommendation},
journal = {Journal of Multidisciplinary & Translational Research},
year = {2026},
doi = {10.4038/jmtr.v11i4.324}
}
EUVIMED