Real-Time Vibration Isolation Strategies for Vision-Based Agricultural Drones Using Optical Flow and Spectral Analysis

Unmanned Aerial Vehicles (UAVs) and computer vision based systems are widely used for real-time crop health monitoring in precision agriculture. However, the vibration-induced image blur is still one of the difficult problems for the automatic detection algorithms. A comprehensive evaluation of five mounting methods of the camera for the vibration damping in a lightweight agricultural drone is demonstrated. Based on real-world flight testing over cotton fields, we evaluated the performance of various isolation strategies in terms of Lucas-Kanade optical flow and Fast-Fourier Transform (FFT) spectral analysis. From basic bar to our latest double bar design rubber mounts to cush drive rubber mounts, the evolution denied several other options. The results indicate that the two-stage vibration dampers led to reduction of 88% in peak vibration amplitude of the baseline and enhanced the object detection of YOLOv10s by 53% to 88.5% mAP@50 in the nutrient deficiency identification. These results present useful guidelines for practical use of robust vision-based agricultural monitoring systems under actual field conditions.