Pith. sign in

REVIEW 3 major objections 5 minor 45 references

AgriSentinel is the first privacy-enhanced embedded-LLM crop disease alerting system: it adds pixel noise to protect farmer data, classifies the disease on-device, and generates management advice with a fine-tuned language model.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

An integrated mobile system for rice disease alerting that adds Gaussian noise to images for privacy, classifies with a CNN, and answers farmer questions with a fine-tuned GPT-2, but its privacy mechanism is not formally validated.

T0 review reviewed 2026-08-04 challenge →

load-bearing objection A working prototype that mistakes Gaussian obfuscation for differential privacy; the central claim does not survive, but the on-device classification + LLM integration is a real, if incremental, idea. the 3 major comments →

arxiv 2509.09103 v1 pith:NGPE7FUW submitted 2025-09-11 cs.CR

AgriSentinel: Privacy-Enhanced Embedded-LLM Crop Disease Alerting System

Chanti Raju Mylay , Bobin Deng , Zhipeng Cai , Honghui Xu This is my paper
classification cs.CR
keywords Large Language ModelPrivacy and SecuritySmart AgricultureMobile ComputingCrop Disease ClassificationDifferential PrivacyGaussian Noise ObfuscationRice Disease
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

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 proposes AgriSentinel, an on-device system that runs a three-stage pipeline: inject Gaussian noise into crop images to obfuscate sensitive data, classify the disease with a lightweight CNN, then feed the predicted class into a fine-tuned on-device LLM that answers farmers' questions using a curated knowledge pool. The paper's central claim is that this is the first privacy-enhanced embedded-LLM crop disease alerting system, and that the noise injection protects farmers from privacy attacks while preserving usable classification accuracy. On a 1,200-image rice dataset covering blast, brown spot, and bacterial leaf blight, the experiments report accuracy between about 60% at the strongest privacy setting and 91% at the weakest, with the fine-tuned LLM producing answers that closely track expert ground truth. If the system works as described, it offers a concrete path to giving smallholder farmers private, actionable disease management on a mobile device, protecting their image data from being exploited in agricultural markets.

Core claim

The central discovery is architectural: AgriSentinel combines pixel-level Gaussian noise, a lightweight CNN classifier, and a fine-tuned on-device LLM (GPT-2 with a curated Q&A knowledge pool) into one privacy-enhanced alerting pipeline. The paper argues that the noise-injection step constitutes differential privacy, with privacy loss ε = Δf/σ tunable across five levels from 0.06 (very high obfuscation) to 30.0 (very low obfuscation). On the three-class rice disease dataset, the classifier retains most of its utility under moderate noise, and the fine-tuned LLM produces answers that closely track expert-provided ground truth for the same questions. The paper positions this as the first work

What carries the argument

The load-bearing mechanism is the Gaussian noise injection of Eq. (1), x_obf = x + N(0, σ^2), together with the ratio ε = Δf/σ used to quantify privacy loss. This identity carries the entire privacy argument: sensitivity Δf and noise scale σ are combined to produce five obfuscation levels, and the resulting obfuscated images are what the CNN classifier sees. The classification side rests on a small CNN (two 3x3 convolutional layers with max pooling, a 128-unit dense layer, dropout, softmax) trained with categorical cross-entropy and Adam, and the alerting side on a fine-tuned GPT-2 that takes the predicted class as input and returns answers from a curated knowledge pool K = (Q, A).

Load-bearing premise

The whole privacy claim rests on the assumption that adding Gaussian noise to images, with ε reported as Δf/σ in Section 4.1.6 and Table 1, actually blocks attackers from inferring whether a farmer's photo was used in the system—yet the paper derives no sensitivity bound for the classifier and runs no attack to test it.

What would settle it

Run a membership-inference attack on the CNN trained on the obfuscated images: give an attacker the noised image and the model's prediction, and ask whether that image was in the training set. At the noise level claimed to give ε = 0.06, chance-level attack accuracy would support the privacy claim; significantly above-chance accuracy would falsify it. An alternative check is to compute the true sensitivity of the classifier's output to changing one training image and compare it with the Δf values in Table 1.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the privacy claim holds, farmers can receive disease alerts without uploading raw crop images, removing a major vector for membership-inference attacks that could reveal planting patterns and undercut farmers' market pricing power.
  • The reported tables give operators a concrete privacy-utility dial: choosing the noise level for ε = 2.0 keeps accuracy in the 0.81–0.85 range, close to the clean baseline, while ε = 0.06 drops it to 0.60–0.65.
  • Because the LLM runs on-device, the raw image need not leave the phone, and the fine-tuned GPT-2's answers align with expert ground truth on the tested rice disease questions, moving the system beyond simple alerts to management advice.
  • The lightweight CNN survives moderate noise, indicating that mobile deployment is feasible while still providing usable disease classification.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Extension: the reported ε values are computed from image-pixel sensitivity and noise, not from the sensitivity of the classifier's output; a membership-inference attack on the trained CNN would be needed to confirm the guarantee, and the paper runs no such attack.
  • Extension: the same three-stage pipeline could be tested on a larger multi-crop dataset with more disease classes to see whether the accuracy/privacy tradeoff at high noise holds when symptoms are more visually subtle.
  • Extension: since moderate noise slightly improved accuracy over baseline in one of the three classes, an unexplored explanation is that pixel noise acts as implicit data augmentation or regularization; this could be isolated by training with and without noise at fixed epochs.
  • Extension: the design is not tied to GPT-2; newer on-device LLMs could be swapped into the same knowledge-pool fine-tuning step, potentially changing answer quality, though the paper does not address this.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. AgriSentinel is an on-device system that combines Gaussian pixel-noise obfuscation, a lightweight CNN for rice disease classification, and a fine-tuned GPT-2 model that generates management advice from a curated knowledge pool. The paper claims that the noise-injection step is a differential privacy mechanism and that AgriSentinel is the first privacy-enhanced embedded-LLM crop disease alerting system. Experiments are conducted on a 1,200-image rice dataset with three disease classes; results are reported at five obfuscation levels for classification accuracy, precision, recall, and F1, together with qualitative examples of LLM-generated answers.

Significance. If the privacy claim were sound, the paper would provide a practically oriented integration of on-device classification, LLM-based advice, and privacy protection for agricultural data. The system architecture is clearly described and the qualitative LLM outputs suggest a reasonable fine-tuning effort. However, the central contribution rests entirely on the assertion that Eq. (1) constitutes differential privacy and that the values in Table 1 are privacy-loss guarantees. That assertion is not supported by any standard DP analysis: no sensitivity is derived, no δ is given, and the ε=Δf/σ formula is not a recognized DP privacy-loss expression. The claimed privacy guarantee is therefore not established, and removing it leaves a standard CNN plus a fine-tuned GPT-2 with no formal privacy property. The empirical evaluation also lacks statistical rigor, consisting of single-run metrics on a small dataset. The paper's significance is thus primarily as an application/system description, not as a validated privacy mechanism.

major comments (3)
  1. [Section 3.1, Eq. (1); Section 4.1.6, Table 1] The privacy mechanism is not differential privacy. Eq. (1) adds isotropic Gaussian noise to pixel values, which is standard image obfuscation, not a DP guarantee. The text defines ε=Δf/σ without specifying the query or function f, without deriving Δf, and without introducing δ. Under the standard Gaussian mechanism, the identity query over a 224×224×3 normalized image has ℓ2 sensitivity at least sqrt(224·224·3)≈388; achieving ε=0.06 with δ=1e-5 would require σ≈2.2×10^4, not σ=1.5 as in Table 1. Thus the reported ε values are not credible as DP guarantees under any standard reading. This is load-bearing because the abstract, introduction, and conclusion all claim that AgriSentinel 'incorporates a differential privacy mechanism' and 'safeguards crop image privacy.'
  2. [Section 4.1.6, Table 1] The sensitivity values in Table 1 are hand-picked to match the desired ε labels; no function f is defined and no sensitivity calculation is provided. The argument is circular: Δf is chosen so that ε=Δf/σ equals 0.06, 0.30, 2.0, etc. The resulting 'Privacy Loss' is therefore not a measured privacy property but a renamed noise-to-signal ratio. No membership-inference or reconstruction experiment is run to provide empirical support, so the paper offers neither a formal nor an empirical basis for its central privacy claim.
  3. [Section 4.1.1, Tables 2–4] The classification evaluation is not sufficiently rigorous to support the 'comprehensive experiments' claim. The dataset contains only 1,200 images, split once into 80/20 train/test, and all reported metrics appear to come from a single run with no error bars, confidence intervals, or repeated trials. The observation that noise 'enhances robustness' below medium obfuscation rests on small differences (e.g., 0.85 vs. 0.84 in Table 2) that could easily be due to random variation. This is secondary to the privacy issue but affects the paper's claims about maintaining classification performance.
minor comments (5)
  1. [Section 3.1, Eq. (1)] The notation is inconsistent: the equation uses x_obf = x + N(0,σ²), but the surrounding text refers to 'Iobf' and 'Lap(0,σ²)', conflating Gaussian and Laplace noise. This should be corrected.
  2. [Section 4.1.6] No δ parameter is mentioned anywhere in the privacy discussion. Differential privacy requires either (ε,δ) accounting or a pure ε guarantee; the absence of δ makes the privacy-loss formulation incomplete even as an informal description.
  3. [Tables 2–4] The baseline row is labeled '0 (Baseline)' but is not clearly defined as no-noise; including it in the same table as obfuscation levels makes the comparison ambiguous. Clarify whether the baseline uses original images and whether all other rows share the same model architecture.
  4. [Section 4.3, Tables 5–6] The LLM evaluation is entirely qualitative and covers only Brown Spot examples. No quantitative metrics (e.g., BLEU, ROUGE, human ratings) are reported, and no examples for Rice Blast or Bacterial Leaf Blight appear in Tables 5–6. The authors should either add quantitative evaluation or soften the claim that the LLM 'closely aligns' with ground truth.
  5. [General] The manuscript lacks details on GPT-2 fine-tuning: knowledge-pool size, number of training epochs, learning rate, computational cost, and model size. The ACM Reference Format lists the year as 2018 while the submission is dated 2025; this metadata should be corrected.

Circularity Check

2 steps flagged

Privacy loss is self-defined as ε=Δf/σ with no standard DP accounting, and LLM answers are evaluated against the very knowledge pool used for fine-tuning; both claims reduce by construction.

specific steps
  1. self definitional [Section 4.1.6 (Performance Metrics), Eq. ε=Δf/σ; also Section 3.1 Eq. (1)]
    "To compute the overall privacy loss, both sensitivity and noise components are combined in the following manner: ε= Δf/σ. This equation illustrates that privacy loss ε is directly proportional to the sensitivity Δf and inversely proportional to the noise σ applied."

    The paper's 'differential privacy' guarantee is established solely through its own formula ε=Δf/σ, with Δf and σ in Table 1 chosen to hit target ε values. No standard (ε,δ)-DP definition is used, no actual query sensitivity is derived, no δ is given, and no composition over the 224×224×3 pixel values is considered. The reported privacy-loss values (0.06, 0.30, 2.0, 10.0, 30.0) are true by construction (Δf divided by σ), so the central 'privacy-enhanced' claim reduces to the formula the authors wrote down rather than to any measured or proven privacy property.

  2. fitted input called prediction [Section 4.1.5 and Section 4.3 (Tables 5 and 6)]
    "We utilize GPT-2 as the foundational LLM model and fine-tune it using our curated knowledge pool on crop disease management and the predicted crop disease results. ... As shown in Table 5 and Table 6, the fine-tuned LLM generates accurate responses closely aligned with expert-provided ground truth knowledge."

    The knowledge pool K=(Q,A) serves both as the dataset for fine-tuning the LLM (Section 3.3, 4.1.5) and as the 'groundtruth' for evaluating it (Section 4.3, Tables 5–6). The LLM is trained on these exact question–answer pairs, so any alignment with the groundtruth is a measure of memorization on the training set, not predictive generalization. No held-out or unseen questions are reported, making the LLM's success forced by the training data by construction.

full rationale

AgriSentinel's crop disease classification is evaluated against external disease labels, so that component is empirically grounded and not circular. However, the paper's central privacy claim rests on a self-defined formula ε=Δf/σ with arbitrary sensitivity values and no (ε,δ)-DP derivation, making the privacy guarantee equivalent to its own definition. The LLM evaluation is also circular because the fine-tuning dataset doubles as the evaluation groundtruth. Together these are load-bearing and give a score of 6. No self-citation chain is involved; the circularity is via definition and fitted input.

Axiom & Free-Parameter Ledger

2 free parameters · 4 axioms · 1 invented entities

The central privacy claim rests on an invented metric and unverified assumptions. The classification and LLM components rely on standard ML assumptions and external labels, but the privacy component is self-referential.

free parameters (2)
  • Noise scale σ per obfuscation level = 0 (baseline), 0.1, 0.2, 0.5, 1.0, 1.5
    Hand-picked in Table 1 to produce specific ε values; no sensitivity analysis ties σ to the data or query.
  • Sensitivity Δf per obfuscation level = 0.1, 0.3, 1.0, 2.0, 3.0
    Arbitrary values in Table 1; the sensitivity of the image classification function is never derived, so these are free parameters inserted to construct the privacy metric.
axioms (4)
  • ad hoc to paper Adding Gaussian noise to images yields a differential privacy guarantee for the trained model
    Section 3.1 asserts this without proof; standard DP requires noise on the mechanism output with calibrated sensitivity, not arbitrary pixel perturbation.
  • domain assumption The 1,200-image rice dataset is representative and correctly labeled
    Section 4.1.1 uses this dataset; no external validation or error analysis supports generalization.
  • domain assumption The curated knowledge pool provides expert-validated answers
    Section 3.3 fine-tunes GPT-2 on this pool; the correctness of the advice is asserted from references rather than tested.
  • standard math Cross-entropy loss with Adam converges to a usable classifier on noisy images
    Standard deep learning background assumption.
invented entities (1)
  • Privacy loss metric ε = Δf/σ no independent evidence
    purpose: To quantify privacy protection in Tables 1-4 and claim tunable privacy levels
    This metric is introduced by the authors and is not an established DP measure; it is not validated against any privacy attack or formal bound.

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of AgriSentinel: Privacy-Enhanced Embedded-LLM Crop Disease Alerting System." pith.science (2026). https://pith.science/paper/NGPE7FUW

@misc{pith2026250909103,
  author       = {Pith},
  title        = {Pith review of: AgriSentinel: Privacy-Enhanced Embedded-LLM Crop Disease Alerting System},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NGPE7FUW}},
  note         = {Machine review of arXiv:2509.09103}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Crop diseases pose significant threats to global food security, agricultural productivity, and sustainable farming practices, directly affecting farmers' livelihoods and economic stability. To address the growing need for effective crop disease management, AI-based disease alerting systems have emerged as promising tools by providing early detection and actionable insights for timely intervention. However, existing systems often overlook critical aspects such as data privacy, market pricing power, and farmer-friendly usability, leaving farmers vulnerable to privacy breaches and economic exploitation. To bridge these gaps, we propose AgriSentinel, the first Privacy-Enhanced Embedded-LLM Crop Disease Alerting System. AgriSentinel incorporates a differential privacy mechanism to protect sensitive crop image data while maintaining classification accuracy. Its lightweight deep learning-based crop disease classification model is optimized for mobile devices, ensuring accessibility and usability for farmers. Additionally, the system includes a fine-tuned, on-device large language model (LLM) that leverages a curated knowledge pool to provide farmers with specific, actionable suggestions for managing crop diseases, going beyond simple alerting. Comprehensive experiments validate the effectiveness of AgriSentinel, demonstrating its ability to safeguard data privacy, maintain high classification performance, and deliver practical, actionable disease management strategies. AgriSentinel offers a robust, farmer-friendly solution for automating crop disease alerting and management, ultimately contributing to improved agricultural decision-making and enhanced crop productivity.

Figures

Figures reproduced from arXiv: 2509.09103 by Bobin Deng, Chanti Raju Mylay, Honghui Xu, Zhipeng Cai.

Figure 1
Figure 1. Figure 1: AgriSentinel Framework 3.1 Differential Privacy Noise Injection Mechanism We propose a differential privacy mechanism to protect crop data by employing Gaussian noise as an effective obfuscation technique. Gaussian noise introduces controlled random variations to the pixel values of an image, masking identifiable features while preserving the image’s utility for analysis. This approach strikes a balance be… view at source ↗
Figure 2
Figure 2. Figure 2: Three Primary Categories of Rice Plant Diseases [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

45 extracted references · 7 linked inside Pith

  1. [1]

    Meenakshi Aggarwal, Vikas Khullar, Nitin Goyal, and Thomas André Prola. 2024. Resource-efficient federated learning over IoAT for rice leaf disease classification.Computers and Electronics in Agriculture221 (2024), 109001

    2024

  2. [2]

    Aleksandr Andreev, Anton Kotsenko, Oleg Varlamov, Radmir Kim, and Boris Goryachkin. 2024. Text processing using LLM for automatic creation of agricultural crops knowledge bases. InBIO Web of Conferences, Vol. 130. EDP Sciences, 01029

    2024

  3. [3]

    Ons Aouedi, Thai-Hoc Vu, Alessio Sacco, Dinh C Nguyen, Kandaraj Piamrat, Guido Marchetto, and Quoc-Viet Pham. 2024. A survey on intelligent Internet of Things: Applications, security, privacy, and future directions.IEEE Communications Surveys & Tutorials(2024)

    2024

  4. [4]

    Saikat Banerjee, Soumitra Das, and Abhoy Chand Mondal. 2024. A Study of the Application Domain of a Large Language Models in the Agricultural Sector.International Journal of Innovative Research in Computer Science & Technology12, 5 (2024), 74–78

    2024

  5. [5]

    Manoj Kumar Barnwal, A Kotasthane, Nicole Magculia, Prasun K Mukherjee, Serge Savary, Anand Kumar Sharma, Hanwant B Singh, US Singh, AH Sparks, Mukund Variar, et al . 2013. A review on crop losses, epidemiology and disease management of rice brown spot to identify research priorities and knowledge gaps.European Journal of Plant Pathology136 (2013), 443–457

    2013

  6. [6]

    Muluadam Berhan Ejigu. 2021.Morphological characterization of bipolaris oryzae and screening of brown spot disease resistant lowland rice (oryza sativa l.) genotypes under rainfed condition at fogera plain, south gondar administration zone in amhara region, Ethiopia. Ph.D. Dissertation. Bahir Dar University

    2021

  7. [7]

    Hanxiao Chen, Hongwei Li, Guishan Dong, Meng Hao, Guowen Xu, Xiaoming Huang, and Zhe Liu. 2020. Practical membership inference attack against collaborative inference in industrial IoT.IEEE Transactions on Industrial Informatics 18, 1 (2020), 477–487

    2020

  8. [8]

    Wei Chen and Zhiyuan Li. 2024. Octopus v2: On-device language model for super agent.arXiv preprint arXiv:2404.01744 (2024)

    Pith/arXiv arXiv 2024

  9. [9]

    Wei Chen and Zhiyuan Li. 2024. Octopus v3: Technical report for on-device sub-billion multimodal ai agent.arXiv preprint arXiv:2404.11459(2024)

    Pith/arXiv arXiv 2024

  10. [10]

    Wei Chen and Zhiyuan Li. 2024. Octopus v4: Graph of language models.arXiv preprint arXiv:2404.19296(2024)

    Pith/arXiv arXiv 2024

  11. [11]

    Wei Chen, Zhiyuan Li, Zhen Guo, and Yikang Shen. 2024. Octo-planner: On-device language model for planner-action agents.arXiv preprint arXiv:2406.18082(2024)

    Pith/arXiv arXiv 2024

  12. [12]

    Prameeladevi Chillakuru, D Divya, and K Ananthajothi. 2022. Enhanced segmentation with optimized nine-layered CNN-based classification of leaf diseases: an automatic approach for plant disease diagnosis.cybernetics and systems (2022), 1–36

    2022

  13. [13]

    Pat Crill, FL Nuque, BA Estrada, and JM Bandong. 1982. The role of varietal resistance in disease management. Evolution of the Gene Rotation Concept for Rice Blast Control(1982), 103–121

    1982

  14. [14]

    Soumia Zohra El Mestari, Gabriele Lenzini, and Huseyin Demirci. 2024. Preserving data privacy in machine learning systems.Computers & Security137 (2024), 103605

    2024

  15. [15]

    Sam Fujisaka, Larry Harrington, and Peter Hobbs. 1994. Rice-wheat in South Asia: systems and long-term priorities established through diagnostic research.Agricultural Systems46, 2 (1994), 169–187

    1994

  16. [16]

    Sahana D Gowda. 2024. Secure Multiparty Computation: Protocols, Collaborative Data Processing, and Real-World Applications in Industry. InCloud Security. Chapman and Hall/CRC, 160–182. 14 AgriSentinel: Privacy-Enhanced Embedded-LLM Crop Disease Alerting System ACM IH&MMSec’25, June 18–20, 2025, San Jose, CA, USA

    2024

  17. [17]

    Maanak Gupta, Mahmoud Abdelsalam, Sajad Khorsandroo, and Sudip Mittal. 2020. Security and privacy in smart farming: Challenges and opportunities.IEEE access8 (2020), 34564–34584

    2020

  18. [18]

    Hamid Jalalzai, Elie Kadoche, Rémi Leluc, and Vincent Plassier. 2022. Membership Inference Attacks via Adversarial Examples.arXiv preprint arXiv:2207.13572(2022)

    Pith/arXiv arXiv 2022

  19. [19]

    Mohd Javaid, Abid Haleem, Ravi Pratap Singh, and Rajiv Suman. 2022. Enhancing smart farming through the applications of Agriculture 4.0 technologies.International Journal of Intelligent Networks3 (2022), 150–164

    2022

  20. [20]

    Daniel Jimenez, Hugo Dorado, James Cock, Steven D Prager, Sylvain Delerce, Alexandre Grillon, Mercedes Andrade Be- jarano, Hector Benavides, and Andy Jarvis. 2016. From observation to information: data-driven understanding of on farm yield variation.PloS one11, 3 (2016), e0150015

    2016

  21. [21]

    2023.DISEASES OF FIELD AND HORTICULTURAL CROPS AND THEIR MANAGE- MENT VOLUME–II

    BHUPENDRA SINGH KHARAYAT. 2023.DISEASES OF FIELD AND HORTICULTURAL CROPS AND THEIR MANAGE- MENT VOLUME–II. PHI Learning Pvt. Ltd

    2023

  22. [22]

    Chen Liang and Tufail Shah. 2023. IoT in agriculture: The future of precision monitoring and data-driven farming. Eigenpub Review of Science and Technology7, 1 (2023), 85–104

    2023

  23. [23]

    2023.Reaction of Rice Germplasm to Blast and Brownspot Under Natural and Controlled Environment in Mwea, Kirinyaga County, Kenya

    Simon V Lirova. 2023.Reaction of Rice Germplasm to Blast and Brownspot Under Natural and Controlled Environment in Mwea, Kirinyaga County, Kenya. Ph.D. Dissertation. University of Nairobi

    2023

  24. [24]

    John Bethany M Macasero, Nancy P Castilla, Ireneo B Pangga, Leonardo V Marquez, Edwin C Martin, Ulysses G Duque, and Alice G Laborte. 2024. Influence of production situation on the incidence of brown spot of rice (Oryza sativa) caused by Bipolaris oryzae in the Philippines.Plant Pathology73, 2 (2024), 390–403

    2024

  25. [25]

    Shiva Mehta, Vinay Kukreja, and Amit Gupta. 2023. Revolutionizing Maize Disease Management with Federated Learning CNNs: A Decentralized and Privacy-Sensitive Approach. In2023 4th International Conference for Emerging Technology (INCET). IEEE, 1–6

    2023

  26. [26]

    Dipali Mishra, D Deepa, et al. 2021. Automation and integration of growth monitoring in plants (with disease prediction) and crop prediction.Materials Today: Proceedings43 (2021), 3922–3927

    2021

  27. [27]

    2013.Diseases Of Vegetable Crops And Their Integrated Management: A Colour Handbook

    RK Mishra. 2013.Diseases Of Vegetable Crops And Their Integrated Management: A Colour Handbook. New India Publishing Agency

    2013

  28. [28]

    Jianguo Mo, Qingyuan Chen, and Fei Yu. 2012. Analysis of meteorological condition for infection of the pathogen of tobacco brown spot (Alternaria alternate).Plant Diseases and Pests3, 2 (2012), 24

    2012

  29. [29]

    RP Narmadha, N Sengottaiyan, and RJ Kavitha. 2022. Deep transfer learning based rice plant disease detection model. Intelligent Automation & Soft Computing31, 2 (2022)

    2022

  30. [30]

    David F Nettleton, Dimitrios Katsantonis, Argyris Kalaitzidis, Natasa Sarafijanovic-Djukic, Pau Puigdollers, and Roberto Confalonieri. 2019. Predicting rice blast disease: machine learning versus process-based models.BMC bioinformatics 20 (2019), 1–16

    2019

  31. [31]

    Julienne Nguefack, G Ednar Wulff, J Blaise Lekagne Dongmo, F Romain Fouelefack, Daniel Fotio, Joseph Mbo, and Jan Torp. 2013. Effect of plant extracts and an essential oil on the control of brown spot disease, tillering, number of panicles and yield increase in rice.European Journal of Plant Pathology137 (2013), 871–882

    2013

  32. [32]

    Nguyen Bang Nong and Mohammad Tarek. 2023. Surveillance Approaches to Intrusion Detection, Crop Health, and Disease Prevention in Agriculture.Quarterly Journal of Emerging Technologies and Innovations8, 3 (2023), 1–17

    2023

  33. [33]

    2020.Diseases of field crops and their management

    S Parthasarathy, G Thiribhuvanamala, and Kandasamy Prabakar. 2020.Diseases of field crops and their management. CRC Press

    2020

  34. [34]

    Mohammad Fatin Fatihur Rahman, Shurui Fan, Yan Zhang, and Lei Chen. 2021. A comparative study on application of unmanned aerial vehicle systems in agriculture.Agriculture11, 1 (2021), 22

    2021

  35. [35]

    Jean B Ristaino, Pamela K Anderson, Daniel P Bebber, Kate A Brauman, Nik J Cunniffe, Nina V Fedoroff, Cambria Finegold, Karen A Garrett, Christopher A Gilligan, Christopher M Jones, et al. 2021. The persistent threat of emerging plant disease pandemics to global food security.Proceedings of the National Academy of Sciences118, 23 (2021), e2022239118

    2021

  36. [36]

    Ranjan Sapkota, Rizwan Qureshi, Syed Zohaib Hassan, John Shutske, Maged Shoman, Muhammad Sajjad, Fayaz Ali Dharejo, Achyut Paudel, Jiajia Li, Zhichao Meng, et al. 2024. Multi-Modal LLMs in Agriculture: A Comprehensive Review.10.36227/techrxiv. 172651082.24507804/v1(2024)

    arXiv 2024

  37. [37]

    Richa Sharma et al . 2024. Revolutionizing Rice Agriculture: A Machine Learning Approach to Fungal Disease Management for Economic Sustainability. In2024 International Conference on Communication, Computer Sciences and Engineering (IC3SE). IEEE, 798–805

    2024

  38. [38]

    Bruno Silva, Leonardo Nunes, Roberto Estevão, Vijay Aski, and Ranveer Chandra. 2023. GPT-4 as an agronomist assistant? Answering agriculture exams using large language models.arXiv preprint arXiv:2310.06225(2023)

    Pith/arXiv arXiv 2023

  39. [39]

    Vaibhav K Singh and Shipli Chawla. 2012. Cultural Practices: An Ecofriendly Innovative Approach in Plant Disease Management.International Book Publishers and Distributers, Dehradun–2481 (2012), 01–20

    2012

  40. [40]

    Vaibhav K Singh, Yogendra Singh, and Prabhat Kumar. 2012. Diseases of ornamental plants and their management. Eco-friendly innovative approaches in plant disease management(2012), 543–572. 15 ACM IH&MMSec’25, June 18–20, 2025, San Jose, CA, USA Xu, et al

    2012

  41. [41]

    Michael E Sykuta. 2016. Big data in agriculture: property rights, privacy and competition in ag data services.Interna- tional Food and Agribusiness Management Review19 (2016), 57–74

    2016

  42. [42]

    Nicoleta Tantalaki, Stavros Souravlas, and Manos Roumeliotis. 2019. Data-driven decision making in precision agriculture: The rise of big data in agricultural systems.Journal of agricultural & food information20, 4 (2019), 344–380

    2019

  43. [43]

    Asaf Tzachor, Medha Devare, Catherine Richards, Pieter Pypers, Aniruddha Ghosh, Jawoo Koo, S Johal, and Brian King. 2023. Large language models and agricultural extension services.Nature food4, 11 (2023), 941–948

    2023

  44. [44]

    2019.Evaluation of adjuvants in fungicide spray application for the control of alternaria brown spot in South African citrus orchards

    Johannes Gideon Van Zyl. 2019.Evaluation of adjuvants in fungicide spray application for the control of alternaria brown spot in South African citrus orchards. Ph.D. Dissertation. Stellenbosch: Stellenbosch University

    2019

  45. [45]

    Jiajun Xu, Zhiyuan Li, Wei Chen, Qun Wang, Xin Gao, Qi Cai, and Ziyuan Ling. 2024. On-device language models: A comprehensive review.arXiv preprint arXiv:2409.00088(2024). 16

    Pith/arXiv arXiv 2024

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.