A novel approach based on deep learning techniques and UAVs to yield assessment of paddy fields

Yield assessment is one of the main interests at regional and national levels of agriculture management. The accuracy of the assessment is not highly expected inherently and, moreover, some costly and sophisticated tools (e.g., satellite images) have been usually involved without a careful considera...

Full description

Saved in:
Bibliographic Details
Published in2017 9th International Conference on Knowledge and Systems Engineering (KSE) pp. 257 - 262
Main Authors Nguyen Cao Tri, Duong, Hieu N., Tran Van Hoai, Tran Van Hoa, Nguyen, Vu H., Nguyen Thanh Toan, Snasel, Vaclav
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.10.2017
Subjects
Agriculture
Convolution
Deep Learning
Electronic mail
Image resolution
Image segmentation
Machine learning
Neural networks
Paddy Field
Unmanned Aerial Vehicles
Yield Assessment
Online AccessGet full text

Cover

More Information
Summary:Yield assessment is one of the main interests at regional and national levels of agriculture management. The accuracy of the assessment is not highly expected inherently and, moreover, some costly and sophisticated tools (e.g., satellite images) have been usually involved without a careful consideration of their investment. In this paper, a novel approach has been proposed orienting to farmers (especially in Southeastern Asia) who are only affordable for low cost and easy-to-use tools. With this vision, the paper presents an idea to co-operate certain, powerful and modern technologies-deep neural networks (DNNs) and unmanned aerial vehicles (UAVs)-with respect to solving the problem of yield assessment. The imagery of paddy fields acquired by UAVs at low altitudes is then used for training DNNs and assessing the yield of the paddy fields. Four DNN models are applied to figure out the most appropriate one for this approach. To test the applicability of the proposed approach, some experiments are conducted on a dataset which was collected at paddy fields in Tay Ninh province, Vietnam. The yield assessment is not directly predicted from DNNs, instead being achieved by a combination of the classification results and a bush-level yield estimation of agricultural experts in the fields. The paper indicates that the proposed approach is quite potential for precision agriculture, along with some remaining challenges.
DOI:10.1109/KSE.2017.8119468