Journal article Open Access

A Pre-Trained Deep Convolutional Neural Network for the Detection of Tungro in Rice Plants

Atole, Ronnel R.; Alarcon, Karen Michelle A.; Dacillo, Garry P.

This paper presents a computer vision application of transfer learning in the detection of ‘Tungro’ among rice plants, using pre-trained deep convolutional neural networks. An AlexNet network, consisting of 5 convolution layers and 3 fully connected layers of neurons, was customized and fine-tuned to accommodate as inputs, images of rice plants representing two (2) classes: those afflicted with Tungro, and those that are healthy. The fine-tuned network was trained on five hundred twenty (520) images of rice plants, three hundred sixty-eight (368) of which belong to the group without infestation, and one hundred fifty-two (152) are infested with Tungro. Both the training and testing dataset-mages were captured from rice fields around the district and validated by technicians in the field of agriculture. Applying stochastic gradient descent as the learning algorithm, the two-class classifier achieved a very high accuracy of 98.17% at mini batch size of twenty (20) and learning rate of 0.0001.

Files (602.3 kB)
Name Size
602.3 kB Download
All versions This version
Views 1111
Downloads 88
Data volume 4.8 MB4.8 MB
Unique views 1010
Unique downloads 88


Cite as