AI-BASED CHANGE DETECTION DUE TO HUMAN ACTIVITIES AND FEATURE EXTRACTION SYSTEM FOR REMOTE SENSING IMAGES USING TIME SERIES DATA

Abstract

This invention presents an AI-driven system for automated change detection and feature extraction in remote sensing (RS) images using spatial and temporal data analysis. The system addresses the need for efficient and accurate identification of geographic changes over time, leveraging satellite imagery from sources like Bhuvan and NRSA. It utilizes deep learning models: a Convolutional Neural Network (CNN) for spatial feature classification and a Convolutional Recurrent Neural Network (ConvRNN) for detecting temporal correlations across different timestamps. The system is supported by a FastAPI-based server and a structured database (PostGIS and PostGreSQL) that stores, organizes, and enables quick retrieval of RS images. Users interact with the system through a web-based interface where they can specify locations and time spans to query changes. Results are displayed with semantic segmentation, extracted features, and a GeoJSON-formatted visualization, providing actionable insights. This innovation streamlines the traditionally manual and labor-intensive process of change detection, offering a scalable, automated solution that enhances decision-making in fields like environmental monitoring, urban planning, and disaster response. The system’s scalable architecture includes Docker containers for deployment, a Redis message broker for managing model-server communication, and robust data security. By integrating advanced AI models with accessible web-based tools, this invention provides an effective approach for large-scale geographic data analysis, making it highly applicable for use by organizations like ISRO in monitoring and managing regional changes over time.

Specification

Description:Field of the Invention
The invention lies at the intersection of computer science and remote sensing technology, particularly focusing on applications of artificial intelligence (AI) and deep learning in satellite image analysis. The invention is designed for automated change detection and feature extraction in remote sensing (RS) images by integrating spatial and temporal data processing techniques. This invention is suitable for numerous applications, including urban planning, environmental conservation, resource management, and disaster monitoring, where precise change tracking over time is critical.

Background of the Invention
Remote sensing technology provides significant data on geographical regions, capturing valuable insights into Earth's surface over time. While manual methods for analyzing changes in RS images exist, they are often limited in speed, accuracy, and scalability due to the sheer volume of data and the need for rapid information processing. Current AI-based systems typically focus on either spatial or temporal analysis separately, limiting their effectiveness in dynamic environments where both spatial and temporal elements are crucial for accurate change detection. For organizations like the Indian Space Research Organization (ISRO), which manage extensive satellite datasets, there is an urgent need for a system that can automate the detection of changes across geographic landscapes by analyzing both the spatial distribution of objects and their temporal evolution.

Summary of the Invention
This invention introduces an AI-based system designed to automate change detection and feature extraction in RS images. By employing two advanced deep learning models-a Convolutional Neural Network (CNN) for spatial analysis and a Convolutional Recurrent Neural Network (ConvRNN) for temporal analysis-the system provides accurate, real-time insights into geographic changes across multiple timestamps.

The system retrieves high-resolution satellite images from sources such as Bhuvan, NRSA, and other RS data repositories, storing them in a structured database for quick access. CNNs are used to identify spatial characteristics and classify various objects within each image. Concurrently, the ConvRNN model processes temporal correlations, detecting changes over time in specific regions. This dual-model approach enables precise detection of new structures, deforestation, urban expansion, and other changes in land usage.

Accessible through a user-friendly web interface, users can specify a location and time span to query the system. The results are formatted in a GeoJSON structure and displayed with detailed segmentation, object classifications, and a clear visual representation of changes. A FastAPI-based server with Docker containerization facilitates the processing, with a Redis message broker ensuring efficient communication between components.

Detailed Description of the Invention
The invention comprises several interlinked components:

1. Data Collection and Storage:
• Data Sources: High-resolution RS images are obtained from platforms such as Bhuvan and NRSA, specifically using LISS III and AWiFS images, which capture multi-band data.
• Storage Architecture: Images are stored in a PostgreSQL database with PostGIS extensions for efficient spatial and temporal data handling. This structure enables rapid data retrieval based on user-specified locations and timeframes.

2. Deep Learning Models:
• Convolutional Neural Network (CNN): This model is used to analyze the spatial structure of each image, identifying and classifying objects of interest, such as buildings, water bodies, vegetation, and infrastructure.
• Convolutional Recurrent Neural Network (ConvRNN): Designed to handle time series data, the ConvRNN model processes images from different timestamps, detecting changes by analyzing temporal correlations within image sequences. This enables the system to monitor gradual or sudden alterations in geographic regions accurately.

3. Server and Processing Infrastructure:
• Backend Server: The backend is built on FastAPI, which provides a fast and efficient framework for managing API requests. Docker containers encapsulate the AI models, enhancing scalability and making it easier to deploy updates without service interruptions.
• Message Broker: A Redis message broker manages query exchanges between the web server and AI models, ensuring low latency and efficient communication, which is essential for handling large datasets and real-time user queries.

4. User Interface:
• Web-Based Dashboard: Through an interactive dashboard, users input location coordinates and timeframes for querying. The interface displays output data with segmented images, labeled objects, and detected changes formatted in GeoJSON, enabling easy integration with GIS applications for further analysis.

5. Data Processing and Feature Extraction:
• After initial preprocessing, including manual labeling and standardizing image formats, the CNN model performs pixel-level labeling for object classification, while the ConvRNN model detects changes in spatial data across time intervals, effectively performing semantic segmentation and change detection.

6. Output Generation:
• GeoJSON Format: Results are outputted in GeoJSON format, allowing easy overlay with other geospatial data systems. This structure enables users to visualize changes geographically and extract actionable insights for planning and response strategies.

Technology Stack used for Implementation
The system leverages modern frameworks and technologies to support high-speed processing and scalable deployment:
• Technology Stack: Uses Python with FastAPI for server management, Docker containers for deploying models, and Redis for message handling. GIS libraries and Plotly are employed for visualizing segmented results and mapped changes.
• Challenges Addressed: The model accounts for variable image sizes and atmospheric conditions, such as cloud coverage, by integrating robust training datasets and adapting to diverse satellite image formats (HDF5, multi-band).
• Scalability and Resource Management: Designed to scale across cloud-based environments, this system can manage extensive RS datasets, providing rapid, accurate change detection suitable for real-time applications in both urban and rural monitoring scenarios.
, Claims:We Claim
1. Claim 1: A method for automated detection and classification of changes in remote sensing images, comprising:
• Acquiring satellite images for a designated location and specified time intervals;
• Employing a CNN to classify and label objects within each image based on spatial data;
• Utilizing a ConvRNN to analyze temporal changes across multiple timestamps, detecting alterations in the spatial data.

2. Claim 2: The system according to claim 1, wherein the ConvRNN model is trained to detect temporal correlations across high-resolution satellite images, providing accurate change detection for various types of environmental and infrastructural data.

3. Claim 3: A scalable storage and retrieval solution for remote sensing data using PostGIS and PostgreSQL, enabling efficient spatial and temporal queries for high-volume RS image datasets.

4. Claim 4: A web-based interface allowing users to input specific geographic coordinates and time periods, where results include segmented images with object classifications, temporal changes, and output in GeoJSON format for geospatial analysis.

5. Claim 5: A system of containerized deployment for deep learning models, utilizing Docker containers and a Redis message broker to facilitate scalable and real-time processing of remote sensing data.

Documents