과일의 신선도를 나타내는 색 변화 기반 신선도 지시계 개발 및 지시계 해석을 돕는 모바일 소프트웨어 개발 = Development of a color change-based freshness indicator for fruit and mobile software for indicator interpretation

다국어 초록 (Multilingual Abstract)

Ⅰ. Assessing the freshness of bananas and kiwis with a potassium permanganate-based ethylene indicator using deep learning-based mobile software

This study investigated the applicability of a potassium permanganate (KMnO4)-containing ethylene indicator in sensing the freshness of bananas and kiwis and explored the feasibility of the use of deep learning to interpret the indicator outcomes. This research also developed a mobile software that adapted the trained deep learning model. The concentrations of KMnO4 impregnated in the indicator base material (filter paper) were 0.1, 0.5, and 1.0% (w/v). Bananas or kiwis were placed in a polypropylene pouch attached to the ethylene indicator and stored at 25 ﾟC for 10 days. The color of the indicator, total aerobic bacterial count, yeast and mold count, ethylene gas concentration in the pouch, total soluble solid content, titratable acidity, firmness, and weight loss of the fruits were analyzed during storage. The indicator image dataset demonstrating freshness and spoilage was labeled and trained using ResNet50. High correlations were observed between the ethylene gas concentration and color change in the “0.5% KMnO4 indicator.” The color of the indicator changed from purple to brown during the spoilage of bananas and kiwis from day 7 to 10. The spoilage time points determined by microbial counts were validated by principal component analysis. The ResNet50 model demonstrated 100% accuracy in predicting the freshness of both fruits. The mobile software developed using the trained model quickly identified the indicator images on fruit packages. Overall, the ethylene indicator was found suitable for indicating the freshness of bananas and kiwis, and the mobile software rapidly and accurately assessed the fruit freshness.


Ⅱ. Establishment of a freshness identification system for climacteric fruits using a dual system ethylene-carbon dioxide indicator with deep learning-based mobile software in the food supply chain


This research developed a dual system freshness indicator, consisting of an ammonium molybdate (AM)-based ethylene indicator and a methyl red-based carbon dioxide indicator, which changed its color reacting with ethylene and carbon dioxide gases. Banana or kiwi was placed in a PP attached with the dual system freshness indicator, stored at 25 ﾟC for 10 days or 12 days, respectively, and analyzed for changes in the color of dual system freshness indicator, as well as the total aerobic bacterial counts and yeast and mold counts in the fruits and ethylene gas concentration, and carbon dioxide gas concentration inside the pouch. Indicator images representing freshness and spoilage were labelled as a dataset and trained using a deep learning model (ResNet50). The trained model was installed in mobile software that identifies the freshness of the fruits. Among the concentrations of AM in the PdSO4-AM solution (1, 2, and 3%), the 1% solution resulted in the highest distinctive color change responding to different concentrations of ethylene gases. The PP pouch was more suitable for the carbon dioxide indicator application with appropriate carbon dioxide permeation than nylon PE pouch. The color of the ethylene indicator prepared with the PdSO4-AM solution containing AM at 1%. For bananas, spoilage occurred on the 9th day of storage based on the total bacterial count criterion. At this point, the ethylene indicator changed from yellow to blue, while the carbon dioxide indicator changed from red to yellow. while the color of the carbon dioxide indicator shifted from red to yellow and then to orange. In the case of kiwis, spoilage was observed on the 12th day of storage based on the yeast and mold count threshold. At this stage, the ethylene indicator exhibited a color change, while the carbon dioxide indicator showed no color change. The ResNet50 model demonstrated 100% accuracy in predicting freshness of both fruits. Moreover, the mobile software integrated with the trained ResNet50 model was rapidly judging the freshness of fruit. The results demonstrated that the dual system freshness indicator effectively assessed the freshness of bananas and kiwis and the mobile software incorporating the trained deep learning model enabled rapid evaluation of freshness of the fruits.

목차 (Table of Contents)

• Ⅰ. 딥러닝이 통합된 모바일 소프트웨어를 이용한 potassium
• permanganate 기반 에틸렌 지시계를 통한 바나나와 키위의 신선도
• 판독
• 1. 서론 1
• 2. 재료 및 방법 4
• 2.1 바나나 및 키위 준비· 4
• 2.2 에틸렌 지시계 제작 4
• 2.3 Potassium permanganate 농도 결정 7
• 2.4 과일의 부패 시점 결정 7
• 2.5 저장 중 과일 품질 결정 8
• 2.6 주성분 분석을 이용한 부패 시점에서의 과일 품질 특성의 다변량
• 분석· 9
• 2.7 에틸렌 지시계 신선/부패 이미지를 이용한 ResNet50 모델 학습과
• 모델 작동 평가 10
• 2.8 ResNet50 모델 학습을 적용한 모바일 어플리케이션 개발 11
• 2.9 통계 분석 11
• 3. 결과 및 고찰· 12
• 3.1 Potassium permanganate 농도 결정· 12
• 3.2 과일의 부패 시점 결정· 18
• 3.3 주성분 분석을 이용한 부패 시점에서의 과일 품질 특성의 다변량
• 분석 21
• 3.4 에틸렌 지시계 신선/부패 이미지를 이용한 ResNet50 모델 학습과
• 모델 평가 25
• 3.5 ResNet50 모델 학습을 적용한 모바일 어플리케이션 개발 28
• 4. 결론 30
• Ⅱ. 딥러닝이 통합된 모바일 소프트웨어를 이용한 dual system
• 에틸렌-이산화탄소 지시계를 통한 바나나와 키위의 신선도 판독
• 1. 서론 32
• 2. 재료 및 방법· 35
• 2.1 에틸렌 지시계 개발 35
• 2.1.1 PdSO4-AM solution 제작 35
• 2.1.2 PdSO4-AM solution에서 AM 농도 결정· 35
• 2.1.3 에틸렌 지시계 제작 36
• 2.1.4 에틸렌 지시계의 특성 분석 37
• 2.2 이산화탄소 지시계 제작 38
• 2.3 지시계 적용 포장 재료 결정 39
• 2.4 Dual system 신선도 지시계 적용 과일 포장재 준비 40
• 2.5 과일의 품질 평가 44
• 2.6 지시계 이미지를 이용한 ResNet50 모델 학습 45
• 2.7 ResNet50 모델 학습을 통한 소프트웨어 개발 46
• 2.8 통계 분석 46
• 3. 결과 및 고찰· 47
• 3.1 에틸렌 지시계 개발 47
• 3.1.1 PdSO4-AM solution에서 AM 농도 결정 47
• 3.1.2 에틸렌 지시계 제작 50
• 3.1.3 에틸렌 지시계의 특성 분석 52
• 3.2 이산화탄소 지시계 제작 57
• 3.3 지시계 적용 포장 재료 결정 60
• 3.4 Dual system 신선도 지시계의 적용성 확인· 65
• 3.5 지시계 이미지를 이용한 ResNet50 모델 학습 69
• 3.6 ResNet50 모델 학습을 통한 모바일 소프트웨어 개발 72
• 4. 결론 74
• 참고문헌 76
• Abstracts 89