심층신경망을 이용한 딸기 생육상태 추정 = Strawberry growth status estimation using deep neural network

다국어 초록 (Multilingual Abstract)

For high-quality, high-yield cultivation, especially work in protected horticulture such as the control of an appropriate environment and physical actions are essential. Recently, automating cultivation technique for agricultural work using long cultivation experience and artificial intelligence has been proposed. And automating work should be based on crop statuses such as age and condition. In particular, since fruits and vegetables have both vegetative and reproductive growth stages, it is important to understand more precisely the status of crops. The purpose of this study is to estimate the growth status of strawberry plant after transplanting from RGB images using a deep neural network. The images were acquired using a GoPro camera (GoPro HERO 10, GoPro Inc., USA) of the strawberry plant ‘Seolhyang’, which was transplanted in a venro-type glass greenhouse. Videos were filmed for 8 weeks at the interval of 1 week from July 6th. The images were augmented using ImageDataGenerator in keras module and inputted to a deep neural network model of various structures. The models used for training are MLP, CNN, and vision transformer. The model showed a performance of more than 95% accuracy. In the future, for the deep neural network structure that showed the highest performance, We are going to proceed ablation test for each RGB channel and compare performance for the different numbers of layers. The application of the deep neural network in this study shows that it could be the basis for the design of work for each growth status in the construction of an automated system for high quality and high yield in strawberry cultivation.