Deep Pre-Trained Model Using ResNet-50 Architecture for Palm Oil Seedlings Classification
Cholid Mawardi, Gema Sukmawati Suryadi, Antinah Latif, Sumanto

Politeknik Negeri Media Kreatif
Universitas BSI

Abstract

One of the causes of declining palm tree production is palm tree plant diseases. Palm tree seedling plant diseases include Root Rot (Ganoderma), Base Rot (Basal Stem Rot), Leaf Spot Disease, Leaf Rot (Phytophthora Palmivora), and White Root. Different types of diseases require different treatments, but not all farmers understand the nature of the disease which can lead to mishandling. To make it easier to solve the problems that exist in the identification of palm tree seedling diseases, it is necessary to innovate a classification system for leaf pests. This system is based on the Convolutional Neural Network (CNN) method with a pre-trained learning scheme that has been tested with a large imagenet dataset, a deep learning method or model developed to cover the weaknesses of machine learning methods. However, deep learning also has the disadvantage that the computation time in the training process is very long and the data size is large. Therefore, a pre-training model is needed to improve the accuracy and performance of deep-learning and facilitate the development of the model structure without building it from scratch. This pretrained model has been implemented. The model used produces the best accuracy rate of 95% and 93% validation using a learning rate system for the weights applied with an iteration time of 30 epochs.

Keywords: palm oil, resnet-50, pre-trained, transfer learning, plant diseases

Topic: Palm Oil Product Diversification in Creative Industries