Scribd
Upload
17 views

SIH2024 IDEA CyberPunk PS

bnvvjmn

Uploaded by

pankajkolage2
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
17 views

SIH2024 IDEA CyberPunk PS

bnvvjmn

Uploaded by

pankajkolage2
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

SMART INDIA HACKATHON 2024

• Problem Statement ID - 1705


• Problem Statement Title - Development
and Optimization of Al model for Feature
identification/ Extraction from drone
orthophotos.
• Theme - Robotics and Drones
• PS Category - Software
• Team ID - WCOE20
• Team Name - Team CyberPunk
Team
CyberPunk AI Model for Feature Extraction from
Drone Imagery
❖ Proposed Solution : -
• Detailed Explanation :
• This is a classic problem of semantic segmentation of aerial imagery data.
• Develop an CNN model to automatically identify key features such as building footprints, road
features, and waterbodies from SVAMITVA drone imagery dataset .
• The model used is derived from U-Net model architecture.

• How it Addresses the Problem :


• The solution will eliminate manual land surveying, reducing human errors and time consumption.
This will create precise digital records and enhance the efficiency of rural land management.

• Innovation and Uniqueness :


• The model will operate on a cloud-based platform, enabling wide-scale accessibility and deployment.
Achieving high accuracy (95%) ensures reliable feature extraction for rural planning purposes.

@SIH Idea submission- Template 2


Team
CyberPunk TECHNICAL APPROACH
• Methodology
• Utilize the SVAMITVA Scheme drone-labeled
datasets for 10 villages to train and validate the
model.
• The U-net model widely used for image segmentation
problems is employed for this project and transfer
learning is used .
• Train the model for extracting various features (roads,
water bodies, roofs, etc.) and then fine tuning it for the
provided dataset.
• Test the model on other drone orthomosaics to test its
performance.
• Technologies Used:
• Programming Languages: Python
• Frameworks & Libraries :TensorFlow, Keras, Django,
GDAL, Geopandas
• Tools: Google Colab, Jupyter Notebooks, Kaggle
• Frontend : HTML5, CSS, JS

@SIH Idea submission- Template 3


Team
CyberPunk FEASIBILITY AND VIABILITY
• Feasibility
• Access to accurate and high-resolution SVAMITVA drone data, which provides clear and labeled
images.
• Existing technology, frameworks and cloud infrastructure, supports model development and
deployment.
• Challenges
• Achieving a target accuracy of 95% while maintaining efficient processing speed requires careful
tuning of the model.
• Model size should be as small as possible for implementing on edge devices or browsers.
• Overcoming Strategies
• We will optimize the model by reducing complexity fine-tuning and tuning hyperparameters.
• Usage of regularization techniques (dropout layers, L1 & L2 regularization) and model pruning.

@SIH Idea submission- Template 4


Team
CyberPunk IMPACT AND BENEFITS

• Impact :
• This supports better rural planning, such as solar
installations and property tax assessments, and aids
in resolving property disputes, thereby benefiting the
community.

• Benefits :
• By improving the accuracy of land records, the
solution will help property owners access bank loans
and enable village-level development projects.
• It will also enhance the quality of life in rural areas by
supporting sustainable development and efficient
resource management.

5
Team
CyberPunk RESEARCH AND REFERENCES

• AI-Assisted Manual Segmentation Web Application for Geospatial


Satellite and Imagery Data - Lingyan Rana, , Cheng Jib, Shizhou Zhangb,c, Xiaoqiang
Zhangd , Yanning Zhang
• An Unsupervised Domain Adaption Framework for Aerial Image
Semantic Segmentation Based on Curriculum Learning - Xun Thong Cham,
Chen Fung Choy Oliver
• A brief introduction to satellite image segmentation with neural
networks – Robin Cole ( Medium)

@SIH Idea submission- Template 6

You might also like