347
APPLICATION OF FUZZY REASONING TO ASSESSMENT OF CROP STRESS LEVEL
BASED ON MODIS DATA: A FOCUS ON HEAVY METAL POLLUTION
Xiangnan Liu*, Nan Jiang, Xiaodong Wang, Chenxia Han
School of Information Engineering, China University of Geosciences, Beijing, China, 100083 - liuxncugb@163.com
KEY WORDS: Crop, Pollution, Hyper Spectral, Fuzzy Logic, Dynamic, Algorithms
ABSTRACT:
This paper reviews the application of fuzzy theory and its combination with artificial neural-network technology for remote sensing
information extraction. A dynamic fuzzy neural-network model is presented for crop heavy metal stress level assessment based on
MODIS data. Hyperspectral vegetation indices, including NDVI, EVI and NDVIg, were used as input variables in this model for the
purpose of enhancing and extracting weak information of crop heavy metal stress obtained from large-scaled farmland under
complex circumstances. The output error and the root mean square error were considered as system performance evaluation factors.
250 samples, which contained values of hyperspectral vegetation indices and heavy metal stress levels, were prepared for the
training process. And fuzzy reasoning rules were generated and evaluated based on their significance. At the end of the training
process, this dynamic fuzzy neural-network model generated a total number of seven fuzzy rules. Another dataset, with 60 testing
samples, was applied to evaluate the performance of this trained system. The result of this experiment indicated that this model was
capable of extracting stress information with reasonable accuracy, which is over 95%, and thus it could be used as an effective tool
in monitoring and managing agricultural environment.
1. INTRODUCTION
The estimation of crop heavy metal stress level in large scale
farmland is essential for the management and protection of
agricultural environment. Besides conventional methods, such
as crop tissue analysis and soil sampling analysis, remote
sensing technology has been applied to obtain more complete
and accurate information. However, crop heavy metal stress is
a kind of weak information without significant representation
under complex circumstance. Therefore, it’s difficult to
classify crop stress level by traditional remote sensing
information extraction methods. To solve this problem, fuzzy
theory is applied in this study, and a dynamic fuzzy neural-
network model is built and trained to classify crop heavy metal
stress level.
Fuzzy neural-network (FNN) is defined as a combination of
fuzzy theory and artificial neural-network technology, which is
composed of fuzzy neurons, including fuzzy neurons
describing “if-then” rules, fuzzy neurons with fuzzy output
values and fuzzy neurons with fuzzy input values (Shao Dong
et ah, 1999). The greatest advantage of FNNs is their ability to
model complex, non-linear process without having to assume
the form of the relationship between input and output variables,
that is to say, it's unnecessary to apply expert knowledge in
these systems (Kwokwing Chau, 2006). Performance of a
FNN system can be improved by means of modifying network
architecture, such as “if-then” rules, membership functions and
the significance of each rule. Considering that it is a useful
technique for regression and classification problems, increasing
attention has been paid in recent years to its application in
remote sensing area. Researchers resorted to various network
structures and learning algorithms to improve its efficiency and
accuracy in extracting thematic information from remotely
sensed data. And it has been found that FNNs have several
advantages over traditional information extraction methods.
* Corresponding author.
Firstly, they are non-linear models and thus have the capability
to analyze complex data patterns. Secondly, they can process
data at varied measurement scales such as continuous, ordinal
and categorical data. So, they can describe and analyze fuzzy
phenomena which are often encountered in practical
applications (D.P. Kanungo et ah, 2006). Thirdly, because of
their ability to integrate non-spectral information into the
networks in the form of additional input variables, they allow a
better discrimination between vegetation spectral reflectance
and plant stress level (Jesus Favela et al., 1998). By combining
the power of ANNs for modelling complex phenomena, FNNs
can provide better results than pure fuzzy logical approach. It is
the particular intention of this study to remark upon the crop
heavy metal stress detection and classification by means of
fuzzy neural-network modeling based on hyperspectral
remotely sensed data.
In this paper, a dynamic fuzzy neural-network (DFNN) model
is presented to extract heavy metal contamination information
in large scale areas under complex circumstance based on
MODIS data. Values of hyperspectral vegetation indices,
including NDVI, EVI and NDVIg, were used as input variables.
Information to be represented by this network was fuzzy rules
describing the relationship between input variables listed above
and output crop heavy metal stress level. It was built and learnt
form training data sets collected form typical heavy metal
contaminated farmlands. According to experimental result, it
was verified that this system was capable of extracting stress
information from hyperspectral remote sensing data of large
scale farmland with reasonable accuracy, and thus it could be
used as an effective tool in monitoring and managing
agricultural environment.
