Persimmon maturity index for chronological analysis using AI object detection algorithm

Recent progress in IoT technology has made it easier to obtain time-series environmental data. Attempts to save labor and optimize plant growth by monitoring meteorological and physical indicators are being made in many parts of the world. For persimmons, analysis of chronological indicators for predicting proper harvest time and fruit yield helps to improve the efficiency of farm management. However, the traditional technique for observation of the fruits’ mutuality process, which is linked to harvest time and fruit yield, causes difficulties in terms of time and skill for proper data acquisition. In this study, a new maturity index was developed using the object detection technique with ‘You Only Look Once’ (YOLO) algorithm which applies the deep learning algorithm to track fruit maturation of persimmon, ‘Tone Wase’. The developed index was tested on greenhouse persimmon fruits in Nara, Japan. Digital images of the fruit maturity process were monitored during the 2020 season using fixed point cameras installed in the greenhouses. Images were taken four times a day and sent to the server through the LTE network. These images were thereafter processed and analyzed using YOLO for quantification of the time-series maturity process to create the index. Results of the analysis showed that the number of detected fruits on one image increased during the fruit enlargement and maturation stages and decreased during the period of the physiological fruit drop, fruit thinning and harvesting. The developed maturity index was able to reflect the growth of fruits in the greenhouse. The index made it possible to quantify a relationship between amount of irrigation application, fruit yield and harvest time and indicated a less yield and earlier harvest time in a house with a less amount of irrigation application. The current study continues to collect more data to increase the accuracy of the persimmon-fruit maturity index.