Mtechphase 1

Uploaded by

keerthipriyanuthi

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PPTX, PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

3 views

Mtechphase 1

Uploaded by

keerthipriyanuthi

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PPTX, PDF, TXT or read online on Scribd

You are on page 1/ 14

Deep Learning-Based Classification Methods

for Remote Sensing Images in Urban Built-Up

Areas
UNDER THE ESTEEMED GUIDANCE OF
Dr. A.RAJANI

SUBMITTED BY:

NUTHI
KEERTHI PRIYA– 20011P0417
AIM
To develop and evaluate deep learning methods for the classification of urban built-up areas using high

spatial resolution remote sensing images.

OBJECTIVES
 To design and implement Convolutional Neural Networks (CNN) and Capsule Networks (CapsNet) for the

classification of urban built-up areas using high-resolution remote sensing images.

 To evaluate the performance of these deep learning models on dataset, with a focus on improving the

accuracy, precision, and reliability of urban area classification.

 To develop and assess the Same Model with Different Training Rounding based on CNN (SMDTR-CNN)

and CapsNet (SMDTR-CapsNet) to further enhance the overall classification accuracy and robustness for

various urban functional zones.

INTRODUCTION
 Urban areas have become very important in remote sensing applications due to their significant role in
reflecting and scattering characteristics similar to built-up areas, making them challenging to classify using
traditional methods.
 Remote sensing with high spatial resolution (HSRRS) images is crucial for urban mapping, but traditional
methods often struggle with the complexity and variability in urban areas.
 Deep learning models like CNN and CapsNet offer superior performance over traditional methods,
enabling the extraction of multi-layer features from HSRRS images.
 Deep learning enhances the classification of urban built-up areas by considering urban functions, aiding in
urban planning, ecological evaluation, and emergency response.
Fig: Block Diagram
QGIS Software
 QGIS (Quantum GIS) is an open-source Geographic Information System (GIS) that provides tools for
viewing, editing, and analyzing geospatial data.
 It is widely used in remote sensing projects for its robust feature set and ease of use.
 Data Labeling: QGIS can be utilized to label and annotate high spatial resolution remote sensing
(HSRRS) images.
 Shapefile Creation: Shapefiles can be created to mark and classify various urban features (e.g., buildings,
roads, airports) directly on the satellite imagery.
 Integration with Deep Learning: Labeled data can be exported from QGIS, which can be used in training
convolutional neural networks (CNNs) and Capsule Networks (CapsNets).
METHODOLOGY
1. HSRRS Dataset Composition:
The HSRRS image dataset for urban built-up area classification includes various scene types, such as building,
avenue, road, airport, storeroom, roadside tree, residents, bridge, parking lot, and marina images collected from
the RSI-CB dataset. Each image is 128x128 pixels with 3 color channels.

The HSRRS image dataset. (a)-(j) Building, avenue, road, airport,

storeroom, roadside-tree, residents, bridge, parking lot and marina.
2. Data Augmentation:
 Remote sensing images can vary widely due to differences in lighting, angles, seasons, and urban layouts.
 Data augmentation helps create a more diverse dataset by simulating these variations, which makes the model
more robust and capable of generalizing to new, unseen images.
 Augmentation artificially increases the size of the dataset, reducing the risk of overfitting.
 In our project, we used
• Image Normalization
• Scaling
• Rotation
• Width Shifting
• Height Shifting
3.CNN Architecture:
4. CapsNet Architecture:
5. Model Ensemble Architecture:
 Model ensemble refers to the process of combining multiple models to produce a more robust and accurate
final prediction.
 SMDTR (Same Model with Different Training Rounds) involves training multiple instances of the same model
but with different training rounds, resulting in slightly varied models.
 The predictions from these multiple models (SMDTR-CNN and SMDTR-CapsNet) are combined using
weighted averaging.
 The performances of proposed methods are evaluated in terms of overall accuracy, kappa coefficient, precision
and confusion matrix.
Performance Evaluation
 Accuracy: The ratio of correctly predicted instances to the total number of instances.
 Kappa Coefficient: A measure of agreement between predicted and actual classifications, adjusted for
chance
 Precision: The ratio of correctly predicted positive instances to the total predicted positives.
 Confusion Matrix: A table that shows the counts of true positives, true negatives, false positives, and false
negatives for each class.
REFERENCES
[1] Wenmei Li, Haiyan Liu and Guan Gui "Deep Learning-Based Classification Methods for Remote Sensing
Images in Urban Built-Up Areas," IEEE Trans. vol. 7, 2019.
[2] Chen,W. Gong, Y. Chen, andW. Li, "Learning a two-stage CNN model for multi-sized building detection
in remote sensing images,'' Remote Sens. Lett., vol. 10, no. 2, pp. 103-110, 2019.
[3] G. A. Canetti, M. C. Gárrastazu, P. P. de Mattos, E. M. Braz, and S. P. Netto, “Understanding multi-
temporal urban forest cover using high resolution images,” Urban Forestry Urban Greening, vol. 29, pp. 106-
112, Nov. 2018.
[4] P. B. Pan, Z. Shi, and X. Xu, ``MugNet: Deep learning for hyperspectral image classi_cation using limited
samples,'' ISPRS J. Photogramm. Remote Sens., vol. 145, pp. 108-119, Nov. 2018.
Thank you!