請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/90727
標題: | 超甜玉米生產之生理支持型專家系統之建立 Establishment of Physiology-based Decision Supporting System in Super Sweet Corn Production |
作者: | 姚箴 Chen YAO |
指導教授: | 盧虎生 Huu-Sheng Lur |
關鍵字: | 甜玉米,資料驅動之校正法,決策輔助系統,物候學,生育有效積溫,智慧農業,脈絡訪查, Contextual inquiry,Data-driven optimization,Decision support system,Growth degree day,Phenology,Smart agriculture,Sweet corn, |
出版年 : | 2023 |
學位: | 碩士 |
摘要: | 超甜玉米是臺灣重要的雜糧作物,佔據國內食用玉米將近6成5的生產量。在臺灣的超甜玉米中,主要包含三大類:甜軟殼、甜硬殼、超甜白,又以超甜白(又名:水果玉米)為最高經濟價值的品項,其品質不僅極度容易受到天氣變化影響,也對採收期十分敏感。然而,目前國內尚缺乏針對超甜白玉米農企業實務操作難點之智慧系統,故本試驗目的即透過分析實際需求,建立數位系統來解決產業經營問題。
通常在玉米果穗發育的R3到R4階段中,糊粉層中色素堆積造成轉色。轉色是農民判斷收穫的重要依據。然而,超甜白玉米的籽粒為白色、並無轉色發生,是故使得農民難以判斷適採期。一般而言,可以透過線性積溫法來建置對應的產期預測模式,但臺灣地屬熱帶,夏季時線性積溫法並無法反應正確的作物生長情況,致此無法滿足產業全年栽培的操作需求;引入對高溫有修正的有效積溫式或許是可行解方,但現狀帶來的其它挑戰卻也隨之而來──不僅現場品種與文獻中不同、須挑出合適的作物參數外,同時超甜白玉米種子源為從國外進口、遺傳背景常有變動,產業上也常有試種新的品種的市場需要。這些複雜、立體的問題結構,對產期預測之建立帶來十分嚴峻的挑戰。 為克服現狀,本研究除引入梯形積溫法外,更開發了一個可利用現場生育資料來校正作物參數之參數最佳化演算法,藉此保證作物參數的效力。經最佳化之積溫所得參數為Tbase = 10℃、Topt1 =24℃、Topt2 = 33℃、Tmax = 37℃,在超過300個田間數據點的驗證中,此方法顯著地優於原先產業使用之線性積溫法的效果,R square從0.77提升至0.92,預測值的標準差亦從±6.7天縮減至±3.2天。此外,本研究亦針對實際應用場景來做更進一步的測試。模擬結果中顯示:產期預測之誤差在進入輪生期到開花期時收斂達到穩定,代表此時資訊以開始足夠可靠。而在學習新品種的模擬中,從3個物候資料位點開始,搭配著參數最佳化之此方法流程便恆優於現行使用之線性積溫法。 最後,本試驗再根據於脈絡訪查法(Contextual Inquiry)探訪之脈絡,規範出設計限制,並以此設計系統架構、使用場景、技術導入,再以R shiny套件開發最小可行性產品(Minium Viable Product)之原型(Prototype)。功能上以產期預測為核心,除了提供田區之生育期查詢外,並能生成可匯入至Google日曆的巡田提醒文件,藉此嵌入於鮮綠農產田間人員現行工作流程中。於資料輸入上,則可與農企業內部文件相容外,亦在格式與行為操作上考量了不同的補救機制,來讓系統在發生問題時能依然具有能夠運作的韌性。 本系統為第一個國內甜玉米之數位智慧系統,有潛力以更精準之產期預測從而改善現行產業中的操作難點。同時,本研究也是當今農業智慧系統設計中第一個使用脈絡訪查法的案例,提出之最佳化演算法亦是物候學領域中第一個能利用田間既有資料來校正多個積溫參數之演算方法。期待透過此研究之提出,成為未來其他應用之拓展上重要的基石,並與更多場域合作,一起為現場難題創造解方。 Sweet corn is an important vegetable crop in Taiwan, with approximately 65% of the total production falling under the category of super sweet corn. Taiwan's super sweet corn can be classified into two major types: tian-rang-ke (yellow-kernel super sweet corn) and chao-tian-bai (specifically referring to white-kernel super sweet corn). Chao-tian-bai, known for its higher unit price, faces quality challenges due to its sensitivity to harvest timing, posing difficulties in cultivation. However, there is currently a lack of digital decision support tools specifically for maize in Taiwan. The study aims to develop a decision support system to overcome existing challenges. The color-changing event, resulting from the accumulation of pigments in the aleurone layer during the transition between R3 and R4 stages, holds significant importance for farmers in accurately determining the optimal harvesting time. However, in the chao-tian-bai, the kernels are white and do not accumulate pigments. This lack of pigment accumulation poses a challenge for farmers in determining the developmental stage without the presence of color-changing events, making it difficult to ensure the quality of the crop. Establishing a corresponding phenological stage model is an integral part of addressing this situation. However, Taiwan belongs to a tropical region where the linear growing degree-day (GDD) method fails to accurately reflect crop growth during high-temperature in summers, resulting in suboptimal predictive performance that does not meet industry needs. To overcome this, we introduced the trapezoidal GDD method and developed a parameter optimization algorithm, which is based on the greedy method, using existing phenological data to select appropriate parameters. During validation with over 300 field data points, the new method showed substantial improvement over the linear growing degree-day (GDD) method. The R-square value increased from 0.77 to 0.92, and the standard deviation of predicted values decreased from ±6.7 days to ±3.2 days. In further testing, it was found that the prediction errors for the harvest period converged and stabilized after the whorl stage and the flowering stage, indicating the utility of predictions during the growing season. A simulation was also conducted to evaluate the algorithm's performance in the context of learning new varieties. It was observed that starting from three phenological data points, the proposed algorithm consistently outperformed the conventional linear growing degree day method currently in use. In addition, a contextual inquiry was conducted for a year and a half, involving user interviews with stakeholders in the agricultural industry. The insights obtained from this process were used to build a comprehensive framework for the decision support system. Subsequently, we developed a minimum viable product (MVP) prototype using the shiny package in the R language. The MVP prototype was evaluated through user feedback and iterative testing, ensuring that it meets the requirements and expectations of the stakeholders. This decision support system is the first of its kind in the maize production in Taiwan and has the potential to enhance cultivation practices, optimize harvest timing, and ultimately improve crop quality and economic outcomes for maize farmers. Additionally, the developed algorithm in this study holds promise in rapidly assessing the phenological stages of unknown maize varieties, making significant contributions to the exploration of new varieties. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/90727 |
DOI: | 10.6342/NTU202303877 |
全文授權: | 同意授權(全球公開) |
顯示於系所單位: | 農藝學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-111-2.pdf 此日期後於網路公開 2025-08-14 | 2.8 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。