應用影像處理技術於蘭花溫室自動化害蟲監測系統

Abstract

蘭花是世界上重要的經濟作物之一，臺灣蘭花產業於2014年有高達1.83億美元的出口產值，且蘭花在臺灣外銷的栽種作物中有相當大的份量。然而蘭花是一種脆弱的栽種作物，蘭花本身相當容易受到病蟲害的侵擾，例如黑翅蕈蠅，導致蘭花生長遲緩或葉片受損，進而影響蘭花外觀美觀與銷售價格。因此為了減低經濟上的損失，本研究致力於發展一套基於物聯網概念下的自動化監測系統以協助防治人員進行害蟲防治，該系統的主要功能是於蘭花溫室中進行環境監測與害蟲監測。在環境監測部分，由無線感測器網路測量環境的溫度與濕度。在害蟲監測部分，透過特定波長的黃色黏紙吸引害蟲，並透過相機模組於固定的時間拍攝影像，拍攝到的影像將於分析後被傳到FTP資料庫內。分析影像時將利用光源修正、適應性二值化、Canny邊緣檢測與減少雜訊等方式計算害蟲數量。此外本研究也設計一個能自動更換捕蟲黏紙的機構，一旦監測得到的害蟲數量超過設定的閥值便會更換捕蟲黏紙。 本研究的實驗每天擷取一份樣本，目前為止我們總共擷取了從2015年8月16日至2015年11月30日的樣本，總共107份樣本，並利用均方根誤差RMSE、相對誤差與信賴區間得出計數誤差蟲數為2.06隻，相對誤差為4.91 %，在95 %的信心水準下相對誤差的信賴區間為0.86 %。實驗結果顯示本研究所設計的害蟲計數演算法具備相當程度的精確性。將來這些感測數據將被用來找出害蟲與環境間的關係，本研究提出的監測系統也可以被用於物聯網中，提供蘭花花農能透過更科學的方法進行相關害蟲防治。

Orchids are one of the essential commercial crops in the world. In Taiwan, according to the official statistics, the orchid has a large proportion of the crop planted in export and the orchid export reached 1.83 million USD in 2014. Moreover, orchids are easily damaged by pests, such as dark-winged fungus gnats (Bradysia sp.), resulting in growth retardation or damaged leaves and eventually influencing the price of orchids. To reduce the loss caused by pests and the cost of manual labor, this study developed an automatic monitoring system for pests based on Internet of Things (IoT) to assist floral famers in pest prevention. Two main functions of the proposed system deployed in an orchid greenhouse nursery are environmental monitoring and pest monitoring. In environmental monitoring, a wireless sensor network is responsible for measuring ambient temperature and relative humidity. In pest monitoring, sticky papers with the yellow color of a specially designed wavelength are utilized to attract pests. A camera module is responsible for capturing the images of the papers in a fixed schedule. Furthermore, the images are transmitted to an FTP server. A pest counting algorithm based on light source correction, adaptive binarization, the Canny edge detection and noise reduction is responsible for calculating the number of pests. In addition, the proposed system can automatically change the sticky papers when the number of the pests on the papers exceeds a designed threshold. From Aug. 16 to Nov. 30, 2015, the proposed system has obtained 107 samples, the root mean square error is 2.06, the relative error is 4.91 % ± 0.86 % at a 95 % confidence level. The experimental results show that the counting accuracy of the pest counting algorithm is high. In the future, the sensing data will be utilized to find the relation between the pests and the environment. It is expected that using the proposed monitoring system can provide prescriptive cultivation decisions to farmers based on IoT such that they can prevent pests by using scientific technologies to obtain high yields.