Original scientific paper

https://doi.org/10.31803/tg-20240910214215

Detection and Classification of Growth Stages in Rice Using Artificial Neural Networks

Durodola Folasade ; Department of Mechatronics Engineering, Federal University of Agriculture, Alabata Road, Abeokuta, Ogun State, 110111, Nigeria *
Owoeye Samuel ; Department of Mechatronics Engineering, Federal University of Agriculture, Alabata Road, Abeokuta, Ogun State, 110111, Nigeria
Kamil-Bello Furqan ; Department of Mechatronics Engineering, Federal University of Agriculture, Alabata Road, Abeokuta, Ogun State, 110111, Nigeria
Daodu Sakira ; Department of Mechatronics Engineering, Federal University of Agriculture, Alabata Road, Abeokuta, Ogun State, 110111, Nigeria
Makinde Kayode ; Department of Mechatronics Engineering, Federal University of Agriculture, Alabata Road, Abeokuta, Ogun State, 110111, Nigeria
Folaranmi Olaniyi ; Department of Mechatronics Engineering, Federal University of Agriculture, Alabata Road, Abeokuta, Ogun State, 110111, Nigeria

* Corresponding author.

Abstract

This project focused on the development of a machine learning model to classify rice plants based on their growth stages, specifically identifying whether the plants are in the "Raw" or "Ripe" stage. The research was conducted using a dataset obtained from Roboflow, which provided annotated images of rice plants. The dataset was divided into training, testing, and validation subsets to ensure the model's robustness and generalization capability. The project involved a comprehensive data preparation process, which included consolidating images into class-based folders, handling file conflicts, and stratifying the dataset into appropriate splits for training, testing, and validation. Several convolutional neural networks (CNN) architectures were explored, including ResNet50, InceptionV3, and MobileNetV2, each leveraging transfer learning from pre-trained models on the ImageNet dataset. ResNet50 achieved an accuracy of 87.3% with a log loss of 0.33, demonstrating good performance but with some misclassifications between similar classes. InceptionV3 outperformed the other models, achieving an accuracy of 95.1% and a log loss of 0.13, indicating superior classification capability and better calibration of predicted probabilities. MobileNetV2 also performed well with an accuracy of 93.5% and a log loss of 0.22, offering a balance between accuracy and computational efficiency. The results highlight InceptionV3 as the most effective model for this task, with a strong ability to differentiate between the rice growth stages. The findings underscore the importance of model selection and data preparation in developing accurate and reliable machine-learning models for agricultural applications. The project demonstrates the potential of CNNs in improving agricultural practices through precise crop monitoring and classification.

Keywords

convolutional neural networks (CNNs); image classification; InceptionV3; machine learning; MobileNetV2; ResNet50; rice plant

Hrčak ID:

348860

URI

https://hrcak.srce.hr/348860

Publication date:

15.9.2026.

Visits: 0 *