透過無人機早期多光譜影像判斷水稻植株位置

Plant Position Determination of Rice by Early-Stage Multispectral Images Obtained from Unmanned Aerial Vehicles

裴昭如(Jauru Pei)

指導教授 ： 劉力瑜

國立臺灣大學/生物資源暨農學院/農藝學研究所/碩士(2019年)

https://doi.org/10.6342/NTU201901190

摘要

植物密度對於作物生長是一重要的影響因子，其可影響作物產量、水分吸收、肥料需求，甚至是對病蟲害的敏感度。前人研究指出，決定最佳的植物密度及行株距對作物而言是一重要的管理決策。而在傳統上，植物密度可由人力至田區中直接以目視進行植株個數計算進而得知，但以人力直接進行田間調查是耗時且費力的。隨著科技發展，空拍影像也是幫助獲取植物密度的一個新方法。相較於更為傳統的衛星影像，無人機影像的解析度提升，並且可依使用者需求訂定飛行高度及路線。本研究對桃園農業改良場四月初期之水稻田區多光譜影像，分別利用像素影像分析及物件影像分析，透過常態化差值植生指標計算田區中的植株位置及個數，藉以推估田區之植物密度。在本文的方法中，像素影像分析主要是以分群法來進行；而物件影像分析，則利用eCognition 中的多尺度影像切割及回歸線來輔助進行。研究結果顯示，像素影像分析以分群法估計田區影像的植株位置之數量而言，可達到85.56 % 的準確率，並且可看出田區中植株初期生長較不佳的作物位置；而物件影像分析準確率則可達96.01 %，並從中可得知缺株位置個數。雖然兩種方法仍有部分需人為操作，但我們的結果提供了對於無人機空拍影像分析之部分自動化的一些可能性，亦有機會用於計算植物密度。

關鍵字

無人機 ； 多光譜影像 ； 植物密度 ； 植生植標 ； 影像分析

並列摘要

Plant density is a crucial factor influencing crop yields, water assimilation, and fertilizer requirement as well as sensitivity to pathogens. Previous studies presented that determining optimal plant density and row spacing is a critical management decision for crop production. Traditionally, plant density can be obtained from labors by visual counting of crops in field directly, whereas it is time-consuming and laborious. With the advance of technology, aerial images become a new approach to assist in precision agriculture and acquisition of plant density. Compared to the conventional satellite images, images of unmanned aerial vehicles (UAVs) offer higher resolution, and the flight height and route can be defined by the request of users. Several previous studies have investigated the crop growth in the field by UAV images. This study utilized the pixel-based image analysis and object-based image analysis to analyze the multispectral rice field images of Taoyuan District Agricultural Improvement Station. With the Normalized Difference Vegetation Index, the number of position of rice in the field could be computed, and the plant density of the field could be known. There were two primary methods in our study, pixel-based image analysis was mainly conducted by density-based clustering, while object-based image analysis exploited the multiresolution segmentation algorithm in eCognition and the assistance of regression lines. According to the number of field crop positions, results presented that the accuracy of clustering method reached 85.56 %, and this method could previously detect the crop positions requiring the attention from managers. In terms of regression line method, the accuracy achieved 96.01 %, and number of the miss-planted positions could be found. Although manual operation was demanded in this work, our results still provided possibilities of partially automated analysis of UAV images, and it also had opportunity to evaluate plant density.

並列關鍵字

unmanned aerial vehicle(UAV) ； multispectral images ； plant density ； vegetative index ； image analysis

參考文獻

Assefa, Y., Vara Prasad, P. V., Carter, P., Hinds, M., Bhalla, G., Schon, R., ... & Ciampitti, I. A. (2016). Yield responses to planting density for US modern corn hybrids: A synthesis-analysis. Crop Science, 56(5), 2802-2817.

Baatz M. and Schäpe A. (2000). “Multiresolution segmentation: an optimization approach for high quality multi-scale image segmentation,” in Angewandte Geographische Informationsverarbeitung XII, Karlsruhe, Germany, pp. 12–23

Baloch, A. W., Soomro, A. M., Javed, M. A., Ahmed, M., Bughio, H. R., Bughio, M. S., & Mastoi, N. N. (2002). Optimum plant density for high yield in rice (Oryza sativa L.). Asian J. Plant Sci, 1(1), 25-27.

Blaschke, T. (2010). Object based image analysis for remote sensing. ISPRS journal of photogrammetry and remote sensing, 65(1), 2-16.

Barmeier, G., & Schmidhalter, U. (2016). High-throughput phenotyping of wheat and barley plants grown in single or few rows in small plots using active and passive spectral proximal sensing. Sensors, 16(11), 1860.