Paper Key : IRJ************373
Author: Kavana P
Date Published: 04 Jul 2024
Abstract
The paper addresses the significant issue of detecting illegal buildings using satellite imagery, a critical concern for urban planning and regulation. This involves the complex challenge of interpreting remote sensing data and verifying it against cadastral maps. To tackle this, you propose a methodology that combines imageprocessing techniques with deep learning tools to automate the detection process. The methodology consists of three main components: region proposal generation, building classification, and legality verification. Region proposals are generated by integrating four computer vision-based techniques, including edge-detection methods like Canny-edge-detection to identify building boundaries, and segmentation methods such as thresholding and clustering (e.g., K-means) to distinguish buildings from their surroundings. Then these praposals are classified using a retrained GoogLeNet convolutional-neural-network (CNN), and finally, the detected building areas are validated for legality against state cadastral maps. Your tests with satellite imagery the datasets demonstrates that this approach achieves acceptable results in both building detection and legality assessment.