Scene classification is one of the important tasks in the analysis of remote sensing images. This article examines the application of machine learning algorithms for forming image descriptors and classifying scenes. Within the scope of the study, low-, mid-, and high-level descriptors were used, including Local Binary Pattern (LBP), Histogram of Oriented Gradients (HOG), Scale-Invariant Feature Transform (SIFT), Oriented FAST and Rotated BRIEF (ORB), BRISK, and BRIEF descriptors. For scene classification, SVM, Random Forest, Decision Tree, Naive Bayes, kNN, and Logistic Regression algorithms were employed. The results of the study demonstrated the effectiveness of the combinations of descriptors and classifiers, with the SIFT descriptor providing the best results. However, the overall results indicated the need to improve methods to achieve higher accuracy in scene classification. This study explores the development and application of algorithms based on low- and mid-level descriptors for scene classification in remote sensing images. For accurate image classification, it is important to form descriptors that express key attributes (such as image dimensions and formats), visual features (such as color and textures), semantic information (contextual relationships), and geometric consistency. The study consists of two main stages: feature extraction and training a classifier based on machine learning algorithms (kNN, Naive Bayes, Random Forest, SVM, and Logistic Regression). The study was conducted on the NWPU-RESISC45 dataset, and classification was performed across 45 different scene categories. The results show that SIFT and other descriptors possess high accuracy, but further research is required to improve classification performance.
| Mualliflar | Yusupov, O.R., Xandamov, Y.X., Eshonqulov, E.Sh. |
|---|---|
| Jurnal | Рақамли технологияларнинг назарий ва амалий масалалари |
| Nashr sanasi | 2025-04-06 |
| Jild | 8 |
| Son | 1 |
| Betlar | 7-21 |
| Til | Rus |
| DOI | 10.62132/ijdt.v8i1.229 |
DOI: 10.62132/ijdt.v8i1.229 · Maqolaning asl sahifasi
sahnali tasniflash, masofaviy zondlash tasviri, belgilar, deskritor, klassifikator, NWPU-RESISC45, scene classification, remote sensing image, feature, descriptor, classifier, NWPU-RESISC45
The paper presents a summary of the work done on the application of underground mining methods in the backfilling of ore deposits. In addition, basic information for future research is provided. The analysis of the…
In this research, the splitting into physical processes approach is used to numerically predict the propagation of radioactive contaminants in the environment. The movement, diffusion, and absorption of radioactive…
In our study, we developed and tested a comprehensive approach to assess flood risk zones in Samarkand city by integrating data from digital elevation models of different resolutions (NASA SRTM - 30 m, ALOS PALSAR DEM…
The mobile communication industry has seen rapid growth in the number of users, and the globally most popular GSM network is facing a number of security vulnerabilities. While some of these issues have been addressed in…
Modeling of nonlinear filtration processes in multilayer heterogeneous porous media is a complex area that combines various mathematical and physical principles to understand the dynamics of fluid in porous structures…
A computer-aided diagnosis system model is developed using fuzzy subsets. In this case, symbols are treated as fuzzy subsets of the corresponding data set. The degree of relevance of a value in the fuzzy subset of a…
In this article developed a numerical algorithm for solving a multidimensional mathematical model of joint heat and moisture transfer processes in heterogeneous porous bodies shaped as rectangular parallelepipeds. The…
This research paper proposes that integrated approaches to image processing, the use of machine learning algorithms in segmentation, the U-Net method will help to achieve high accuracy in diagnosing various retinal…
The article discusses the development of an algorithm and software for calculating the hydraulic resistance coefficient of complex gas pipeline systems. The methodology for selecting the initial data, the mathematical…
The work is devoted to the development and evaluation of a situational cause-and-effect model of the gas distribution network performance indicators in emergency situations. A method for modeling various emergency…
Рақамли технологияларнинг назарий ва амалий масалалари — barcha maqolalar