應用於溫室環境作物監測之自主導航無人機多機協同系統

Abstract

本研究開發了一套自主多機無人機系統，旨在進行溫室中洋香瓜作物的生長監測。透過多機無人機的協同作業，進行了多角度的影像拍攝，並設計了三種不同的飛行路徑。利用UWB定位系統進行無人機自主飛行的精度比較，並利用收集的影像進行三維重建，對洋香瓜植物進行分析。收集的影像經過處理後，用於提取關鍵的植物表型數據，本研究主要分析了植株的高度和展幅，透過高度和展幅進一步對植物的生長進行監測，通過擬合生長曲線，並將其與實際生長數據進行比較。在無人機的飛行精度方面，平行飛行路徑的誤差範圍為7至12公分，閉環飛行路徑為5至9公分，而多高度路徑的誤差範圍為4至11公分，顯示出穩定的飛行精度。接著，我們進一步比較了在相同覆蓋面積下，多機系統與單機系統的效能。結果顯示，多機系統能夠將任務時間縮短73%，並將電池消耗降低5%。在作物三維重建方面，我們比較了三種軌跡所收集的三種不同的重建方法，分別是單面、合併以及三個高度。根據評估重建結果的指標，使用PSNR、SSIM和LPIPS三個指標進行比較，結果顯示三個高度方法在重建質量上表現最佳，PSNR為0.37，SSIM為9.48，LPIPS為0.65。在植物高度的測量上，合併方法達到了最低的MAE誤差為6.6公分，而在展幅測量方面，單面方法則達到了最低的MAE誤差為5.8公分。本研究展示了多機無人機系統在溫室作物監測中的應用潛力，還證明了不同重建方法和測量策略在提高農業監測精度和效率方面的有效性。

This study developed an autonomous multi-drone system for muskmelon crop growth monitoring in a greenhouse. Through collaborative multi-drone operations, multi-angle images were captured and three flight paths were designed. The UWB positioning system was used to compare the accuracy of the UAV autonomous flight, and the collected images were used for 3D reconstruction to analyze muskmelon plants. The images were processed to extract key phenotypic data, focusing on plant height and canopy span, and growth monitoring was performed by fitting growth curves and comparing them with actual growth data. In terms of flight accuracy, the parallel flight path had an error range of 7 to 12 cm, the closed-loop path had an error range of 5 to 9 cm, and the multi-altitude path had an error range of 4 to 11 cm, demonstrating stable flight precision. We also compared the performance of multi-drone and single-drone systems over the same coverage area. The multi-drone system reduced mission time by 73% and battery consumption by 5%. For 3D reconstruction, we compared three methods collected along three different paths: Single-side, Merged, and Three-height. Evaluation metrics showed that the Three-height method provided the best reconstruction quality with PSNR of 0.37, SSIM of 9.48, and LPIPS of 0.65. For height measurement, the Merged method achieved the lowest MAE of 6.6 cm, and for canopy span measurement, the Single-side method achieved the lowest MAE of 5.8 cm. This study demonstrates the potential of multi-UAV systems in greenhouse crop monitoring and proves the effectiveness of different reconstruction methods and measurement strategies in improving monitoring accuracy and efficiency.