JIMMA UNIVERSITY JIMMA INSTITUTE OF
TECHNOLOGY
FACULTY OF COMPUTING AND INFORMATICS
DEPARTMENT OF INFORMATION TECHNOLOGY
GIS Group Assignment
Image Classification in Remote Sensing: Unsupervised Techniques
Group Five
Name ID
1. Ayenew Amare RU 0507/14
2. Dinkayeu Daniel RU 2126/14
3. AMANUEL AYELE RU 1067/14
4. Yonas Aweke RU 0728/14
5. Abdurehman Yusuf RU 0336/14
6. Hanan Akmel RU 4657/14
Feb 4, 2025
Jimma, Ethiopia
Table of Contents
Table of Contents ...................................................................................................... I
1. Introduction .......................................................................................................... 1
2. Fundamentals of Image Classification ............................................................... 1
2.1 Spectral Resolution ...........................................................................................1
2.2 Spatial Resolution .............................................................................................1
2.3 Preprocessing Techniques ................................................................................ 1
3. Unsupervised Classification Techniques ............................................................2
3.1 K-Means Clustering ......................................................................................... 2
3.2 ISODATA (Iterative Self-Organizing Data Analysis Technique) .................. 2
3.3 Hierarchical Clustering .....................................................................................2
4. Applications of Unsupervised Classification in Remote Sensing .................... 3
4.1 Land Use and Land Cover Mapping ................................................................ 3
4.2 Disaster Management ....................................................................................... 3
4.3 Agricultural Monitoring ................................................................................... 3
4.4 Environmental Conservation ............................................................................3
Conclusion ................................................................................................................. 4
References ..................................................................................................................5
I
1. Introduction
Image classification is an essential technique in remote sensing and Geographic Information
Systems (GIS), widely used to interpret and analyze satellite imagery. The classification process
involves assigning pixels in an image to specific land cover categories such as vegetation, water
bodies, and urban areas based on their spectral, spatial, and temporal characteristics. Accurately
classifying images is crucial for various applications, including land use planning, disaster
management, agricultural monitoring, and environmental conservation [1].
Image classification techniques are broadly categorized into supervised and unsupervised
methods. Supervised classification relies on labeled training data to guide the classification
process, whereas unsupervised classification identifies patterns and clusters in the data without
prior knowledge. The advantage of unsupervised classification is its ability to detect unknown
patterns in remote sensing data, making it useful for exploratory analysis [2]. This document
focuses on unsupervised classification, its methodologies, advantages, and applications in remote
sensing.
2. Fundamentals of Image Classification
Image classification transforms raw remote sensing data into meaningful information by
categorizing pixels based on their spectral signatures. The effectiveness of image classification
depends on several factors, including spectral resolution, spatial resolution, and preprocessing
techniques [3].
2.1 Spectral Resolution
Spectral resolution refers to a sensor’s ability to distinguish between different wavelengths of
electromagnetic radiation. Higher spectral resolution enhances the differentiation between land
cover types, thereby improving classification accuracy. Hyperspectral sensors provide fine
spectral resolution, allowing detailed discrimination of vegetation species, water quality, and soil
properties [4].
2.2 Spatial Resolution
Spatial resolution determines the smallest object that can be detected in an image. High-
resolution images, such as those from commercial satellites like WorldView or QuickBird,
provide detailed classification results but increase computational complexity. Conversely,
medium and low-resolution images, such as those from Landsat or MODIS, are more suitable for
large-scale studies but may lack fine details [5].
2.3 Preprocessing Techniques
Before classification, images often undergo preprocessing to enhance their quality. Common
preprocessing techniques include:
1
Radiometric Correction: Adjusts for sensor inconsistencies and atmospheric effects to
standardize image brightness values.
Geometric Correction: Aligns images to a standard coordinate system, ensuring spatial
accuracy.
Noise Reduction: Removes distortions caused by sensor noise, improving classification
performance [6].
Understanding these fundamental aspects is crucial for implementing effective image
classification techniques in remote sensing applications.
3. Unsupervised Classification Techniques
Unsupervised classification techniques automatically group image pixels into clusters based on
spectral similarity without requiring predefined training data. These methods identify natural
patterns in the data and later associate them with meaningful land cover classes [7].
3.1 K-Means Clustering
K-Means is one of the most widely used unsupervised classification algorithms. It partitions
image pixels into K clusters by iteratively minimizing intra-cluster variance. The algorithm
follows these steps:
1. Initialize K cluster centroids randomly.
2. Assign each pixel to the nearest cluster based on spectral similarity.
3. Recalculate the cluster centroids as the mean of assigned pixels.
4. Repeat steps 2 and 3 until centroids stabilize [8].
The primary advantage of K-Means is its computational efficiency. However, its performance
depends on the initial selection of K and is sensitive to outliers.
3.2 ISODATA (Iterative Self-Organizing Data Analysis Technique)
The ISODATA algorithm extends K-Means by dynamically adjusting the number of clusters
based on statistical criteria. It performs the following steps:
i. Starts with an initial set of clusters.
ii. Merges small clusters and splits large clusters iteratively.
iii. Recalculates cluster centroids until a stopping criterion is met.
This adaptive clustering approach improves classification accuracy in heterogeneous landscapes
by allowing the number of clusters to change based on the data structure [9].
3.3 Hierarchical Clustering
2
Hierarchical clustering builds a tree-like structure of nested clusters. Each pixel is initially
treated as an individual cluster, and clusters are progressively merged based on similarity
measures. There are two main types:
a. Agglomerative Clustering: Begins with individual pixels and merges them iteratively.
b. Divisive Clustering: Starts with all pixels in one cluster and splits them iteratively [10].
Hierarchical clustering is useful for producing detailed classification results but is
computationally intensive.
4. Applications of Unsupervised Classification in Remote
Sensing
Unsupervised classification techniques are widely applied in various domains, including:
4.1 Land Use and Land Cover Mapping
Unsupervised classification helps map land use and land cover (LULC) changes over time. This
is critical for urban expansion studies, deforestation monitoring, and resource management [11].
4.2 Disaster Management
Satellite imagery classification aids in identifying disaster-affected areas. For example, flood
mapping using unsupervised classification can help emergency responders allocate resources
efficiently [12].
4.3 Agricultural Monitoring
Crop classification and yield estimation benefit from unsupervised techniques, enabling large-
scale assessment of agricultural conditions without field surveys [13].
4.4 Environmental Conservation
Monitoring ecosystems, detecting deforestation, and assessing biodiversity loss rely on
unsupervised classification to analyze remote sensing data over vast regions [14].
3
Conclusion
Unsupervised classification techniques are a cornerstone of remote sensing and Geographic
Information Systems (GIS), enabling automated interpretation of satellite and aerial imagery
without requiring prior knowledge or labeled training data. These techniques, including K-
Means clustering, ISODATA (Iterative Self-Organizing Data Analysis Technique), and
hierarchical clustering, provide significant advantages in applications such as land cover
mapping, agricultural monitoring, disaster response, and environmental conservation. Their
ability to detect patterns and categorize pixels based solely on spectral properties makes them
invaluable tools for large-scale and exploratory remote sensing studies.
Despite their numerous benefits, unsupervised classification methods also present certain
challenges. One of the primary issues is computational complexity, particularly when
processing high-resolution or hyperspectral imagery. The iterative nature of clustering
algorithms, such as ISODATA, requires substantial computational resources, making real-time
classification difficult for large datasets. Additionally, the lack of predefined class labels
means that post-classification refinement is often necessary to ensure meaningful
interpretations. Analysts must manually assign class labels to clusters, which introduces
subjectivity and the potential for misclassification, especially in heterogeneous landscapes with
mixed land cover types.
To address these challenges, future research should focus on integrating unsupervised
classification techniques with advanced machine learning and deep learning models.
Convolutional Neural Networks (CNNs) and other artificial intelligence-based approaches have
shown great promise in improving classification accuracy by leveraging spatial, spectral, and
contextual information. Combining traditional unsupervised clustering methods with deep
learning could enhance feature extraction, reduce classification ambiguity, and minimize the
need for manual post-processing. Additionally, advancements in cloud computing and high-
performance computing infrastructure will facilitate more efficient processing of large remote
sensing datasets, making unsupervised classification more accessible and scalable for real-world
applications.
Overall, while unsupervised classification remains an essential method for automated image
interpretation, continuous advancements in computational techniques, hybrid classification
approaches, and artificial intelligence will further refine its effectiveness. By integrating these
emerging technologies, remote sensing professionals can achieve more accurate, efficient, and
meaningful land cover classifications, ultimately supporting informed decision-making in
environmental management, urban planning, and global sustainability efforts.
4
References
[1] P. Mather and M. Koch, Computer Processing of Remotely-Sensed Images: An
Introduction, 4th ed., John Wiley & Sons, 2011.
[2] P. A. Longley, M. F. Goodchild, D. J. Maguire, and D. W. Rhind, Geographic Information
Systems and Science, 3rd ed., Wiley, 2015.
[3] J. Campbell and R. Wynne, Introduction to Remote Sensing, 5th ed., Guilford Press, 2011.
B. Lillesand, R. W. Kiefer, and J. W. Chipman, Remote Sensing and Image Interpretation, 7th
ed., Wiley, 2015.
[4] P. Bolstad, GIS Fundamentals: A First Text on Geographic Information Systems, 5th ed.,
XanEdu Publishing, 2016.
[5] R. Jensen, Introductory Digital Image Processing: A Remote Sensing Perspective, 4th ed.,
Pearson, 2015.
[6] T. Blaschke et al., "Object-based image analysis for remote sensing," ISPRS Journal of
Photogrammetry and Remote Sensing, vol. 65, no. 1, pp. 2-16, 2010.
[7] J. A. Richards and X. Jia, Remote Sensing Digital Image Analysis, 5th ed., Springer, 2013.
[8] T. M. Lillesand and R. W. Kiefer, Remote Sensing and Geographical Information System,
3rd ed., Wiley, 2016.
[9] P. Bolstad, GIS Fundamentals, 5th ed., XanEdu Publishing, 2016.
