Automated Land Classification Using AIML
Automated Land Classification Using AIML
Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Automated Land Classification using AI/ML
Juhi Sharma, Pal Varma, Yashasvi Verma, Mithilesh Tank
Students, Department of Computer Science and Engineering, Parul Institute of Engineering and Technology, Vadodara, India
Abstract:- The wide unfold of mobiles as hand-held resolution. Land cowl maps for environmental designing,
devices end up in varied innovative applications that land use amendment detection, and transportation designing
create use of their ever-increasing presence in our are often done mistreatment temporal resolution. knowledge
existence. One such application is location chase and integration and analysis of urban areas mistreatment
monitoring; chase exploitation Geographical Positioning mediums resolution remote sensing representational process
System (GPS) and world System for Mobile chiefly focuses on the documentation of engineered up areas
Communication (GSM) technology. Land information or is employed for differentiating between residential,
systems (LIS) provide a technological foundation for business and industrial zones. to collect numerous
decision-making in a wide range of natural resource geographic knowledge in numerous ways and techniques
applications, including scientific, environmental, area unit obtainable. Geographic’s mapping and earth
engineering, and public policy. The LIS provides a resources inventory and mapping represent one of the most
‘framework to mix land surface models, relevant important activities in numerous geosciences and
knowledge and computing tools and resources’ (Kumar engineering disciplines. Combined analysis of map
et al., 2006). Typically, LIS represents a processed info knowledge set at completely different scales is important in
repository for holding geospatial elements, comprising modelling and understanding processes associated with the
‘mapping unit’ pure mathematics, and connected geo- Earth’s physical setting, managing earth resources, etc. of
referenced materials like satellite imaging, earth science these activities involving geographic knowledge sets area
observations, and predictions, and scanned heritage unit qualified because the four Ms: mensuration, mapping,
mapping, that's wherever the thought of analysing the monitoring, and modelling. The term Geographic data
land via the assistance of satellite imaging enters. The Systems (GIS) is sometimes utilized, particularly since the
system displays the item moving path on the monitor and go’s, to explain computerized systems dedicated to the
also the same info may also be communicated to the analysis of varied geographic knowledge sets, hold on
user's cellular phone, thus we'd like an associate degree within the style of digital maps, and also the extraction and
application to form one such mil model that offers North distribution of geographic data. Numerous tools are
American nation automatic classification of land developed permitting localized searches in geographic
exploitation mil options. With a marker feature with databases and knowledge extraction, operations, map
draggable practicality on Google maps, survey officers overlay, and combined knowledge sets analysis. Additional
will regulate coordinate positions additional exactly with specialized GIS has conjointly been developed permitting
real-world things on digital maps. full 3D illustration of the geographic house and new
approaches area unit being explored for representing and
Keywords:- Machine Learning [ML], Artificial Intelligence analysing phenomena evolution in house and time. GIS area
[AI], GPS, LIS unit currently thought-about customary technology for
geographic knowledge analysis and mapping. Pattern
I. INTRODUCTION recognition and laptop vision were introduced in numerous
geosciences and engineering disciplines within the early
Recent advances in computing technology, cloud ’70s as a method of mechanically extracting pertinent
computing, and superior computing area unit paralleled with geographic data from remote sensing representational
those in advanced computer science (AI) algorithms and process.
important investment within the European Copernicus Earth
Observation program and its watch satellite missions. AI II. LITERATURE REVIEW
allows automatic detection of spatial patterns in
environmental knowledge like satellite pictures supported Over the previous couple of years, deep learning (DL)
coaching knowledge. The paradigm of searching for spatial algorithms have exploded in quality for remote-sensing
patterns rather than the historic target spectral data in the image process. the most decilitre ideas applicable to remote
satellite representational process permits the detection of the sensing area unit given during this report, and quite two
latest sorts of land cowl and land use. The free and open hundred publications during this field area unit reviewed and
convenience of world coverage Earth observation examined, the bulk of that were revealed among the last 2
knowledge consistently collected and archived by house years. A meta-analysis was initial conducted to assess the
agencies. Remotely perceived satellite pictures and state of remote sensing decilitre analysis in terms of study
knowledge embrace spectral, spatial, and temporal targets, decilitre model(s) used, image special resolution(s),
Resolutions. Spectral statistics involves components of study space sort, and classification accuracy achieved.
remotely perceived image classification. The most facet that Following that, a comprehensive study of however decilitre
influences the accuracy of a ground object is spatial has been used for remote sensing image analysis tasks like
IJISRT21MAR430 www.ijisrt.com 635
Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
image fusion, image registration, scene classification, object 3.3 DEEP LEARNING MODULE
detection, land use and land cowl (LULC) classification, Deep learning rekindled the pursuit of artificial
segmentation, and object-based image analysis is performed intelligence in the direction of a general-purpose computer
(OBIA). From pre-processing to mapping, this study covers capable of automating any human-related operation. This is
nearly each application and technology within the field of primarily due to a burst of interest in deep machine learning,
remote sensing. Land cowl classification is one in all the which uses hierarchical feature representations rather than
popular ways used for remote sensing because it lays the human-designed features or rules to model high-level
muse of the many world and environmental applications. abstractions, demonstrating great promise in recognizing
Before the advancements of the technology, algorithms like and characterizing LC and LU patterns from VFSR imagery.
most probability classifier, SVM, Random forest and object
primarily based classification. With the assistance of IV. WORKING
ArcGIS API for Python, showcases associate degree finish
to finish land cowl classification progress. There are 3 major At the very initial stages, we will be training the deep
steps within the workflow: (1) extract coaching knowledge, learning data sets using satellite imagery for ArcGIS pro. In
(2) train a deep learning image segmentation model, and (3) which the further segmentation and Labelling of the images
deploy the model for illation and map creation. We'll use will take place. To export training data, there is a labelled
World imagination and high-resolution tagged knowledge imagery layer with the class label for each position, as well
from the Chesapeake Conservancy land cowl project to as a raster input with all the original pixels and band
assist justify this operation. information. This land cover classification situation will use
a subset of the one-meter resolution Kent County, Delaware
III. METHODOLOGY dataset as the named imagery layer and World Imagery:
Color Infrared as the raster input. With the feature class and
ArcGIS professional provides information preparation raster layer in place, the module is ready to use the export
tools for deep learning workflows, conjointly for improved training data() method in the ArcGIS. To export training
support for deploying qualified models for feature extraction data, use the Learn module. In addition to feature class,
and classification. Similar capabilities are on the market in raster layer, and output folder, we must also specify tile size
ArcGIS Image Server within the ArcGIS Enterprise ten.7 (image chip size), strid size (distance to transfer each time
update that permits users to deploy deep learning models at when creating the next image chip), chip format (TIFF,
scale mistreatment distributed computing. Learn modules PNG, or JPEG), and metadata format (how we are going to
within the ArcGIS API for Python to coach deep learning store those training labels). Depending on the size of the
models employing a easy, intuitive API. ArcGIS Notebooks data, tile and stride size, and computing resources, this
offers a ready-to-use atmosphere for deep learning model operation took 15 minutes to 2 hours in our experiment.
coaching. For object detection and classification workflows After the modules have been trained and published as a deep
mistreatment CNTK, Keras, PyTorch, fast.ai, and learning package, the deep learning package will export and
TensorFlow, ArcGIS provides intrinsically Python connect the maps to Android Studio for user access.
formation functions. You’ll be able to conjointly produce
your own Python formation feature that employs your most
popular deep learning library or a specific deep learning
model/architecture.
3.1 ArcGIS PRO:
Every part of the information science progress may be
motor-assisted by ArcGIS tools, as well as information
preparation and preliminary information analysis; model
training; special analysis; and at last, scattering results
through internet layers and maps and driving field operation.
ArcGIS professional additionally provides information
preparation tools for deep learning workflows, additionally
as improved support for deploying qualified models for
feature extraction and classification.
3.2 ANDROID STUDIO
Linking of the deep learning trained modules from
ArcGIS Pro to Android Studio to work as an front-end app
giving the satellite imagery, data and also providing the user
the benefit of choosing their specific location through the Fig. 1: A subset of the labelled data for Kent County,
ArcGIS satellite map. Delaware.
IJISRT21MAR430 www.ijisrt.com 636
Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
V. CONCLUSION
Once the maps are exported and synced with the
Android Studio, the users can easily drag the marker on the
map to their selected locations. National Agriculture
Imagery Program (NAIP) images are categorised into six
major land cover classes by the qualified model: 1)
structures, 2) roads or parking lots, 3) water, 4) cultivated,
open land or bare land, 5) woodland, and 6) planted or dark
cropland, among others. The use of ArcGIS Pro with the
ArcGIS API for Python and combine it with deep learning
tools (e.g., Keras) to make data planning and modelling
easier and integrating it with Android studio for the user
output .Using NAIP bands to identify NAIP, the U-Net
model performs very well.
REFERENCES
[1]. Abdi, A.M., 2020. Land cover and land use
classification performance of machine learning
algorithms in a boreal landscape using Sentinel-2 data.
GIScience & Remote Sensing, 57(1), pp.1–20.
[2]. Anon, 2018. Study Data from University of Putra
Malaysia Update Knowledge of Data Mining
(Integrative image segmentation optimization and
machine learning approach for high quality land-use
and land-cover mapping using multisource remote
sensing data). Information Technology Newsweekly,
p.324.
[3]. Carranza-García, M., García-Gutiérrez, J. & Riquelme,
J., 2019. A Framework for Evaluating Land Use and
Land Cover Classification Using Convolutional Neural
Networks. Remote Sensing, 11(3), pp.Remote Sensing,
Jan 2019, Vol.11(3).
[4]. Díaz-Alcaide, S. & Martínez-Santos, P., 2019.
Review: Advances in groundwater potential mapping.
Hydrogeology Journal, 27(7), pp.2307–2324.
[5]. Du, F. et al., 2015. Soil property variation mapping
through data mining of soil category maps.
Hydrological Processes, 29(11), pp.2491–2503.
[6]. Galeana-Zapien, H., Torres-Huitzil, C. & Rubio-
Loyola, J., 2014. Mobile Phone Middleware
Architecture for Energy and Context Awareness in
Location-Based Services. Sensors, 14(12), pp.23673–
23696
[7]. Kirtan Jha et al., 2019. A comprehensive review on
automation in agriculture using artificial intelligence.
Artificial Intelligence in Agriculture, 2, pp.1–12.
[8]. Ozdogan, M., 2014. A practical and automated
approach to large area forest disturbance mapping with
remote sensing. PloS one, 9(4), p.e78438.
