The Role of CV Methods in Drone Image Classification: Analysis of Time and Accuracy Efficiency

Abstract

Drone images are an important source of information for agriculture, green area monitoring, urban infrastructure analysis, and land cover object detection. However, real drone images may contain uneven illumination, shadows, sensor noise, texture variations, and blurred object boundaries, which reduce classification accuracy. This paper analyzes the effect of Gaussian, Median, and Bilateral filtering algorithms on the time and accuracy efficiency of drone image classification. Real drone images are processed in unfiltered and filtered forms, RGB, HSV, and texture features are extracted, and a single Random Forest model is used for classification. The results are evaluated using Accuracy, Precision, Recall, F1-score, IoU, and processing time. The experiments show that filtering improves classification accuracy, while some methods increase computational cost.