A Monitoring, Evaluation, and Prediction System for Slight Water Stress in Citrus Seedlings Based on an Improved Multilayer Perceptron Model
Abstract
:1. Introduction
- (1)
- We conducted experiments on different gradients of water stress in citrus seedlings. Through these experiments, we developed a method to determine water stress in citrus seedlings and collected electronic data on water stress. We used electromagnetic sensors, LoRa technology, and a host computer to establish a sensor network, enabling non-destructive and remote data collection. Additionally, we observed changes in both aboveground and underground characteristics of seedlings after stress and formulated a set of criteria for assessment. Based on the VWC dataset obtained from experiments, we provide the delineation intervals for water stress in citrus seedlings. These intervals serve as reference points for other research endeavors.
- (2)
- We constructed a classification model for water stress. Based on the experimental dataset, we augmented the dataset through techniques such as data augmentation and class balancing, bootstrap sampling, and feature engineering. This augmentation aimed to more accurately reflect the soil moisture conditions during actual seedling cultivation processes, thereby enhancing the model’s generalization ability. After analyzing various classification models, we chose the MLP model, which exhibits similarities in computational processes to the accumulation process of water stress. We optimized the model’s performance by considering the dataset’s features and the specific performance of the loss function, achieving satisfactory results.
- (3)
- We conducted experiments on slight water stress, and the actual status of the seedlings matched the predictions of the model. Additionally, the model’s results exhibited strong confidence, demonstrating the model’s excellent performance.
2. Materials and Methods
2.1. Total System Design
2.2. Data Acquisition of VWC
2.2.1. Root Experiment under Water Stress
2.2.2. Wireless Sensor Network Based on LoRa
2.2.3. VWC Experiment of Water Stress under Different Gradients
- (1)
- Collecting VWC datasets under different gradients of water stress: The experiment gathered VWC data from citrus seedlings subjected to varying degrees of water stress. These data provide insights into changes under different water stress conditions.
- (2)
- Determining the boundaries of drought, health, and waterlogging: By analyzing the collected data, this experiment identified the boundaries between drought and health or waterlogging and health. When evaluating the ideal water stress edge, we tried to find two gradual VWC lines as reference points. Any horizontal lines above or below these two lines were considered to indicate the presence of water stress. By establishing this framework, the experiment identified and quantified the occurrence of water stress based on distinct variations in VWC and established a reliable method for identifying the boundaries of water stress based on VWC.
- (3)
- Obtaining relevant information about slight water stress: This experiment captured and analyzed data related to slight water stress in citrus seedlings. By studying the changes in VWC and other relevant parameters, this experiment produced insights into the characteristics and impacts of slight water stress on plants.
- (4)
- Validating the necessity and benefits of electronic monitoring systems for slight water stress.
- (1)
- Preparation: Before entering the seedling rearing phase, the seedlings are carefully removed from the original seedling rearing bags to ensure the root system remains healthy.
- (2)
- To maintain the stability of the water stress gradient, small amounts of irrigation are selectively carried out twice a day, at 10 a.m. and 4 p.m. The water stress gradient is influenced by both time and VWC, and it is essential to consider both factors to improve the accuracy of the gradient assessment. To ensure reliable results, we maintain the experimental data within a predefined range of [−5%, 5%], thus avoiding multiple gradients within a single experiment.
- (3)
- Upon completion of the seedling rearing phase, the sensor probes are carefully removed from the substrate. The substrate is then poured out, and the roots of the seedlings are gently washed using a small water flow. This process helps in assessing the state of the plants and observing any visible changes or effects resulting from the water stress experiment.
- (4)
- After the experiment, the seedlings need to be re-implanted in the new seedling bag. Before the next experiment, the seedlings should be restored to a healthy state. By following these steps, the experiment ensures proper care and monitoring of the citrus seedlings during the experiment process, allowing for an accurate assessment of the effects of water stress on the plants.
2.2.4. VWC Dataset Production
- (1)
- Data Augmentation: The dataset underwent minor vertical shifting to enhance the role of original data during the training process, enabling the model to better learn data features and improving its robustness. Introducing data augmentation helps reduce the model’s excessive reliance on training data, thereby enhancing its performance on unseen data. Seven-day datasets were obtained, and overall numerical values were subtracted from 0.1 to 0.9, with the original data labels retained. Due to the relatively smooth data characteristics and small variations in this subset, it was more easily classified, thereby the quantity of this subset is relatively small.
- (2)
- Bootstrap Sampling: This is a commonly used resampling method in statistics, involving the generation of numerous sub-samples (bootstrap samples) with replacement from the original dataset. Inference is then made using confidence intervals. However, since all datasets resulting from this method will have identical labels (as they are derived from the same source), we only utilize its conceptual framework for sampling. Within each individual-day dataset for different states, a 7-day dataset is generated by randomly shuffling the data, with the corresponding state serving as the label. The generated datasets exhibit more pronounced and steeper changes compared to the original dataset, thereby increasing the gradient variation of numerical values within the same dataset. This emulates the fluctuating moisture conditions in real-world seedling environments.
- (3)
- Feature Engineering: In this study, feature construction was primarily utilized, involving the creation of new features through combinations, transformations, or derivations to enhance the expressive power and predictive performance of machine learning models. Feature construction leverages domain expertise or prior information to construct specific features relevant to the problem at hand, enabling the model to better capture the underlying structure and patterns of the data, thereby improving accuracy and generalization capability. Seven daily datasets were randomly selected and merged to generate a 7-day dataset. Based on experimental findings, we supposed that the health state could be considered healthy if it persisted for at least 5 days, whereas a state was deemed unhealthy if it persisted for at least 3 days. It is important to note that the unhealthy state labels in the mixed dataset do not necessarily represent actual unhealthy states (these unhealthy labels signify suboptimal seedling quality). For instance, a curve exhibiting 4 days of health and 3 days of an unhealthy state does not reflect an ideal seedling growth process and thus is labeled as unhealthy. Through this approach, more datasets representing mild water stress were created to enhance the differentiation between mild stress and healthy data.
2.3. Selection and Improvement of the MLP Model
2.3.1. Model Selection
- -
- The stress level is represented by Di, a non-negative number.
- -
- The stress correction coefficient is denoted as Ti, also a non-negative number, and Si represents the stress correction constant.
- -
- Di is positively correlated with the deviation from the ideal environment.
- -
- Ti is negatively correlated with the deviation from the average value of the 7-day dataset. A greater deviation indicates a shorter duration of VWCi’s effect, thus reducing its impact on Li. Consequently, when the deviation is significant, Si takes a negative value, causing Li to approach 0; conversely, Si is positive, increasing Li.
- -
- Overall, L is cumulatively formed by Li.
2.3.2. MLP Model Optimization
3. Results
3.1. Root Experiments
- (1)
- Drought group: In this group, there are minimal or no new shoots or lateral roots at the top. The main root system appears dry and pale, and the new shoots at the end of the lateral roots are slender.
- (2)
- Health group: In contrast to the other groups, the health group exhibits significant growth in the lateral roots at the top. There is an abundance of flexible and robust new shoots. The main roots appear light yellow and white, indicating healthy growth. Additionally, there is noteworthy growth in new shoots at the end of the lateral roots.
- (3)
- Waterlogging group: The growth of lateral roots at the top of this group is slow, with short and scarce new shoots. The main root system is black and sticky, indicating unfavorable conditions. The growth of new shoots at the end of the lateral roots is also sluggish.
- (1)
- Drought group: After experiencing drought damage, the entire root system becomes dry and stiff. The color of the roots is predominantly white with some light yellow areas. The root crown appears slender, and there is limited extension and enlargement in the elongation zone and meristem zone.
- (2)
- Health group: The entire root system of the healthy group is predominantly white. The mature area appears light yellow, and the roots have a flexible texture. The root crown is round and full, indicating overall health. Furthermore, the elongation zone and meristem zone exhibit clear signs of extension and enlargement, indicating active growth.
- (3)
- Waterlogging group: The root system of the waterlogging-damaged group exhibits weakness and has a mucous texture. The color of the roots is dark, including black areas. The root crown appears slender and fragile. The extension and enlargement of the elongation zone and meristem zone are not significant. Additionally, the whole root system emits an unpleasant odor.
3.2. VWC Experiment
3.2.1. VWC and Root Changes
- (1)
- The experiments identified both boundaries for drought and waterlogging. The drought edge falls between lines 20% and 30%, and the waterlogging edge lies between lines 50% and 55%. Line 45% is closest to the ideal seedling environment, and the best seedling conditions exist between lines 45% and 50%.
- (2)
- The experiments revealed that even slight water stress can have a significant impact on the normal growth of seedlings. Seedlings after the 20% drought test were subjected to the 50% experiment after a 5-day recovery period. From the overall results of the experiment, our takeaway is that the 50% VWC should have been closer to the ideal environment, but the actual seedlings still showed a significant cessation of growth, as well as a small amount of leaf abscission. This emphasizes the importance of studying slight water stress to improve seedling quality.
3.2.2. Leaf Change in Water Stress under Different Gradients
3.3. VWC Dataset Production
3.4. MLP Model Optimization
3.4.1. Selection and Analysis of ReLU
3.4.2. Selection and Analysis of Sigmoid
3.4.3. Selection and Analysis of BatchNorm1d
4. Discussion
5. Conclusions
- Healthy seedlings typically exhibited a predominantly white root system with a light yellow mature area. The roots possessed flexibility, and the root crown appeared round and full. Additionally, significant elongation and enlargement were observed in the elongation and meristem areas. In contrast, drought damage resulted in a dry and stiff root system characterized by white and light yellow coloration. The root crown became slender, with minimal extension and growth in the elongation and meristem areas. Moreover, drought-induced abscisic acid production led to the degradation of leaf chlorophyll. Waterlogging damage manifested as a flagging and mucous root system, exhibiting a black and dark color. The root crown became fragile and slender, with limited extension and enlargement in the elongation and meristem areas. The entire root system emitted an unpleasant odor.
- The optimal VWC for seedlings was [45%, 50%], while the boundary between drought and health was [20%, 25%], and the boundary between waterlogging damage and health was [50%, 55%].
Author Contributions
Funding
Data Availability Statement
Conflicts of Interest
References
- Chand, L.; Sharma, S.; Dalal, R.; Poonia, A.K. In vitro shoot tip grafting in citrus species-A review. Agric. Rev. 2013, 34, 279–287. [Google Scholar] [CrossRef]
- Rodríguez-Gamir, J.; Ancillo, G.; González-Mas, M.C.; Primo-Millo, E.; Iglesias, D.J.; Forner-Giner, M.A. Root signalling and modulation of stomatal closure in flooded citrus seedlings. Plant Physiol. Bioch. 2011, 49, 636–645. [Google Scholar] [CrossRef] [PubMed]
- Syvertsen, J.P.; Garcia-Sanchez, F. Multiple abiotic stresses occurring with salinity stress in citrus. Environ. Exp. Bot. 2014, 103, 128–137. [Google Scholar] [CrossRef]
- Wu, Q.; Zou, Y.; Huang, Y. The arbuscular mycorrhizal fungus Diversispora spurca ameliorates effects of waterlogging on growth, root system architecture and antioxidant enzyme activities of citrus seedlings. Fungal Ecol. 2013, 6, 37–43. [Google Scholar] [CrossRef]
- Liu, C.Y.; Zheng, M.M.; Xiong, S.J.; Zeng, Q.H.; Liu, J.; Liu, G.D. Effects of flooding stress on morphological and physiological characteristics of ‘Newhall navel orange’/trifoliate orange seedlings. J. Fruit Trees 2019, 36, 1494–1504. [Google Scholar]
- Rodríguez-Gamir, J.; Primo-Millo, E.; Forner, J.B.; Forner-Giner, M.A. Citrus rootstock responses to water stress. Sci. Hortic. 2010, 126, 95–102. [Google Scholar] [CrossRef]
- Kosola, K.R.; Eissenstat, D.M. The fate of surface roots of citrus seedlings in dry soil. J. Exp. Bot. 1994, 45, 1639–1645. [Google Scholar] [CrossRef]
- Espeleta, J.F.; Eissenstat, D.M. Responses of citrus fine roots to localized soil drying: A comparison of seedlings with adult fruiting trees. Tree Physiol. 1998, 18, 113–119. [Google Scholar] [CrossRef] [PubMed]
- Zaher-Ara, T.; Boroomand, N.; Sadat-Hosseini, M. Physiological and morphological response to drought stress in seedlings of ten citrus. Trees 2016, 30, 985–993. [Google Scholar] [CrossRef]
- Raghavendra, A.S.; Gonugunta, V.K.; Christmann, A.; Grill, E. ABA perception and signalling. Trends Plant Sci. 2010, 15, 395–401. [Google Scholar] [CrossRef]
- Iglesias, D.J.; Cercós, M.; Colmenero-Flores, J.M.; Naranjo, M.A.; Ríos, G.; Carrera, E.; Ruiz-Rivero, O.; Lliso, I.; Morillon, R.; Tadeo, F.R. Physiology of citrus fruiting. Braz. J. Plant Physiol. 2007, 19, 333–362. [Google Scholar] [CrossRef]
- Munné-Bosch, S.; Alegre, L. Die and let live: Leaf senescence contributes to plant survival under drought stress. Funct. Plant Biol. 2004, 31, 203–216. [Google Scholar] [CrossRef] [PubMed]
- Roopaei, M.; Rad, P.; Choo, K.R. Cloud of things in smart agriculture: Intelligent irrigation monitoring by thermal imaging. IEEE Cloud Comput. 2017, 4, 10–15. [Google Scholar] [CrossRef]
- Fares, A.; Alva, A.K. Evaluation of capacitance probes for optimal irrigation of citrus through soil moisture monitoring in an entisol profile. Irrig. Sci. 2000, 19, 57–64. [Google Scholar] [CrossRef]
- Rusu, C.; Krozer, A.; Johansson, C.; Ahrentorp, F.; Pettersson, T.; Jonasson, C.; Rösevall, J.; Ilver, D.; Terzaghi, M.; Chiatante, D. Miniaturized wireless water content and conductivity soil sensor system. Comput. Electron. Agr. 2019, 167, 105076. [Google Scholar] [CrossRef]
- Jin, C.H. Preliminary study on water requirement index of citrus. Southwest China J. Agric. Sci. 1988, 3, 4. Available online: https://www.cnki.com.cn/Article/CJFDTotal-XNYX198802003.htm (accessed on 10 March 2020).
- Matilla, D.M.; Murciego, A.L.; Jiménez-Bravo, D.M.; Mendes, A.S.; Leithardt, V.R. Low-cost Edge Computing devices and novel user interfaces for monitoring pivot irrigation systems based on Internet of Things and LoRaWAN technologies. Biosyst. Eng. 2022, 223, 14–29. [Google Scholar] [CrossRef]
- Williams, M.L.; James, W.P.; Rose, M.T. Variable segmentation and ensemble classifiers for predicting dairy cow behaviour. Biosyst. Eng. 2019, 178, 156–167. [Google Scholar] [CrossRef]
- Krisnawijaya, N.N.K.; Tekinerdogan, B.; Catal, C.; van der Tol, R. Data analytics platforms for agricultural systems: A systematic literature review. Comput. Electron. Agr. 2022, 195, 106813. [Google Scholar] [CrossRef]
- Melnyk, C.W.; Meyerowitz, E.M. Plant grafting. Curr. Biol. 2015, 25, R183–R188. [Google Scholar] [CrossRef] [PubMed]
- Hussain, S.; Curk, F.; Anjum, M.A.; Pailly, O.; Tison, G. Performance evaluation of common clementine on various citrus rootstocks. Sci. Hortic. 2013, 150, 278–282. [Google Scholar] [CrossRef]
- Forner-Giner, M.A.; Continella, A.; Grosser, J.W. Citrus rootstock breeding and selection. In The Citrus Genome; Springer: Cham, Switzerland, 2020; pp. 49–74. [Google Scholar]
- Kang, S.; van Iersel, M.W.; Kim, J. Plant root growth affects FDR soil moisture sensor calibration. Sci. Hortic. 2019, 252, 208–211. [Google Scholar] [CrossRef]
- Wang, K.; Li, W.; Deng, L.; Lyu, Q.; Zheng, Y.; Yi, S.; Xie, R.; Ma, Y.; He, S. Rapid detection of chlorophyll content and distribution in citrus orchards based on low-altitude remote sensing and bio-sensors. Int. J. Agric. Biol. Eng. 2018, 11, 164–169. [Google Scholar] [CrossRef]
- Savla, A.; Israni, N.; Dhawan, P.; Mandholia, A.; Bhadada, H.; Bhardwaj, S. Survey of classification algorithms for formulating yield prediction accuracy in precision agriculture. In Proceedings of the 2015 International Conference on Innovations in Information, Embedded and Communication Systems (ICIIECS), Coimbatore, India, 19–20 March 2015; IEEE: New York, NY, USA, 2015; pp. 1–7. [Google Scholar]
- Kotsiantis, S.B.; Zaharakis, I.D.; Pintelas, P.E. Machine learning: A review of classification and combining techniques. Artif. Intell. Rev. 2006, 26, 159–190. [Google Scholar] [CrossRef]
- Kotsiantis, S.B.; Zaharakis, I.; Pintelas, P. Supervised machine learning: A review of classification techniques. Emerg. Artif. Intell. Appl. Comput. Eng. 2007, 160, 3–24. [Google Scholar]
- Chauhan, V.K.; Dahiya, K.; Sharma, A. Problem formulations and solvers in linear SVM: A review. Artif. Intell. Rev. 2019, 52, 803–855. [Google Scholar] [CrossRef]
- Apicella, A.; Donnarumma, F.; Isgrò, F.; Prevete, R. A survey on modern trainable activation functions. Neural Netw. 2021, 138, 14–32. [Google Scholar] [CrossRef] [PubMed]
- Lubana, E.S.; Dick, R.; Tanaka, H. Beyond batchnorm: Towards a unified understanding of normalization in deep learning. Adv. Neural Inf. Process. Syst. 2021, 34, 4778–4791. [Google Scholar]
- Khan, H.; Haq, I.U.; Munsif, M.; Mustaqeem; Khan, S.U.; Lee, M.Y. Automated Wheat Diseases Classification Framework Using Advanced Machine Learning Technique. Agriculture 2022, 12, 1226. [Google Scholar] [CrossRef]
No. | Date | Preset VWC/% | Information Explanation |
---|---|---|---|
1 | 5/31 | 80 | Bad waterlogging, root system obviously blackened, sticky, smelly |
2 | 5/31 | 40 | Healthy, but insignificant growth |
3 | 6/10 | 20 | Slight drought, little change in chlorophyll, little root growth, and no significant change overall |
4 | 6/10 | 60 | Waterlogging, pronounced growth, new root system of slender roots, older roots look yellowish-brown in color, a small amount of mucus and odor |
5 | 6/23 | 50 | Healthy, a small amount of leaf loss during the experiment due to previous drought experiments on this seedling; the root system was in a normal state but growth was low |
6 | 6/23 | 30 | Healthy, new thick root system but small quantities |
7 | 6/23 | 45 | Healthy, significant increase in root density, a high number of new shoots |
8 | 7/13 | 15 | Drought, significant chlorophyll changes, no root growth |
9 | 7/13 | 10 | Drought, significant chlorophyll changes, no root growth |
10 | 7/13 | 55 | Slight waterlogging, overall yellowish roots, the new root system with slender roots and lush growth, a small amount of mucus but no odor |
Preset VWC/% | Difference in Chlorophyll | Mean | Percentage Difference Value | |||
---|---|---|---|---|---|---|
1 | 2 | 3 | 4 | |||
20 | −1.3 | −1.1 | −5.4 | −3.3 | −2.775 | 4.96% |
15 | −9.8 | −0.8 | −2.7 | −7.0 | −5.075 | 9.06% |
10 | −16.2 | −6.3 | −2.4 | −1.5 | −6.6 | 11.78% |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Dai, Q.; Chen, Z.; Li, Z.; Song, S.; Xue, X.; Lv, S.; Wang, Y.; Guo, Y. A Monitoring, Evaluation, and Prediction System for Slight Water Stress in Citrus Seedlings Based on an Improved Multilayer Perceptron Model. Agronomy 2024, 14, 808. https://doi.org/10.3390/agronomy14040808
Dai Q, Chen Z, Li Z, Song S, Xue X, Lv S, Wang Y, Guo Y. A Monitoring, Evaluation, and Prediction System for Slight Water Stress in Citrus Seedlings Based on an Improved Multilayer Perceptron Model. Agronomy. 2024; 14(4):808. https://doi.org/10.3390/agronomy14040808
Chicago/Turabian StyleDai, Qiufang, Ziwei Chen, Zhen Li, Shuran Song, Xiuyun Xue, Shilei Lv, Yuan Wang, and Yuanhang Guo. 2024. "A Monitoring, Evaluation, and Prediction System for Slight Water Stress in Citrus Seedlings Based on an Improved Multilayer Perceptron Model" Agronomy 14, no. 4: 808. https://doi.org/10.3390/agronomy14040808