8.5. INTEGRATED PEST MANAGEMENT SYSTEMS 279
Table 8.16. Classification results of trash objects
infive test samples using ANFIS: 222-partition
Total number Actual Classified trash type
of objects trash type Bark Stick Leaf Pepper
10 Bark 6 4 0 0
4Stick0202
16 Leaf 0 2 14 0
215 Pepper 0 0 0 215
Table 8.17. Classification rates with 232- and 222-partition
Membership
Partition 232-Partition 222-Partition
Samples Training Test Training Test
Classification
rate (%)97.6526 91.8367 97.1831 95.1020
From the results of this application, we conclude that ANFIS produces far
superior results in terms of its performance as a classifier, and has potential
for on-line implementation based upon its ability to adapt. Due to the ìfuzzyî
nature of the variables, the ANFIS architecture is ideally suited for applications
that require rule-based reasoning as part of the decision-making process. The
adaptive nature of the network architecture makes ANFIS a highly robust tool
for decision-making under uncertainty.
8.5 Integrated pest management systems
Control of pesticides in agricultural ecosystems is essential towards minimizing
environmental pollution. Lowering the use of pesticides requires the implemen-
tation of biological control wherein natural enemies and predators are used to
control the population of insects that are harmful to the agricultural commod-
ity. Therefore, the goal in Integrated Pest Management (IPM) systems is to
implement methodologies that can minimize pesticide use while maximizing the
use of biological controls. Achieving this goal requires carefully monitoring the
populations of specific insect species and determining the appropriate mix of
pesticide and biological controls for the target agricultural ecosystem. In this
context, there is a need to develop methods to identify the class of insects that
populate specific agricultural ecosystems. In this example, we discuss a neuro-
fuzzy approach to insect classification.
Statistical approaches to classifyinginsects have been met with limited suc-
cess. This is due primarily to the fact that the spatial patterns of insects are not
fixed. This problem is made even more complex by changes in insect population
with season and time. Factors such as weather, host plants, soil, predator, para-
sitoid, and behavioral factors of the insect population can all contribute towards
an extremely complex problem in insect classification. Hence it is difficult, if not
impossible, to obtain satisfactory statistical models to predict or classify insect
categories. Figure 8.21 illustrates a framework for developing IPM systems.