Ecuador has been recognized for the export of high-quality plant products for food. Plant leaves disease detection is an important task for increasing the quality of the agricultural products and it should be automated to avoid inconsistent and slow detection typical of human inspection. In this study, we propose an automated approach for the detection of aphids on lemon leaves by using convolutional neural networks (CNNs). We boarded it as a binary classification problem and we solved it by using the VGG-16 network architecture. The performance of the neural network was analyzed by carrying out a fine-tuned process where pre-trained weights are updated by unfreezing them in certain layers. We evaluated the fine-tuning process and compared our approach with other machine learning methods using performance metrics for classification problems and receiver operating characteristic (ROC) analysis, respectively and we evidenced the superiority of our approach using statistical tests. Computational results are encouraging since, according to performance metrics, our approach is able to reach average rates between 81% and 97% of correct aphids detection on a real lemons leaf image dataset.
CITATION STYLE
Parraga-Alava, J., Alcivar-Cevallos, R., Riascos, J. A., & Becerra, M. A. (2021). Aphids Detection on Lemons Leaf Image Using Convolutional Neural Networks. In Advances in Intelligent Systems and Computing (Vol. 1273 AISC, pp. 16–27). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-030-59194-6_2
Mendeley helps you to discover research relevant for your work.