Mahardika Gama Pradana

Work place: Pusat Penelitian Kelapa Sawit, Siantar, Indonesia

E-mail: mahardikagama@gmail.com

Website: https://orcid.org/0000-0003-0477-7055

Research Interests:

Biography

Mahardika Gama Pradana is a dedicated and committed researcher, actively contributing to the Palm Oil Research Center. Specializing in Plant Pests and Diseases, Mahardika has become a key figure in efforts to understand and address challenges related to palm oil. His degree in the relevant field strengthens his knowledge foundation. As a researcher at the Palm Oil Research Center, Mahardika is involved in in-depth research to identify, analyze, and develop strategies for the prevention and control of pests and diseases that can impact the growth and productivity of palm oil.

Author Articles
Optimizing VGG16 for Accurate Pest Identification in Oil Palm: A Comparative Study of Fine-Tuning Techniques

By Muhathir Andre Hasudungan Lubis Dwika Karima Wardani Mahardika Gama Pradana Ilham Sahputra Mutammimul Ula

DOI: https://doi.org/10.5815/ijieeb.2024.05.03, Pub. Date: 8 Oct. 2024

Recent advancements in pest classification using deep learning models have shown promising results in various agricultural contexts. The VGG16 model, known for its robust performance in image classification, has been applied to the task of classifying pests in oil palm plants. This study aims to evaluate the effectiveness of the VGG16 model in identifying pests on oil palm, comparing the performance of default settings with models fine-tuned using grid search and random search techniques. We employed a quantitative approach, training the VGG16 model with three different configurations: default, fine-tuned with grid search, and fine-tuned with random search. Evaluation metrics including precision, recall, F1-Score, and overall accuracy were used to assess model performance across different pest categories: Metisa plana, Setora nitens, and Setothosea asigna. The default VGG16 model achieved precision, recall, and F1-Score values around 90% for Metisa plana, Setora nitens, and Setothosea asigna, with an overall accuracy of 91.00%. Fine-tuning with grid search improved these metrics, with precision, recall, and F1-Score reaching approximately 93.88%, 92%, and 92.93% respectively, and an overall accuracy of 93%. The random search fine-tuning resulted in even higher performance, with precision of about 95.92%, recall of 94%, and F1-Score of 94.95% for Metisa plana, and overall accuracy of 94.67%. The VGG16 model demonstrated strong performance in pest classification on oil palm, with significant improvements achieved through fine-tuning techniques. The study confirms that grid search and random search fine-tuning can substantially enhance model accuracy and efficacy. Future research should focus on expanding the dataset to include more diverse pest species, incorporating attention mechanisms, and leveraging automated control technologies like drones and the Internet of Things (IoT) to further improve pest management practices.

[...] Read more.
Other Articles