Comparative Analysis of Classical Feature Detection Methods for UAV-Based Tomato Detection

Abstract

Remote sensing technologies, especially Unmanned Aerial Vehicles (UAVs), become crucial for Precision Agriculture (PA) to perform different tasks such as crop detection, yield prediction, leaf disease diagnosis, weed detection, and harvest forecasting to ensure higher productivity. Therefore, using various image analytics methods feature detection from the UAV-captured images plays a vital role for conducting these PA tasks. To enhance the effectiveness of the UAV-based image analysis, this study investigates the performance of various classical feature detection algorithms on the UAV-captured images of tomato fields. This study also identifies the standard benchmarks of the evaluation metric used in the feature detection methods. The evaluation considers challenging conditions such as rotation, illumination variation, and scaling. Results show that Oriented FAST and Rotated BRIEF (ORB) and Speeded Up Robust Features (SURF) among the classical methods demonstrated better performance under all these environmental conditions. However, considering the limitations in existing feature detection techniques this study also suggests that integrating classical feature detection with deep learning approaches could significantly improve real-time feature detection efficiency.