Agriculture Reference
In-Depth Information
Another example of data fusion was applied to wirelessly networked data process-
ing for support of automated operations in real time. To support the real-time data
fusion, a software package for a wireless data fusion system was developed to collect
real-time data from multiple sensors installed on moving machinery operating in
the field, transmit the collected raw data from the in-field machine to an in-office
computer performing real-time data fusion, then send the extracted implementation
data back to the in-field machine to perform precise field operations (Guo and Zhang,
2005). In applying the data fusion concept to extract hidden information from the
collected data, Baltazar et al. (2008) developed a software package to implement a
Bayesian classifier to determine the ripening stages of fresh intact tomatoes in terms
of nondestructive testing data. This software package could numerically estimate
the probability of error using Bhattacharyya distance, and verified that through the
multisensorial data fusion it could considerably increase the classification accuracy.
3.4.4 D
ATA
M
INING
Data mining, a process of discovering new patterns from large data sets using either
statistics or artificial intelligence methods, is an effective way of revealing hidden
information from collected raw data. It can process and analyze data from different
perspectives and summarize it into useful information in large relational databases.
Much of current research and development is focused on retrieving and analyz-
ing relevant data, to find the important information from databases carrying large
amount of data (Mucherino et al., 2009). One example is the application of data
mining for fruit sorting and defects detection. As quality plays the most important
role in pricing and marketing, fruit needs to be graded and sorted according to size,
color, shape, and presence of defects at packing houses. Conventionally, such grad-
ing and sorting are performed manually, and largely depends on the knowledge and
experience of a human grader. In developing an automated fruit grading and sort-
ing system, it is essential to convert such human knowledge and experience into an
automatically retrievable way so that a computer-based system could reliably replace
human graders in performing more robust grading and sorting tasks. To accomplish
this goal, Leemans and Destain (2004) developed a hierarchical grading method,
supported by
k
-means algorithm, for sorting Jonagold apples. In their data mining
algorithm, they used a
k
-means clustering to distinguish the blob features from a
set of training images of apples, and used a set of defined clusters as benchmarks in
classifying blobs via a linear discriminant analysis. In detecting defects on cherry
fruit, either visible or invisible, Guyer and Yang (2000) developed a data mining tool
by using enhanced genetic artificial neural networks to identify spectral signatures
of different tissues in cherry images.
One popular application of data mining was the optimization of pesticide usage.
Abdullah et al. (2004) developed a pesticide usage optimization and education tool
to cluster data in a bank of cotton pest scouting data and historic meteorological
data, for showing users the patterns of pesticide usage dynamics. Oakley et al. (2007)
also reported on an initial finding of an ongoing research project to capture differ-
ences in pest management strategies, and decision-making among growers using the
California Pesticide Use Reports database.
