This study is a deep learning-based growth prediction study aimed at year-round production through the creation of an optimal environment for Salicornia herbacea L. growth in a greenhouse environment. Salicornia herbacea L. was grown according to light intensity (PPFD) conditions and root zone environmental conditions using a vertical cultivation bed, growth data was collected according to condition changes, and a DNN model optimized for Salicornia herbacea L. growth prediction was designed. It was applied to the deep learning prediction model designed using the collected data to predict the length growth according to weeks and conditions.
1. According to the reference literature, studies on the standardized cultivation method for stable production of Salicornia herbacea L. and the environment affecting growth have been conducted, but research on how to grow in a greenhouse environment by controlling the growth environment of Salicornia herbacea L. has not been conducted. In order to grow Salicornia herbacea L., which has the efficacy of functional food and health food, in a greenhouse environment, a vertical cultivation bed that can increase production per area was used. We collected Salicornia herbacea L. growth data according to environmental conditions by applying controllable light intensity (PPFD) conditions and root zone environmental conditions to the growth environment.
2. According to the reference literature, there was a study that predicted values by applying statistics, machine learning, and deep learning techniques in crop growth prediction and yield prediction, and compared each technique to derive performance, and the DNN model performed the best. presented Therefore, in this paper, DNN, one of the deep learning models, was selected, and 18-18 was selected by applying the most optimal hidden layer and number of nodes by comparing the number of cases with the number of nodes at 2, 4, and 6 times the number of inputs. The results were derived by optimizing the model for the purpose of predicting Salicornia herbacea L. growth, such as applying Z_score Normalization, applying 4 for Mini Batch, applying 3600 for Batch Size, and selecting Adam of the optimizer. Through the evaluation index, the results showing an average error rate of 11.85% and 95% were confirmed, and it was determined that the DNN model could be used to predict Salicornia herbacea L. growth.
3. Collectable environmental conditions (Weeks (week), PPFD (Photosynthetic Photon Flux Density), Supply_EC (EC of supplied nutrient solution)) were used as independent variables to predict Salicornia herbacea L growth within a limited experimental environment. However, various environmental factors affect the growth of actual Salicornia herbacea L. growth, and it is judged that it will be possible to predict the growth of Salicornia herbacea L. similar to the actual environment if the data are collected by adding various environmental factors and the independent variables that have an effect through analysis are added.
4. The currently collected growth data of Salicornia herbacea L. is data collected through one experiment. In order to grow Salicornia herbacea L. grown in the existing open field in a greenhouse environment, research was conducted to grow it in a limited environment using a vertical cultivation bed. It was difficult to verify the average growth length of Salicornia herbacea L. according to average weeks.
5. In conclusion, as a result of the prediction of Salicornia herbacea L. growth by applying the DNN model to the growth data collected so far, the correlation between the independent variable and the dependent variable was 95 with an average error rate of 11.85% and an explanatory power of 95% coefficient of determination through the evaluation index for each experimental model. It was confirmed that it was %, and it was confirmed that it was possible to predict growth according to changes in environmental conditions by applying deep learning-based Salicornia herbacea L environment and growth data. When the reliability of Salicornia herbacea L. data is secured, it is judged that it can help create an optimal environment by applying a deep learning growth prediction model.