Smart Irrigation IoT Solution using Transfer Learning for Neural Networks

A. Jalili; A. Risheh; E. Nazerfard

REVIEW 5 major objections 6 minor 29 references

Smart Irrigation IoT Solution using Transfer Learning for Neural Networks

T0 review · 5 major / 6 minor · reviewed 2026-08-27 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A transfer-learned neural network predicts deep-soil moisture six to nine minutes ahead within a mean squared error of 0.05, using four moisture sensors and a pre-trained model.

desk verdict A useful public soil-moisture dataset and a strong ANN result, but the reliability claim is built on dry-down data only and the SVR baseline is mislabeled. read the letter →

arxiv 2009.12747 v1 pith:AJEQ24KJ submitted 2020-09-27 cs.LG

A. Risheh , A. Jalili , E. Nazerfard This is my paper · ORCID

classification cs.LG

keywords smartirrigationgreenhousesoilmoisturepredictiontransferlearningartificialneuralnetworksinternetofthingssupportvectorregressionfine-tuning

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper builds a smart-irrigation system for greenhouses that predicts the moisture of the deepest soil layer 6–9 minutes in advance from four current moisture readings, using a single-hidden-layer neural network. It reports that this network reaches a mean squared error below 0.05 on its test data, whereas support vector regression needs many more training samples and fails to move from one soil to another. To make the neural network practical on low-power IoT devices, the paper uses transfer learning: start from a network trained on one soil and fine-tune it on another. The paper claims this gives faster convergence, acceptable error with far fewer samples, and a way to add new climate sensors without retraining from scratch.

What carries the argument

The load-bearing object is a single-hidden-layer feedforward neural network with four input nodes (current moisture at four depths), four rectified-linear hidden units, and one output node that predicts deepest-layer moisture three sampling periods ahead; dropout is used during training to prevent overfitting. Transfer learning reuses weights trained on one soil as the starting point for training on another soil, which the paper shows converges faster and starts from a better error value. For adding new environmental sensors, the paper duplicates the network with additional input and hidden neurons, runs the original and extended models side by side, and blends their predictions as $\alpha Y_1 + \beta Y_2$, updating $\alpha$ and $\beta$ by gradient descent from $\alpha = 0.999$, $\beta = 0.001$, and replacing the old model once $\beta$ exceeds 0.9.

What would settle it

Collect a dataset that includes irrigation events, with moisture sensors recording through watering and subsequent redistribution, and evaluate the same fine-tuned network on those post-irrigation samples; if the mean squared error on those samples exceeds 0.05, the paper's reliability claim is refuted for the controlled-use case.

Watch

Extended reading notes

Core claim

The paper's central claim is that transfer learning makes neural-network irrigation control reliable enough for greenhouse use: when the network trained on Soil1 is used as the initial weights for training on Soil2, it predicts the deepest layer's moisture three sampling periods ahead with $\mathrm{MSE} < 0.05$, the paper's reliability threshold. On 1000 test samples of Soil2 the pre-trained model has sum of errors 2.2272 and MSE 0.0124, while a model trained from scratch on Soil1's 4000 samples reaches MSE $8.4409 \times 10^{-5}$. The transferred model reaches its local minimum in about 8 epochs, versus 18 for a fresh network, and the paper finds that a new model trained on Soil2 alone needs roughly 2600 samples, about five days of collection, to match the pre-trained model's accuracy. The paper concludes that transfer learning is superior in performance and keeps the system within its reliability constraint throughout training.

Load-bearing premise

The model is trained only on soil-moisture dry-down data collected while no irrigation was applied, so the paper assumes that the same learned mapping also predicts moisture correctly during and just after irrigation events.

Editorial extensions

If this is right

A greenhouse can deploy irrigation control after a few days of data collection by fine-tuning a network pre-trained on another soil, instead of waiting the roughly five days needed to train a new model to the same accuracy.
With four moisture sensors, a model trained on one soil can be adapted to a second soil using 4000 samples while keeping prediction error below the 0.05 reliability threshold.
Adding temperature or light sensors does not require rebuilding the system; the parallel source-and-target networks with linear blending allow gradual handover without a performance drop.
Support vector regression is not a reliable fallback across soils in the paper's setting: the Soil1 SVR model applied to Soil2 data gives roughly 40% prediction error, which the paper treats as unacceptable.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The paper's data contain only drying phases because the soils were not irrigated during collection, so the trained model is validated on dry-down trajectories only; whether the same network stays below 0.05 error immediately after an irrigation event is not tested and should be verified before deployment.
The claim that transfer learning is superior 'at all times' refers to the specific training runs reported, not a general guarantee; changing soil type, sensor spacing, sampling interval, or season could alter how many fine-tuning samples are needed.
If the sampling period grows to 30 minutes, as the paper notes can happen with power constraints, the advantage of pre-trained initialization should increase because new-soil training data become even scarcer; this is a direct quantitative extension.
Section IV states that the transfer-learning demonstration was 'very limited' and that constraints prevented a full domain-change experiment, so the strongest measured evidence is the Soil1-to-Soil2 fine-tuning while the climate-sensor extension is presented as a method rather than a fully tested result.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Request a human review

A listed scientist reviews the paper for a fee and the review publishes here regardless of verdict. See the reviewers or get listed.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

This is a short conference paper with a genuinely useful new dataset and one very strong empirical result, but its headline reliability claim is not backed by the experiments as written. The dataset—soil-moisture readings at four depths from two soils, collected over about 26,000 points and posted publicly—is a real contribution. The single-hidden-layer ANN predicts the deepest layer's moisture 6–9 minutes ahead and reports MSE 8.44e-05 on a 1,000-sample test set. That is a lot better than the paper's own SVR baseline (MSE 0.0298), and the authors deserve credit for publishing data and code.

However, the load-bearing issue is the one the stress-test note identifies. Section II explicitly says "none of soils were irrigated during the time intervals." Every training and test sample is therefore from a dry-down phase, where moisture is monotonically decreasing. The deployed system must make decisions during and right after irrigation, when moisture rises and wetting fronts move through the layers. The model has never seen that regime. The phrase "achieves our practical reliability constraints (MSE < 0.05) at all times" in Section IV is an extrapolation, not an experimental result. That is the main gap, and it directly affects the system-level conclusion.

There are also two secondary problems. First, the SVR baseline is not actually SVR. Equations (1)–(2) define a linear model trained with an Adam optimizer on a plain sum of errors—no kernel, no epsilon-insensitive loss, no regularization. Comparing your ANN to that may be fine, but calling it SVR and concluding "SVR requires higher training data" overstates the comparison. Second, the transfer-learning advantage is never quantified by the fine-tuned model's test MSE. The paper shows faster convergence and a better starting point, and a sample-efficiency experiment that uses the pre-trained model zero-shot. That is suggestive but does not measure how much fine-tuning reduces the data requirement. Minor: the ANN result is a single run with no error bars, and the sensor-blending algorithm in Section V is described but not tested.

I want to be fair: the paper is not a waste of time. It identifies a practical problem, publishes a useful dataset, and the transfer-learning direction is reasonable. But as it stands, the central reliability claim is unvalidated, and the SVR comparison is mislabeled. A serious referee could require new data collected during actual irrigation cycles and a proper SVR baseline. Given the public dataset and the clean ANN result, I would not desk-reject this; I would send it to review with a clear request for those experiments. My call: revise and resubmit with real irrigation-transient data.

Referee Report

5 major / 6 minor

Summary. The paper proposes a smart irrigation system for greenhouses based on an artificial neural network (ANN) with four soil-moisture sensors, predicting the deepest layer's moisture 6–9 minutes ahead. The authors collected a dataset from two soils (with no irrigation during collection), compared the ANN against a support vector regression (SVR) baseline, and proposed transfer learning to adapt a model pre-trained on one soil to another with reduced data and lower processing power. They also sketch a method for adding new climate sensors by blending source and target model predictions, and present an IoT architecture using MQTT-SN. The central claim is that transfer learning achieves a practical reliability constraint (MSE < 0.05) at all times for the system.

Significance. If the claims were fully validated, the paper would offer a practical contribution to IoT-based smart irrigation, with the notable strengths of a publicly available dataset, a clear performance target, and a concrete system architecture. The empirical comparison, however, has several load-bearing gaps: the SVR baseline is not a true kernel SVR, the transfer-learning advantage is not measured by final fine-tuned test MSE, and the dataset contains no irrigation events—so the 'at all times' reliability claim is untested under the system's actual operating conditions. The central idea is plausible, but the current evidence does not support the stated engineering guarantees.

major comments (5)

[III-A, Eqs. (1)-(2)] The model defined in Eqs. (1)-(2) is ordinary linear regression, not support vector regression. A proper SVR formulation requires an epsilon-insensitive loss and, typically, a kernel; the optimization in Eq. (2) is a least-squares-style linear fit. Consequently, the claim that 'SVR requires higher training data' (Section III-A) is not established by the experiments; the comparison is between an ANN and a linear regressor, not between an ANN and SVR.
[II] The dataset was collected with no irrigation: 'none of soils were irrigated during the time intervals.' Since the deployed system is an irrigation controller, it must act during and after watering events, when moisture rises and wetting fronts propagate. The model is trained and tested only on dry-down dynamics, so the statement in Section IV that transfer learning 'achieves our practical reliability constraints (MSE < 0.05) at all times for the system' is unvalidated for the system's actual operation. The authors should either validate on data that includes irrigation events or substantially weaken the reliability claim.
[IV] The transfer-learning advantage is never quantified by the final fine-tuned test MSE on Soil2. The zero-shot MSE (0.0124) is reported, but the figures compare convergence speed and starting-point error, not the converged test error of the fine-tuned model. The claim that transfer learning 'achieves our practical reliability constraints (MSE < 0.05) at all times' is therefore unsupported: the reader cannot tell whether the fine-tuned model's MSE on Soil2 test data meets the 0.05 threshold.
[III-B] The ANN result (MSE 8.4409e-05) is a single run on a single train/test split, with no repeated experiments, confidence intervals, or error bars. Given the small network and the limited data, the reported accuracy could be sensitive to initialization or data split. Hyperparameters such as the dropout rate, the number of hidden units (q=4 is stated, but not the dropout probability), learning rate, and training details are not reported, which limits reproducibility and makes it difficult to assess the robustness of the claimed three-orders-of-magnitude improvement over the baseline.
[V, Algorithm 1] The proposed method for adding new sensors via the blending rule αY1+βY2=Yans is presented without any empirical validation. No experiments demonstrate that the procedure converges, that the threshold β>0.9 yields a working replacement, or that the blended model maintains the MSE<0.05 constraint during the transition. The claim that this method allows adding sensors 'without any disruption in its performance' is therefore not supported by the data in the paper.

minor comments (6)

[V] The text contains several typos and misspelled words, including 'denition', 'scientic debates', and 'specic context'; these should be corrected.
[III-A] The term 'Sum of error' is used but never defined. It appears to be the sum of absolute errors, but the units and computation should be specified, and the metric should be stated consistently (e.g., MAE).
[IV] The sentence 'we have seen that a pre-trained model with 4,000 training data of Soil1, on 1000 test data of Soil2, had sum of error less than 3' appears to conflict with the earlier reported sum of error 2.2272; please clarify whether these are different runs or different subsets.
[II] The claim that the two soils 'show clearly that the rate of water absorption varies' is based on visual inspection of Figure 1; a quantitative comparison, such as a statistical test or computed dry-down rates, would be more appropriate.
[III-B] All figures plot MSE multiplied by 10, but this rescaling is not stated consistently in every figure caption; please note it in each relevant caption.
[References] References [8] and [11] are raw Kaggle URLs; please provide full repository descriptions, access dates, and version information for archival reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the dry-down-only data gap is an external-validity concern, not a derivation that reduces to its own inputs.

full rationale

The paper's claims are empirical rather than definitional. The neural network and transfer-learning comparisons are evaluated on held-out test data, and no fitted parameter is renamed as a prediction: the SVR and ANN models are trained on Soil1/Soil2 data and then tested on unseen Soil2 samples, with reported MSE values. The transfer-learning advantage is supported by convergence curves and sample-efficiency experiments, which are standard empirical comparisons rather than constructions that presuppose the conclusion. The paper's own quoted limitation, 'none of soils were irrigated during the time intervals' (Section II), means the reliability claim 'MSE < 0.05 at all times for the system' (Section IV) is not validated for irrigation transients, but this is a correctness/generalization gap, not circularity. The self-references are to the authors' own public dataset and code repositories, and they are used as reproducibility aids rather than as load-bearing mathematical premises; no uniqueness theorem or ansatz is imported from self-citations. Therefore, no circular step can be exhibited with a specific equation-to-equation reduction.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The paper's central claims rest on hand-chosen hyperparameters (q=4, epochs=20, alpha/beta), an unstated dropout rate, and several domain assumptions about the moisture data, most notably that non-irrigated dry-down data support irrigation-control decisions. No new physical entities are introduced.

free parameters (5)

number of hidden nodes q = 4
The single hidden layer size is chosen by hand (Section III-B); no ablation is shown.
dropout rate
The paper says a dropout layer is used but never specifies the dropout rate (Section III-B).
training epochs = 20
All models are trained for 20 epochs by design (Section III-A/B), which may limit convergence claims.
alpha and beta initial weights = 0.999 and 0.001
The blending weights in Algorithm 1 are hand-initialized and the switch threshold beta>0.9 is arbitrary (Section V).
prediction horizon = 3T (6-9 minutes)
The prediction target is the deepest moisture at 3T, chosen to avoid overflow; this design choice is not varied or justified quantitatively.

assumptions (4)

domain assumption Moisture dynamics are predictable from current moisture in four layers via a feedforward network
The entire training approach assumes a static mapping from current readings to future moisture, which neglects history and external forcing.
domain assumption Dry-down data without irrigation are representative of operational conditions
Section II states no irrigation during collection; the paper does not test on wetting events, so this assumption is unverified.
ad hoc to paper The linear blending of two model outputs converges and safely replaces the source model
Algorithm 1 is proposed with no proof, no convergence analysis, and no experiments.
domain assumption The reliability threshold (deep layer within 5%, 20% deviation kills crop) applies to the prediction error
The paper cites [12] for thresholds but does not link MSE directly to crop outcomes.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Smart Irrigation IoT Solution using Transfer Learning for Neural Networks." pith.science (2026). https://pith.science/paper/AJEQ24KJ

@misc{pith2026200912747,
  author       = {Pith},
  title        = {Pith review of: Smart Irrigation IoT Solution using Transfer Learning for Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AJEQ24KJ}},
  note         = {Machine review of arXiv:2009.12747}
}

read the original abstract

In this paper we develop a reliable system for smart irrigation of greenhouses using artificial neural networks, and an IoT architecture. Our solution uses four sensors in different layers of soil to predict future moisture. Using a dataset we collected by running experiments on different soils, we show high performance of neural networks compared to existing alternative method of support vector regression. To reduce the processing power of neural network for the IoT edge devices, we propose using transfer learning. Transfer learning also speeds up training performance with small amount of training data, and allows integrating climate sensors to a pre-trained model, which are the other two challenges of smart irrigation of greenhouses. Our proposed IoT architecture shows a complete solution for smart irrigation.

Figures

Figures reproduced from arXiv: 2009.12747 by the authors.

**Figure 1.** Two types of soil moisture comparison. in Soil2 the water remains longer in the first layer. We also see water passes second and third layers faster in Soil1, but in Soil2 the moisture slowly reaches the last layer. These show clearly that the rate of water absorption varies between the two pots. Moreover, the absorption rate across different layers of each soil are different (similar observations are made in [9]). … view at source ↗

**Figure 2.** Training error, both MSE and sum of errors, of SVR [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Schematic of neural network with dropped unit and [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

Figures from the paper (8 more)

**Figure 4.** Figure 4: Training error, both MSE and sum of errors, of ANN [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

**Figure 5.** Figure 5: New model error result with Soil2 data by sum of error [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]

**Figure 6.** Figure 6: Train pre-trained model with 4000 new data of Soil2 [PITH_FULL_IMAGE:figures/full_fig_p004_6.png]

**Figure 7.** Figure 7: Pre-trained and new model comparison about 8 epochs for pre-trained model and 18 for experiment 1. Comparing these results shows the transfer learning is superior in performance and achieves our practical reliability constraints (MSE < 0.05) at all times for the system…

**Figure 8.** Figure 8: Accuracy by training data size from 50 to 2500 [PITH_FULL_IMAGE:figures/full_fig_p005_8.png]

**Figure 9.** Figure 9: New model generated after new variable We initialize α and β with 0.999 and 0.001 as we know our first model has been trained before. We will train our models base on Yans with gradient descend optimization and the linear regression separately. In this section, we want…

**Figure 10.** Figure 10: An overview of the proposed system’s architecture is depicted. The three major subsystems, namely endpoint, [PITH_FULL_IMAGE:figures/full_fig_p007_10.png]

**Figure 11.** Figure 11: Network stack diagram. The physical and MAC layers [PITH_FULL_IMAGE:figures/full_fig_p007_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 28 canonical work pages

[1]

Smart irrigation: A smart drip irrigation system using cloud, android and data mining,

S. Ghosh, S. Sayyed, K. Wani, M. Mhatre, and H. A. Hingoliwala, “Smart irrigation: A smart drip irrigation system using cloud, android and data mining,” in 2016 IEEE International Conference on Advances in Electronics, Communication and Computer Technology (ICAECCT) , pp. 236–239, IEEE, 2016

work page 2016
[2]

Mobile integrated smart irrigation management and monitoring system using iot,

S. Vaishali, S. Suraj, G. Vignesh, S. Dhivya, and S. Udhayakumar, “Mobile integrated smart irrigation management and monitoring system using iot,” in 2017 International Conference on Communication and Signal Processing (ICCSP) , pp. 2164–2167, IEEE, 2017

work page 2017
[3]

Iot based smart irrigation monitoring and controlling system,

S. B. Saraf and D. H. Gawali, “Iot based smart irrigation monitoring and controlling system,” in 2017 2nd IEEE International Conference on Recent Trends in Electronics, Information & Communication Technology (RTEICT), pp. 815–819, IEEE, 2017

work page 2017
[4]

Neural network- based irrigation control for precision agriculture,

F. Capraro, D. Patino, S. Tosetti, and C. Schugurensky, “Neural network- based irrigation control for precision agriculture,” in 2008 IEEE Inter- national Conference on Networking, Sensing and Control , pp. 357–362, IEEE, 2008

work page 2008
[5]

Machine learning-based irrigation control optimization,

A. Murthy, C. Green, R. Stoleru, S. Bhunia, C. Swanson, and T. Chas- pari, “Machine learning-based irrigation control optimization,” in Pro- ceedings of the 6th ACM International Conference on Systems for Energy-Efﬁcient Buildings, Cities, and Transportation , pp. 213–222, 2019

work page 2019
[6]

A machine learning based smart irrigation system with lora p2p networks,

Y . Chang, T. Huang, and N. Huang, “A machine learning based smart irrigation system with lora p2p networks,” in 2019 20th Asia-Paciﬁc Network Operations and Management Symposium (APNOMS) , pp. 1–4, 2019. 8

work page 2019
[7]

Smart irrigation with embedded system,

K. K. Namala, K. K. Prabhu A V, A. Math, A. Kumari, and S. Kulkarni, “Smart irrigation with embedded system,” in2016 IEEE Bombay Section Symposium (IBSS), pp. 1–5, 2016

work page 2016
[8]

A. Jalili. https://www.kaggle.com/amirmohammdjalili/ soil-moisture-dataset

work page

Show all 29 references

[9]

Do-it-yourself (diy) science instrument water absorption of soil apparatus,

J. Genovia, E. Jarantilla-castro, E. Sedurifa, F. J. Sequete, D. Capilitan, and S. J. Malayao, “Do-it-yourself (diy) science instrument water absorption of soil apparatus,” 05 2016

2016
[10]

A tutorial on support vector regression,

A. J. Smola and B. Sch ¨olkopf, “A tutorial on support vector regression,” Statistics and computing , vol. 14, no. 3, pp. 199–222, 2004

2004
[11]

A. Risheh. https://www.kaggle.com/alirisheh/support-vector-regression
[12]

Understanding soil water content and thresholds for irrigation management,

S. Datta, S. Taghvaeian, and J. Stivers, “Understanding soil water content and thresholds for irrigation management,” 06 2017

2017
[13]

An artiﬁcial neural network (p, d, q) model for timeseries forecasting,

M. Khashei and M. Bijari, “An artiﬁcial neural network (p, d, q) model for timeseries forecasting,” Expert Syst. Appl. , vol. 37, pp. 479–489, 01 2010

2010
[14]

Forecasting with artiﬁcial neural networks: The state of the art,

P. Zhang, E. Patuwo, and M. Hu, “Forecasting with artiﬁcial neural networks: The state of the art,” International Journal of Forecasting , vol. 14, pp. 35–62, 03 1998

1998
[15]

Dropout: A simple way to prevent neural networks from overﬁt- ting,

N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhut- dinov, “Dropout: A simple way to prevent neural networks from overﬁt- ting,” Journal of Machine Learning Research, vol. 15, no. 56, pp. 1929– 1958, 2014

1929
[16]

Efﬁcient classiﬁcation across multiple database relations: a crossmine approach,

X. Yin, J. Han, J. Yang, and P. S. Yu, “Efﬁcient classiﬁcation across multiple database relations: a crossmine approach,” IEEE Transactions on Knowledge and Data Engineering, vol. 18, no. 6, pp. 770–783, 2006

2006
[17]

Classiﬁer ensembles with a random linear oracle,

L. I. Kuncheva and J. J. Rodriguez, “Classiﬁer ensembles with a random linear oracle,” IEEE Transactions on Knowledge and Data Engineering , vol. 19, no. 4, pp. 500–508, 2007

2007
[18]

A lazy approach to associative classiﬁcation,

E. Baralis, S. Chiusano, and P. Garza, “A lazy approach to associative classiﬁcation,” IEEE Transactions on Knowledge and Data Engineering, vol. 20, no. 2, pp. 156–171, 2008

2008
[19]

Semi-supervised learning literature survey,

X. J. Zhu, “Semi-supervised learning literature survey,” tech. rep., University of Wisconsin-Madison Department of Computer Sciences, 2005

2005
[20]

Text classiﬁcation from labeled and unlabeled documents using em,

K. Nigam, A. Mccallum, S. Thrun, and T. Mitchell, “Text classiﬁcation from labeled and unlabeled documents using em,” Machine Learning , vol. 39, pp. 103–134, 05 2000

2000
[21]

Combining labeled and unlabeled data with co-training,

A. Blum and T. Mitchell, “Combining labeled and unlabeled data with co-training,” Proceedings of the Annual ACM Conference on Computational Learning Theory , 10 2000

2000
[22]

Transductive inference for text classiﬁcation using support vector machines,

T. Joachims, “Transductive inference for text classiﬁcation using support vector machines,” ICML, 08 2001

2001
[23]

Class noise handling for effective cost-sensitive learning by cost-guided iterative classiﬁcation ﬁltering,

Xingquan Zhu and Xindong Wu, “Class noise handling for effective cost-sensitive learning by cost-guided iterative classiﬁcation ﬁltering,” IEEE Transactions on Knowledge and Data Engineering, vol. 18, no. 10, pp. 1435–1440, 2006

2006
[24]

Test-cost sensitive classiﬁcation on data with missing values,

Qiang Yang, C. Ling, X. Chai, and Rong Pan, “Test-cost sensitive classiﬁcation on data with missing values,” IEEE Transactions on Knowledge and Data Engineering , vol. 18, no. 5, pp. 626–638, 2006

2006
[25]

A survey on transfer learning,

S. J. Pan and Q. Yang, “A survey on transfer learning,” IEEE Transac- tions on Knowledge and Data Engineering , vol. 22, no. 10, pp. 1345– 1359, 2010

2010
[26]

Identifying climate-smart agriculture research needs,

E. Torquebiau, C. Rosenzweig, A. M. Chatrchyan, N. Andrieu, and R. Khosla, “Identifying climate-smart agriculture research needs,” 2018

2018
[27]

Water conservation poten- tial of smart irrigation controllers on st. augustinegrass,

M. McCready, M. D. Dukes, and G. Miller, “Water conservation poten- tial of smart irrigation controllers on st. augustinegrass,” Agricultural water management, vol. 96, no. 11, pp. 1623–1632, 2009

2009
[28]

Per- formance evaluation of mqtt and coap via a common middleware,

D. Thangavel, X. Ma, A. Valera, H.-X. Tan, and C. K.-Y . Tan, “Per- formance evaluation of mqtt and coap via a common middleware,” in 2014 IEEE ninth international conference on intelligent sensors, sensor networks and information processing (ISSNIP) , pp. 1–6, IEEE, 2014

2014
[29]

Mqtt for sensor networks (mqtt-sn) protocol speciﬁcation,

A. Stanford-Clark and H. L. Truong, “Mqtt for sensor networks (mqtt-sn) protocol speciﬁcation,” International business machines (IBM) Corpo- ration version, vol. 1, p. 2, 2013

2013

Pith tools

Reviewed August 27, 2026 · model on record in the stance chip above.

[1] [1]

Smart irrigation: A smart drip irrigation system using cloud, android and data mining,

S. Ghosh, S. Sayyed, K. Wani, M. Mhatre, and H. A. Hingoliwala, “Smart irrigation: A smart drip irrigation system using cloud, android and data mining,” in 2016 IEEE International Conference on Advances in Electronics, Communication and Computer Technology (ICAECCT) , pp. 236–239, IEEE, 2016

work page 2016

[2] [2]

Mobile integrated smart irrigation management and monitoring system using iot,

S. Vaishali, S. Suraj, G. Vignesh, S. Dhivya, and S. Udhayakumar, “Mobile integrated smart irrigation management and monitoring system using iot,” in 2017 International Conference on Communication and Signal Processing (ICCSP) , pp. 2164–2167, IEEE, 2017

work page 2017

[3] [3]

Iot based smart irrigation monitoring and controlling system,

S. B. Saraf and D. H. Gawali, “Iot based smart irrigation monitoring and controlling system,” in 2017 2nd IEEE International Conference on Recent Trends in Electronics, Information & Communication Technology (RTEICT), pp. 815–819, IEEE, 2017

work page 2017

[4] [4]

Neural network- based irrigation control for precision agriculture,

F. Capraro, D. Patino, S. Tosetti, and C. Schugurensky, “Neural network- based irrigation control for precision agriculture,” in 2008 IEEE Inter- national Conference on Networking, Sensing and Control , pp. 357–362, IEEE, 2008

work page 2008

[5] [5]

Machine learning-based irrigation control optimization,

A. Murthy, C. Green, R. Stoleru, S. Bhunia, C. Swanson, and T. Chas- pari, “Machine learning-based irrigation control optimization,” in Pro- ceedings of the 6th ACM International Conference on Systems for Energy-Efﬁcient Buildings, Cities, and Transportation , pp. 213–222, 2019

work page 2019

[6] [6]

A machine learning based smart irrigation system with lora p2p networks,

Y . Chang, T. Huang, and N. Huang, “A machine learning based smart irrigation system with lora p2p networks,” in 2019 20th Asia-Paciﬁc Network Operations and Management Symposium (APNOMS) , pp. 1–4, 2019. 8

work page 2019

[7] [7]

Smart irrigation with embedded system,

K. K. Namala, K. K. Prabhu A V, A. Math, A. Kumari, and S. Kulkarni, “Smart irrigation with embedded system,” in2016 IEEE Bombay Section Symposium (IBSS), pp. 1–5, 2016

work page 2016

[8] [8]

A. Jalili. https://www.kaggle.com/amirmohammdjalili/ soil-moisture-dataset

work page