Publication: Embedded machine vision system for uav-based oil palm tree healthiness inspection
Loading...
Date
2020-07-01
Authors
Ling, Jia Wen
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Palm oil is the world’s most consumed edible vegetable oil that comprises over 50% of all package products sold in the market. However, the palm oil industries concern the diseases of oil palm trees that may affect the palm oil yield negatively.Although previous works proposed machine learning approaches for oil palm tree diseases classification, there are still some limitations exists such as huge network needed for large image size, recognition rate decrease and difficult to extract the features under various uniform environment. Recently, the improvement of deep learning can be seen clearly on applications in different fields such as face recognition, image classification and document analysis. Therefore, this research proposed to develop a machine vision system by using deep convolution neural network technique to check the healthiness of oil palm tree. A total of 1008 images, 208 healthy and 800 unhealthy oil palm images were used for training and validation. The CNN model was tested with different experiments by altering hyperparameters such as number of epochs and convolutional layers, filter size, pooling layer, ratio of testing and validation set and activation function for fully connected layer. Finally, the best model obtained with 100 epochs, 7 convolutional layers, 3x3 filter size, max-pooling layer, ratio of 9:1 for training to validation set and sigmoid activation function for fully connected layer. The results on the embedded machine vision system with 98.4% accuracy demonstrate that the proposed technique can effectively classify the healthiness of the healthy and unhealthy oil palm tree.