Semi-automatic Labeling of Satellite Images Using Texture Features and Hough Circle Transformation

• Published in: SN computer science 2024-06, Vol.5 (5), p.516, Article 516
• Main Authors: Sundari, L. K. Sowmya, Mallikarjuna, M. K., Halakeri, Pooja, Hebbar, Ramachandra
• Format: Article
• Language: English
• Subjects: Accuracy; Advances in Computational Approaches for Image Processing; Algorithms; Annotations; Automation; Circles (geometry); Classification; Cloud Applications and Network Security; Computer Imaging; Computer Science; Computer Systems Organization and Communication Networks; Crops; Data Structures and Information Theory; Decision making; Deep learning; Efficiency; Environmental monitoring; Food security; High resolution; Image classification; Image enhancement; Image resolution; Information Systems and Communication Service; Labeling; Land use; Methods; Neural networks; Original Research; Pattern Recognition and Graphics; Remote sensing; Sampling methods; Satellite imagery; Software Engineering/Programming and Operating Systems; Spatial filtering; Subtraction; Unmanned aerial vehicles; Urban planning; Vegetation; Vegetation index; Vision; Wireless Networks
• ISSN: 2661-8907; 2662-995X; 2661-8907
• DOI: 10.1007/s42979-024-02834-0

In order to extract valuable ground information from images, supervised classification is an extensively adopted technique. However, it has a substantial downside in that choosing and labeling training samples requires a lot of time and resources. Fortunately, recent developments in automatic information indexing have produced encouraging this development makes it possible to develop an automated method for labelling training samples as opposed to relying just on human interpretation. In this research paper, we present a unique technique for the automated recognition and annotation of training data for high-resolution image classification which will identify tree and non-tree. The proposed methodology consists of methods like pre-processing to enhance the quality, identifying features where distinctive characteristics are extracted from the images, applying NDVI a fundamental vegetation index which quantifies the presence of healthy vegetation, applying a 7*7 mask, spatial filter to enhance specific features in the image, image subtraction to highlights discrepancies between images and Hough circle transformation a robust method for circular shape detection to identify circular patterns associated with tree canopies detecting tree region. The proposed sampling methodology has been assessed using a series of high-resolution remote sensing images. The results indicate that the proposed method can reliably generate a large number of samples and can achieve satisfactory results when used with various classifiers. This research makes a significant contribution to high-resolution image classification by automating the labeling of training data, with a particular emphasis on tree identification.