An Intelligent Variable Spraying Decision-Making System Based on Fuzzy Neural Network for Greenhouse Mobile Robot - Intelligent Computing for Sustainable Energy and Environment - page 260

level of plant diseases and insect pests are input through the computer user in-
terface. On that basis, an oine training fuzzy neural network with self-learning
and adaptive capacity is constructed, which fuses the information of area, dis-
tance and the level of plant diseases and insect pests, and implements quickly
an intelligent decision for the spraying robot on pesticide quantity.

3.2 Design of Fuzzy Neural Network

The variable spraying process is complex non-linear, so it is hard to build the
mathematical model. Multi-input fuzzy rules aimed at the distance, area, and
damage level are not only dicult to establish, but also cant adaptively adjust
with the change of greenhouse environment. Hence, the decision-making method
based on fuzzy neural network is proposed to improve the ability of adapting to
the change of environment and impact factors in the conventional fuzzy decision.
The architecture of the fuzzy neural network is shown in Fig. 3.

Input Fuzzification Rules Defuzzification Output

Area s

Distance
d

Damage
Level n

i=3 j=8 k=18 m=18 n =1

Fig. 3. Architecture of the fuzzy neural network

Use fuzzy neural network [7,8] structure to handle the input fuzzification,
fuzzy inference, fuzzy rules, and defuzzification. There are five layers.
Layer 1 (input layer): there are three nodes representing the crops area 's',
the spraying distance 'd', the damage level of disease and insect 'n'. In the layer,
each neuron represents an input signal, the number of neurons is equal to the
number of variables appeared in the premised fuzzy rules, it passes input vector
directly to the next layer. The i th neuron connects with the i th unit of input
variable X .
Layer 2 (fuzzification input layer): there are eight nodes, each node represents
a language variable: area, the unit is m 2 , given three language variable values,
namely, 'big', 'medium', 'small'. Distance, the unit is m , given three language
variable values, namely, 'near', 'medium', 'far'. Damage level of plant disease and
insect pest, given two language variable values, namely, 'serious', 'not serious'.
In this layer, each neuron is used to simulate the membership function of input
variables, and its role is to calculate the membership functions of input weight