Nvidia DIGITS, users could easily design, train, and visualize the deep neural
network architectures. As a GPU-optimized framework, Nvidia DIGITS could
automatically handle scale training jobs across multiple GPUs which greatly
facilitates the implementation of deep neural network.
The development of artificial intelligence and machine learning tools is very
fast. There are many other open source artificial neural network libraries and
frameworks available. For example, PyBrain (Schaul et al. 2010) is a modular
machine learning library for Python which is short for Python-Based Reinforcement
Learning, Artificial Intelligence, and Neural Network. PyTorch is a machine learning
library providing a wide range of algorithms for deep learning. Theano is a numerical computation Python Library supports for deep learning implementation on both
CPU and GPU architectures.
6.5.4 A Case Study
The National Agricultural Statistics Service (NASS) of the U.S. Department of
Agriculture (USDA) start operating the Cropland Data Layer (CDL) program in
1997. The mission of the program is to provide a comprehensive, raster-formatted,
geo-referenced, crop-specific land cover classification data product to support the
US agricultural monitoring. By far the CDL data products have been widely adopted
as reference data by growers, agricultural industry, government, academy educators
and students, researchers world-wide for crop production, agricultural production
planning and management, government policy formulation and decision making,
teaching, and various research activities.
An overview of CDL program is given by Boryan et al. (2011) which mainly
introduced the background of the program and CDL products of 2009. To generate
CDL product, a series of inputs including imagery data, ground truth data, and
ancillary data are used. The source of imagery data used in the CDL program
includes AWiFS, Landsat TM and ETM+, and MODIS satellite data. Ground truth
data used in the process of supervised classification training include Common Land
Unit (CLU) data from the Farm Service Agency (FSA) as agricultural ground truth,
and National Land Cover Data (NLCD) as nonagricultural ground truth. Ancillary
data sources include the USGS National Elevation Data (NED), NCLD 2011 tree
canopy, and NLCD 2001 imperviousness data layers.
The classification method used in generating CDL products is See5/C5.0 decision
tree algorithm. To derive the state-level decision trees, FSA CLU data (agricultural
ground truth) and NLCD 2001 data (nonagricultural ground truth) are used as
training dataset by See5/C5.0 classifier. Then, the classifier performs the classification on input data such as AWiFS, Landsat, and MODIS imagery. By comparing the
classification result with the independent validation data extracted from the ground
truth data, the accuracies could be derived. Take Nebraska state as an example, the
total crop mapping accuracies of the major crop categories for the 2016 CDL is
6 Image Processing Methods in Agricultural Observation Systems
97
network architectures. As a GPU-optimized framework, Nvidia DIGITS could
automatically handle scale training jobs across multiple GPUs which greatly
facilitates the implementation of deep neural network.
The development of artificial intelligence and machine learning tools is very
fast. There are many other open source artificial neural network libraries and
frameworks available. For example, PyBrain (Schaul et al. 2010) is a modular
machine learning library for Python which is short for Python-Based Reinforcement
Learning, Artificial Intelligence, and Neural Network. PyTorch is a machine learning
library providing a wide range of algorithms for deep learning. Theano is a numerical computation Python Library supports for deep learning implementation on both
CPU and GPU architectures.
6.5.4 A Case Study
The National Agricultural Statistics Service (NASS) of the U.S. Department of
Agriculture (USDA) start operating the Cropland Data Layer (CDL) program in
1997. The mission of the program is to provide a comprehensive, raster-formatted,
geo-referenced, crop-specific land cover classification data product to support the
US agricultural monitoring. By far the CDL data products have been widely adopted
as reference data by growers, agricultural industry, government, academy educators
and students, researchers world-wide for crop production, agricultural production
planning and management, government policy formulation and decision making,
teaching, and various research activities.
An overview of CDL program is given by Boryan et al. (2011) which mainly
introduced the background of the program and CDL products of 2009. To generate
CDL product, a series of inputs including imagery data, ground truth data, and
ancillary data are used. The source of imagery data used in the CDL program
includes AWiFS, Landsat TM and ETM+, and MODIS satellite data. Ground truth
data used in the process of supervised classification training include Common Land
Unit (CLU) data from the Farm Service Agency (FSA) as agricultural ground truth,
and National Land Cover Data (NLCD) as nonagricultural ground truth. Ancillary
data sources include the USGS National Elevation Data (NED), NCLD 2011 tree
canopy, and NLCD 2001 imperviousness data layers.
The classification method used in generating CDL products is See5/C5.0 decision
tree algorithm. To derive the state-level decision trees, FSA CLU data (agricultural
ground truth) and NLCD 2001 data (nonagricultural ground truth) are used as
training dataset by See5/C5.0 classifier. Then, the classifier performs the classification on input data such as AWiFS, Landsat, and MODIS imagery. By comparing the
classification result with the independent validation data extracted from the ground
truth data, the accuracies could be derived. Take Nebraska state as an example, the
total crop mapping accuracies of the major crop categories for the 2016 CDL is
6 Image Processing Methods in Agricultural Observation Systems
97