Search

An ensemble model of convolutional neural network (CNN) and multilayer perceptron (MLP) models was developed to detect sweet pepper (Capsicum annuum var. annuum) fruits in images and predict their development stages. The plants were grown in four rows in a greenhouse, and images were collected from each row. Plant environment and growth data were collected every minute and month, respectively. The fruit development stage was classified into immature, breaking, and mature stages with a CNN using images. The immature stage was internally divided into four stages with an MLP, so a total of six stages were classified using the CNN–MLP ensemble model. The plant growth and environmental data and the information from the CNN output were used for the MLP input. The average accuracy of the six stages was F1 score = 0.77 and IoU = 0.86. The ensemble model showed acceptable performance in predicting fruit development stages. The CNN-only model could classify the mature and breaking stages well, but the immature stages were not distinguished, while the MLP-only model could hardly classify the fruit stage except the immature stages. The most influential factors in classification were the data obtained from CNN and the plant growth and environment data. The ensemble models could help in appropriate labour allocation and strategic management by detecting individual fruits in images and predicting precise fruit development stages.

Hydroponic solutions used in greenhouses or plant factories are usually evaluated based on their electrical conductivity (EC) and pH. However, EC and pH cannot provide sufficient information about ion imbalances in hydroponic solutions, and this may result in wastage of nutrients or poor yields. This paper reports on the development of an on-site ion monitoring system based on ion-selective electrodes (ISEs) that can automatically calibrate sensors and measure the concentrations of individual ions (NO3−, K+, and Ca2+) in hydroponic solutions. This enables farmers to effectively manage nutrients in reused solutions by rapidly identifying any imbalances that appear in the nutrient ratios. The measurement performance of the developed system was evaluated using hydroponic solutions prepared for growing paprika crops in greenhouses. An application test was conducted to investigate the feasibility of using the developed on-site ion monitoring system for the automated measurement of three macronutrients (NO3−, K+, and Ca2+) in a real greenhouse. The results showed that the developed system was able to measure NO3− concentrations, showing an almost 1:1 relationship with the results of a standard instrument, i.e., ion chromatography (slope of 0.99 and R2 of 0.99). Although the developed system overestimated and underestimated the K+ and Ca2+ concentrations with slopes of 1.17 and 0.75, respectively, the high coefficients of determination of 0.99 and 0.97 made it possible to use calibration factors to compensate for differences in estimation. In fact, relatively low RMSEs of

Deep learning is the state-of-the-art application of machine learning in many fields, and this technology has also been applied in agriculture. A large quantity of data needs to be provided to the deep learning models in the training procedure; however, sufficient data may not be provided when considering agriculture applications. Transfer learning, which is a learning strategy for rapid and easy adaptation of a pre-trained model, can be a solution for limited agricultural data. Therefore, the objective of this study is to verify the adaptability of a pre-trained model that predicts the environmental variables of a greenhouse by retraining the model with data from a new cultivation condition, using the transfer learning technique. As a result, the transfer learning methodology was applied to five common deep learning models. Twenty-seven greenhouses (14 sweet peppers and 13 tomato cultivations) in various regions of South Korea provided the experimental dataset to this research. The analyzed environmental variables are the internal temperature, relative humidity, radiation, CO2 concentration, and external temperature. Before the transfer learning procedure is conducted, some layers from pre-trained models were replaced with new layers. The model was, thereafter, re-trained with a new test dataset. The best model in the training procedure was BiLSTM, resulting in an average R2 of 0.69. The models could predict the tendencies of the environmental changes, indicating that they were adequately trained. The most accurate deep-learning model considering the transfer dataset was the transferred BiLSTM, with an average R2 of 0.78 and 0.81 for sweet pepper and tomato datasets, respectively. The accuracies of most transferred models are higher than those of the corresponding deep-learning models. As a result, transfer learning can be used to adapt previously trained deep-learning models, enabling them to predict the microclimates of a greenhouse with scarce data. Furthermore, advanced transfer learning strategies would increase the performance of the transferred models analyzed in this study.

Diffuse fraction, which can be increased by using diffuse films, has been considered to influence light interception and photosynthesis of crops in greenhouses. However, quantifying the influence of diffuse films is challenging owing to the complicated optical interactions between climatic factors inside and outside greenhouses. Thus, versatile methods for evaluating the effect of diffuse films are required. The objective of this study was to evaluate the effect of diffuse films on the improvement of the light profile and photosynthesis of tomatoes in greenhouses according to film diffuseness and regional solar radiation using ray-tracing simulation. The structural and optical properties of the greenhouse components were applied in a 3D-framework combined with a ray-tracing module. The light transmission patterns of diffuse films and solar radiation properties were incorporated. The reliability of the simulation was confirmed by comparing measured and estimated irradiances inside greenhouses covered with films having different haze factors. For scenarios, the diffuse film efficiency was assessed under typically different solar radiations, a low irradiance, high diffuse radiation fraction (LIHD) and a high irradiance, low diffuse radiation fraction (HILD). The light interception was estimated through the simulation and used to calculate the photosynthesis using the Farquhar-von Caemmerer-Berry model. The simulation was found to be reliable with R2 of 0.95 and 0.94 for the two greenhouses covered with different diffuse films. The light distribution on the tomato plants were less affected by film diffuseness under LIHD than HILD. With increasing film diffuseness, carbon uptake and light use efficiency increased by 5.30% and 4.58% under HILD, but did not change under LIHD. The light distribution and photosynthesis in diffuse film-covered greenhouses under different light environments could be reasonably estimated by the simulation. Thus, this method can be used to evaluate the applicability of diffuse films to various regions with diverse meteorological characteristics.

Irrigation strategy has a direct influence on the productivity and the production cost of soilless cultures of paprika, and thus more efficient irrigation methods are required. The objective of this study was to compare water use efficiency and crop productivity between a conventional irrigation method and a modified irrigation method using a precise irrigation control system. Accumulated radiation was used as an irrigation index for the conventional method as a control, whereas the compensated value of accumulated radiation as well as the substrate moisture content and electrical conductivity was used as irrigation indices for the modified one. Plant growth and water use efficiency were compared between both irrigation systems. The experiment was performed in a commercial farm cultivating paprika in Hwasung, Korea. For the modified method, the substrate moisture content was well-controlled in the range of 69–85% and showed a narrow fluctuation compared to the control. Additionally, the substrate electrical conductivity was maintained within 2.4–4.6 dS m −1 during the growth period, whereas the substrate electrical conductivity increased up to 7.9 dS m −1 for the control. Water consumption and water-use efficiency of the shoots and fruits were 6.6% and 3.7% higher, respectively, with the modified method compared to the control. Overall plant growth was better with the modified method. Based on these results, it could be concluded that the modified irrigation method improved the productivity of paprika in soilless cultures.

Plant structure is a significant factor for influencing the light interception and photosynthesis of plants. The light interception on the plant surface can be analyzed by three-dimensional (3D) plant model and optical simulation, but its accuracy is directly affected by the structural accuracy of the 3D model. This study aims to analyze and compare the effect of the accuracy of 3D structural models on light interception and photosynthesis. 3D-scanned plant models with different structural accuracies were constructed, and the light interception and photosynthetic rate were analyzed at single leaf and whole-plant scales. When using a low accuracy model that lacked the fine structural details of the plant, it was overestimated in light interception and photosynthetic rate compared to the 3D-scanned model that has high structural accuracy. At the single leaf scale, the light interception was higher in the low-accuracy model than that in the 3D-scanned model due to self-shadings from higher curvature in the leaf surface. At the whole-plant scale, the light interception and the subsequent photosynthetic rate in the low-accuracy model were 18% and 45 to 58% higher than those in the 3D-scanned model at light intensities of 700–2000 μmol m−2 s−1 at the upper canopy. The 3D-scanned plant model could accurately estimate the light interception and photosynthetic rate of the plants through optical simulation. The presented methodology can contribute to accurate analyses of plant light environment, plant physiological response, and plant growth modelling.

Water management directly affects the productivity of paprika plants and is currently determined based on accumulated radiation levels. However, the amount of water used by the plants, which can be determined by their transpiration rates, does not always increase proportionally to the accumulated radiation levels, depending on the region and climate as well as crop growth stages and development. This effect is observed because the transpiration rate is also related to light intensity, which varies with the time of day and season. To develop a more efficient irrigation strategy, both factors should be analyzed based on the relationship between light intensity and transpiration rate in the short-term. In this study, a sigmoidal relationship between light intensity and transpiration rate at an interval of 10 min was observed using a consecutive transpiration monitoring system. From this relationship, a compensated equation that can calibrate the light intensity was developed. When a modified irrigation was applied using this compensated equation, less water was used compared to a conventional irrigation that supplies water proportional to accumulated radiation, especially in summer. Moreover, there were no significant differences in the transpiration rates and plant growth between plants watered with either the conventional or modified with compensated equation irrigation method. From these results, it was concluded that water was used more efficiently with the modified irrigation method without affecting plant growth. In a region with a high solar radiation in summer, such as Korea, using our equation to calculate for light intensity can prevent water waste, resulting in energy-saving and a reduction of environmental pollution in open-loop soilless cultures.

For analysis of greenhouse environments using big data, measuring data should be continuously collected without data loss caused by sensing and networking problems. Recently, deep learning approach has been widely used for precision agriculture. However, in order to use deep learning methods, the enormous amount of reliable data is necessary. The objective of this study is to compare the interpolation accuracy of greenhouse environment data using multilayer perceptron (MLP) with existing statistical and machine learning methods. Linear and spline interpolations were selected as statistical methods, and linear models, MLP and random forest (RF) were selected as machine learning methods. The raw data used for interpolation were greenhouse environment data collected from October 2, 2016 to May 31, 2018 where Irwin mango (Mangifera indica L. cv. Irwin) trees were cultivated. As a result, the linear interpolation method showed the highest R2 (average 0.96) in short-term data loss conditions, but the MLP showed R2 = 0.95. However, in long-term data loss conditions, the accuracies of the linear, spline, and regression interpolations decreased, but the accuracies of the MLP and RF remained stable. However, MLP showed better accuracies than RF. Therefore, the MLP was better suited to interpolating greenhouse environment data because short- and long-term data loss actually occurred simultaneously when collecting greenhouse environment data. The trained MLP showed the high accuracy in both short- and long-term data interpolations, indicating that MLP can also be complementally used with existing methods. The trained MLP accurately estimated the missing data in the greenhouse and will contribute to the analysis of big data collected from greenhouses.

An environmental prediction model was developed for optimal ventilation in a mushroom house utilising a sensible heat balance and a three-dimensional (3-D) computational fluid dynamics (CFD) model. The respiration of the mushrooms and the use of a low-capacity cooler were considered. A mushroom-house-specific ventilation equation was developed to calculate the ventilation rate for a given environmental condition. Calculated ventilation rates were compared with the experimentally measured data for the indoor temperature set to the optimum for growing mushrooms (16.2 °C) with varying outdoor temperature. There was good agreement between the measured and predicted rates (0.2–5.1% error). Calculated ventilation rates (from the sensible heat balance) were used as an input parameter for 3-D CFD model, eliminating the need for experimental measurement of ventilation rate. 3-D CFD simulations were conducted using the same environmental condition to establish the local heat distribution in a mushroom house. The simulation results for temperature were compared with the experimental data at several different locations in a mushroom house and showed negligible errors. The CFD model was also used to improve heat distribution of a mushroom house. It was predicted that enhanced cooling and more uniform temperature distribution could be achieved just by changing the direction of airflow from air inlet ducts and/or installing small fans onto them, but not by changing the directions of airflow from a cooler. This would be a more economical than replacing a cooler or redesigning the entire structure. The model could be used to predict the environmental conditions over different locations in a mushroom house without the need for experimentally determining the ventilation rate.

Recirculating subirrigation systems are frequently exposed to the risk of plant pathogens transmission, which may deteriorate the growth and quality of the plants. The transmission of Phytophthora nicotianae was examined using Kalanchoe blossfeldiana cv. New Alter in two recirculating subirrigation systems, a nutrient-flow wick culture (NFW) system and an ebb and flow (EBB) system. When the nutrient solution was infested, the pathogen was recovered from roots in both subirrigation systems. However, foliar blights and browning of roots appeared 4 and 7 weeks, respectively, after inoculation in the EBB system. Only a little discoloration appeared in the NFW system. The fresh and dry weights were lower in the EBB system than in the NFW system. When growing medium was inoculated, the pathogen was unable to be isolated from the plants in the NFW system. However, disease symptoms appeared in the EBB system 4 weeks after inoculation, and the pathogen was observed in the basal leaves and roots. Similar to the infested nutrient solution, the plant growth in the EBB system was inhibited. These results suggested that when the nutrient solution was infested, pathogen transmission could occur in plants in both systems, although differences existed with regard to disease symptoms and the time it took for symptoms to appear. However, we observed that when growing medium was inoculated the pathogen was not transmitted to adjacent plants in the NFW system using wick.

The distribution of humidity in a greenhouse was studied using three-dimensional (3-D) computational fluid dynamics (CFD). The calculations were validated using experimental data from a single-span greenhouse without plants. Two types of humidity distribution were considered: humidifying using a fog cooling system, and dehumidifying using refrigerative dehumidifiers in addition to a fog cooling system. The simulation errors of RH were 0.1–18.4% with a fog cooling system and 1.1–13.1% with a fog cooling system and refrigerative dehumidifiers at each observation point. Contour maps were obtained from the 3-D CFD simulations to locate any non-uniformity in humidity distribution. The use of refrigerative dehumidifiers reduced the overall difference of humidity between the middle and bottom zones of a greenhouse, but the local distribution of humidity was uneven, especially close to the dehumidifiers. This study suggests that the developed 3-D CFD model can be a useful tool in designing and evaluating greenhouses with various configurations.

Non-destructive and mathematical approaches of modeling can be very convenient and useful for plant growth estimation. To predict individual leaf area, fresh weight, and dry weight of a cucumber ( Cucumis sativus L.), models were developed using leaf length, leaf width, SPAD value, and different combinations of these variables. Eight regression equations, commonly used for developing growth models, were compared for accuracy and adaptability. The three nonlinear models developed were as follows: individual leaf area (LA) = −210.61 + 13.358 W + 0.5356 LW ( R 2 = 0.980 *** ), fresh weight (FW) = −2.72 + 0.0135 LW + 0.00022 LWS ( R 2 = 0.956 *** ), and dry weight (DW) = 0.25 − 0.00102 LS + 0.000077 LWS ( R 2 = 0.956 *** ), where L is the leaf length, W the leaf width, S the SPAD value, and LWS = L × W × S . For validation of the model, estimated values for individual leaf area, fresh weight, and dry weight showed strong agreement with the measured values, respectively. Leaf dry weight, especially, was estimated with a higher degree of accuracy through the use of a SPAD value, as well as leaf length and width. Therefore, it is concluded that models presented herein may be useful for the estimation of the individual leaf area, fresh weight, and dry weight of a cucumber with a high degree of accuracy.

To compliment the current subirrigation systems used for production of potted plants, a nutrient-flow wick culture (NFW) system was developed and compared with other subirrigation systems, such as an ebb and flow culture (EBB) system and a nutrient-stagnant wick culture (NSW) system in relation to their system characteristics and plant growth. Kalanchoe ( Kalanchoe blossfeldiana cv. New Alter) was cultivated in a 6 cm pot for 10 weeks in each subirrigation system. The water-absorption pattern of the medium, water content of the medium, water loss, algal growth, salt-buildup and plant growth under various culture systems were observed. The water contents of medium under the NFW and EBB systems showed fluctuations from 30 to 40% and from 50 to 60% (by volume), respectively, whereas the water content under the NSW system gradually increased to over 40% without fluctuation. Relative to other systems, the water loss in the NFW system was 50–70% due to the reduction in the evaporation from the surfaces of the trough and medium. Algae appeared in the NSW system because the nutrient solution was always stagnant in the trough, while it was not observed under the NFW system. The dissolved oxygen in the nutrient solution was the highest during the irrigation period and the salinity in the medium was the lowest in the NFW system. With regard to system characteristics, the NFW system was simple, water-saving and efficient. In addition, the growth of kalanchoes in the NFW system was similar to those in the NSW and EBB systems at an irrigation frequency of five times a day.

This paper introduces a new concept of automation for bio-production with tele-operative system. The proposed system showed practical and feasible way of automation for the volatile bio-production process. Based on the proposition, recognition of the job environment with object identification was performed using computer vision system. A man-machine interactive hybrid decision-making, which utilized a concept of tele-operation was proposed to overcome limitations of the capability of computer in image processing and feature extraction from the complex environment image. Identifying watermelons from the outdoor scene of the cultivation field was selected to realize the proposed concept. Identifying watermelon from the camera image of the outdoor cultivation field is very difficult because of the ambiguity among stems, leaves, shades, and especially fruits covered partly by leaves or stems. The analog signal of the outdoor image was captured and transmitted wireless to the host computer by R. F module. The localized window was formed from the outdoor image by pointing to the touch screen. And then a sequence of algorithms to identify the location and size of the watermelon was performed with the local window image. The effect of the light reflectance of fruits, stems, ground, and leaves were also investigated.