應用機器視覺與模糊類神經網路於萵苣採收時間與品質估測之研究

Study on Estimation of Harvest Time and Quality of Lettuce Using Machine Vision and Fuzzy-Neural Network

傅偉倫(Fu Wei-Lun)

指導教授 ： 張仲良

國立屏東科技大學/工學院/生物機電工程系所/碩士(2019年)

https://doi.org/10.6346/NPUST201900343

摘要

因應環境氣候變遷，如何調整產期、穩定產量以及產值是目前農業作物生產端所需面臨的問題，現有農作物的採收時間點以及生長品質大多由種植者依靠其經驗法則決定。而這樣的方式也需要消耗大量的農力資源來觀察作物生長狀態。因而本研究提出了基於模糊-類神經網路技術的植株生長以及品質估測系統，結合了環境數據，用來預測作物收成的時間及其品質，提供農人作為作物產期調整的建議。在生長估測系統中，利用多組模糊估測器以及一組類神經網路估測葉片數葉面積及乾重等基本資料，隨後再將這些資料送進模糊類神經中訓練並推測植株的收成日期以及其品質。此外，本研究也應用機器視覺技術，於植物生長期間，即時辨識植株的葉片數目與葉面積，並同時比較於生長估測系統所輸出估測值的差異。為計算估測的收成日期以及品質之準確率，評定標準採以專家知識所給予的資料為原則做為比對。經實驗證明由估測系統估測出的收成日與實際收成日相比相差5日左右，而品質估測方面，其準確率介於70%~99%之間。至於在機器視覺識別性能方面，葉片數識別率可達93.47%。

關鍵字

生長模型 ； 品質估測 ； 模糊邏輯 ； 類神經網路 ； 機器視覺

並列摘要

Global climate change has become a fact, and farmers have to seek solutions in fixing plant harvest period, stabilizing quantity and quality. Personal experience is the main principle to judge and decide timing of harvesting plants, meaning the farm owners will have label the harvest date and quality. However, this approach is time-consuming out of limited manpower and resources. This study introduces a fuzzy-based neuron network inference system to predict the harvest date and quality of plant combining the environmental factors to generate estimated harvest date and plant quality, for the reference of farmers. In the fuzzy inference system and neural network, the environmental factors and planting time are used to estimate the plant growth data such as number of leaves, leaves area and dry mass. Thus, such data is sent to the fuzzy-neural network to train and generate estimated harvest date and crop quality. Meanwhile, machine vision system can also identify the number of leaves and leaf area. Evaluation factor for harvest and the quality of plant are defined by horticultural expert to assure the accuracy of estimation. It is proven through experiments that the estimated harvest date generated by system having an error less than 5 days in contrast to the actual harvest time based on expert knowledge. The estimated error of plant quality ranges between 70 to 99%. The accuracy rate of number and leaves area estimated by machine vision method is 93.47%.

並列關鍵字

plant growth model ； quality estimation ； fuzzy logic ； neural network ； machine vision

參考文獻

[1]田峻墉，具高效與節能之農業物聯網平台設計與實現，碩士論文，國立屏東科技大學，生物機電工程系，2018年7月。.

[2]C. L. Chang, W. L. FU, and J. Y. TIAN, 2018, “Design of a fuzzy logic-based plant quality estimation system base on morphological feature,” Int. Symp. Machinery and Mechatronics for Agricultural and Biosystems Engineering (ISMAB 2018), Korea, July 2018.

[3]陳曉軍，基於蒸騰速率的溫室滴灌控制系統研究，碩士論文，浙江工業大學，機械電子工程，2010年。

[4]張治安、楊福、陳展宇、徐克章，菰葉片淨光合速率日變化及其與環境因子的相互關係，中國農業科學，第3期，P502-509，2006年。

[5]E.J.V. Henten, “Validation of a dynamic lettuce growth model for greenhouse climate control,” Agricultural System, Vol. 45, pp. 55-72, 1994.