An Iterative Pixel-Based Dimensional Voting Model for High Spatial-Resolution Image Classification

Understanding land use and land cover changes is crucial for effective environmental management, particularly in mixed-land zones and urban areas, which exhibit distinct characteristics influenced by diverse factors. Rapid urbanization, land utilization patterns, green space preservation, clouds, shadows, pollution, and dynamic human activities pose significant challenges in accurate classification. Traditional classification methods often struggle due to ineffective stand-alone data classification, high costs associated with data fusion, and the need for frequent data collection. To address these issues, this paper proposes an Iterative Pixel-Based Dimensional Voting model (Pixel-DMV), which enhances classification accuracy by iteratively analyzing pixel similarities within 3 x 3 neighborhood pixels. The model assigns a label to unknown pixels based on computed similarity measures and ranks to predict the class for the unknown pixel. The performance is then measured using statistical indices such as Overall Accuracy (OA) and the Kappa Index of Agreement (Kappa). Pixel-DMV outperformed conventional methods, including Support Vector Machine (SVM), Random Forest (RF), and Maximum Likelihood (ML) classifications. The findings indicate that Pixel-DMV simplifies the classification process by relying on a single dataset, reducing data preparation costs, and is suitable for frequent data collection tasks. Given its high accuracy, the proposed model is well-suited for applications in agricultural management, urban planning, and disaster response.