有機肥料熟成影像辨識與人工智慧最佳化處理技術

Image Recognition and Artificial Intelligence Optimization Processing Technology of Organic Fertilizer Maturation

林禹宏(Lin Yu-Hong)

指導教授 ： 蔡正發

國立屏東科技大學/管理學院/資訊管理系所/碩士(2022年)

https://doi.org/10.6346/NPUST202200419

摘要

政府積極推廣有機農業，國內外學者也積極研究相關議題，希望藉由有機農業減少農業對環境之影響，然而農民與廠商卻因有機農業所需投入成本較高而卻步，故本論文以降低有機肥料生產成本之角度切入此問題，藉由提供一高精度且低門檻之肥料熟成辨識技術降低有機肥料生產時判別熟成所需之檢測費用與等待時間，鑒於本論文之時間有限與各有機肥之配方、熟成條件與外觀不同，故本文僅探討以禽畜糞為原料之肥料，且本論文之模型僅適用於慶豐興業之雞糞有機肥料之配方，若更改配方則須調整模型以符合其熟成情況與條件。
本論文透過比較各式模型與影像處理及權重更新方法的搭配與否，嘗試找出肥料熟成之影像辨識最適模型，實驗結果顯示本論文研提之CSP ResNeXt搭配Weight evolution深度學習技術，在有機肥料熟成影像辨識上達到87.63%的準確率與84.77的精確率成效頗佳，值得推廣到智慧農業中。

關鍵字

有機肥料 ； 有機肥熟成判定 ； 人工智慧 ； 深度學習 ； 影像處理

並列摘要

The government actively promotes organic agriculture, and scholars at home and abroad are also actively studying related issues, hoping to reduce the impact of agriculture on the environment through organic agriculture. However, farmers and manufacturers are deterred by the high input cost of organic agriculture. Therefore, this study cuts into this problem from the perspective of reducing the production cost of organic fertilizers. By providing a high-precision and low-entry-level fertilizer maturation identification technology, the inspection cost and waiting time required to determine the maturation of organic fertilizers are reduced.
In view of the limited time of this study and the different formulations, aging conditions and appearance of each organic fertilizer. This article only discusses the fertilizers using livestock manure as raw materials and the model of this study is only applicable to the formula of chicken manure organic fertilizer of Qingfeng Xingye corporation. If the formula is changed, the model must be adjusted to match its maturity and conditions
This study attempts to find the optimal model for image recognition of fertilizer ripening by comparing various models with image processing and weight updating methods. The experimental results show that the CSP ResNeXt proposed in this paper combined with Weight Evolution has achieved an accuracy of 87.63% and an accuracy of 84.77 % in the recognition of organic fertilizer mature images, which is worthy of being extended to smart agriculture.

並列關鍵字

organic fertilizer ； rapid ripening determination of organic fertilizer ； artificial intelligence ； deep learning ； image processing

參考文獻

[1] 陳冠佑(2014)。木黴菌酵成有機生物肥料。國立東華大學生命科學系碩士論文，花蓮縣。 取自https://hdl.handle.net/11296/hxqc6c

[2] 張簡佳欣(2020)。廚餘有機肥料對青梗白菜生長效益之研究。朝陽科技大學景觀及都市設計系碩士論文，台中市。 取自https://hdl.handle.net/11296/5pch9x

[3] 劉宇翔(2020)。接種根瘤菌與叢枝菌根菌搭配不同量稻殼生物炭及有機質肥料對土壤性質、紅豆生育及產量之影響。國立屏東科技大學農園生產系所碩士論文，屏東縣。 取自https://hdl.handle.net/11296/9c6w8q

[4] Ding, X., Zhang, X., Ma, N., Han, J., Ding, G., & Sun, J. (2021) Repvgg: Making vgg-style convnets great again. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 13733-13742)

[5] He, K., Zhang, X., Ren, S., & Sun, J. (2016) Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 770-778)