시계열 이미지와 생육 데이터를 활용한 딥러닝 기반 개체 과중의 비파괴적 추정

초록

Image-based deep learning techniques have been increasingly applied in agricultural fields, including disease detection, growth monitoring, and yield assessment. Strawberry (Fragaria × ananassa Duch.), characterized by a short harvest cycle and continuous production, requires frequent and plant-level monitoring for precise cultivation management. This study aims to examine the feasibility of non-destructively estimating plant-level fruit weight at harvest time by integrating time-series image data with growth information using a multimodal deep learning approach. Data were collected from the ‘Seolhyang’ cultivar grown in a glass greenhouse at Gyeongsang National University between September and December 2021. For each plant, time-series images captured at one-day intervals over the nine days preceding the harvest reference day were used, together with non-destructively measurable growth traits (plant height, number of leaves, leaf length, and leaf width) obtained on the reference day. Image data were preprocessed using HSVbased masking to remove background regions. A ConvLSTM network was employed to extract temporal features from the image sequences, while fully connected layers were used to process growth data, and both feature types were combined to estimate fruit weight. Model evaluation on the test dataset resulted in an RMSE of 5.40g and an R2 of 0.44, indicating the potential feasibility of non-destructive fruit weight estimation at harvest time. Additional analysis of RGB color channels revealed a moderate association between pixel intensity in the red channel and estimation error, suggesting that visually derived indicators related to fruit maturity may provide meaningful information for image-based fruit weight estimation. Overall, this study explores the feasibility of integrating time-series images and growth data for non-destructive fruit weight estimation and provides a foundational basis for plant-level productivity management in data-driven precision strawberry cultivation and smart farming environments.

키워드