Development of a Cut Rose Longevity Prediction Model Using Thermography and Machine Learning

• Published in: Weon'ye gwahag gi'sulji Vol. 38; no. 5; pp. 675 - 685
• Main Authors: Choi, So Young, Lee, Ae Kyung
• Format: Journal Article
• Language: English
• Published: 한국원예학회HST 01.01.2020; 한국원예학회
• Subjects: 농학; cut flower; artificial neural network; non-destructive testing; thermal image; vase life
• ISSN: 1226-8763; 2465-8588; 2465-8588; 1226-8763
• DOI: 10.7235/HORT.20200061

To predict the longevity of cut roses (Rosa hybrida L.), we used thermal image analysis on ‘3D’,‘Kensington Garden’, and ‘Hera’ rose cultivars. At blooming stage, the temperatures of leaves andpetals were similar to or slightly lower than the air temperature. When the temperature of leaves andpetals increased by 2°C compared to the air temperature, no symptoms such as senescence werevisible in the leaves and petals. However, three days after the temperature increase, significantvisual senescence was observed and the temperature of leaves and petals decreased back to that ofthe air temperature. Based on this data, we identified three different stages of cut roses: (1) theblooming stage, (2) the last stage with no visual senescence, and (3) the stage with significant visualsenescence. To embody a longevity prediction model for cut roses, the temperature differencebetween the leaf of ‘Hera’ and the air were chosen for the practice data for the model. After themachine learning process, a model with 100% accuracy was obtained. According to the model,when the temperature of a cut rose leaf is lower than the surrounding air, it is undergoing itsblooming stage, while when it is higher it is undergoing the senescence stage. Using logisticregression with machine learning, a value of 1 indicates the senescence stage and a value of 0indicates the blooming stage. This study suggests that current smart farming techniques used for cutroses are first-generation level, which means there are limitations in environmental control whenusing a remote control system and partially automatic system. To upgrade this process andovercome these limitations, an optimal model to predict the longevity of a cut rose is needed. KCI Citation Count: 11