Abstract
This study evaluates a lightweight parking area classification framework based on Gray-Level Co-occurrence Matrix (GLCM) feature extraction, Particle Swarm Optimization (PSO) for feature selection, and K-Nearest Neighbor (KNN) classification. Experimental results indicate that PSO achieved 91.04% accuracy using only nine selected features in five evaluations with a processing time of 329 s. A key contribution of this study is the introduction of a lightweight parking-slot classification framework and the comparative evaluation of PSO, Genetic Algorithm (GA), and FOX. Although the GA achieved the highest accuracy of 92.73%, PSO offered the most favorable trade-off between accuracy, feature reduction, and processing time among the evaluated methods. ANOVA results indicate that feature selection significantly affects both accuracy and processing time (p < 0.001). Optimization was conducted under identical experimental conditions and it was found that PSO is the most computationally efficient feature-selection strategy among the three tested methods.

This publication has 46 references indexed in Scilit: