Next Article in Journal
Design and Experiment of a Universal Harvesting Platform for Cabbage and Chinese Cabbage
Previous Article in Journal
A Daily Reference Crop Evapotranspiration Forecasting Model Based on Improved Informer
 
 
Article
Peer-Review Record

Towards Secure and Efficient Farming Using Self-Regulating Heterogeneous Federated Learning in Dynamic Network Conditions

Agriculture 2025, 15(9), 934; https://doi.org/10.3390/agriculture15090934
by Sai Puppala * and Koushik Sinha
Reviewer 1:
Reviewer 2: Anonymous
Reviewer 3: Anonymous
Agriculture 2025, 15(9), 934; https://doi.org/10.3390/agriculture15090934
Submission received: 20 March 2025 / Revised: 17 April 2025 / Accepted: 21 April 2025 / Published: 25 April 2025
(This article belongs to the Section Artificial Intelligence and Digital Agriculture)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Comments:

The area focused in the paper is very interesting for the efficient upliftment of agriculture processes. The proposed work basically focuses on the interesting topic of federated learning. However, the work has significant comments as follows:

  1. The work proposed an architecture based on federated learning (FL) to deal with the challenges of agriculture. However, the authors are unable to provide any novelty in the work. The work simply applies existing FL in agriculture without specifying any specific issue related to agriculture and resource utilization in agriculture.
  2. The proposed work seems to be puzzling in terms of different issues like data privacy, data security, network-related issues, operational efficiency, and resource utilization. A very clear and concise distinction is required among them, and each should be defined clearly with the scope of work presented in the paper.
  3. The work provides the literature review related to the proposed work. However, it is very hard to find any research gap or limitation that needs to be focused in this paper.
  4. The proposed architecture is very ambiguous and not clearly explained. It requires a clear explanation for each of its components.
  5. The work is based on federated learning. However, there is lack of FL architecture and its working explanation.
  6. The proposed work talks about the clustering of the devices (integrated into tractors). However, it is very hard to understand why clustering is an important part of the proposed work. If this is the case, what is the basis of the clustering? Can’t the proposed work not implemented without clustering the devices?
  7. The work requires a clear explanation of the alpha score and its significance in the proposed work.
  8. The proposed algorithms are not very clear. There is a tremendous amount of requirement to clearly explain and write the pseudo-code of the proposed algorithms.
  9. The work claims for the enhancement of data security in the proposed work. However it has not proposed any novel work in terms of data security, instead it has employed an existing data encryption method. Hence, if the work is supposed to claim for any novel data security methods, it should be clearly explained.
  10. The paper requires significant work to review for the English language and grammar used in the paper.
Comments on the Quality of English Language

The paper requires significant work to review for the English language and grammar used in the paper.

Author Response

Response Letter for Reviewer Concerns

1. The work proposed an architecture based on federated learning (FL) to deal with the challenges of agriculture. However, the authors are unable to provide any novelty in the work. The work simply applies existing FL in agriculture without specifying any specific issue related to agriculture and resource utilization in agriculture.

Response:
Thank you for your gracious time and we sincerely appreciate your feedback and would like to clarify that our architecture is specifically tailored to address the unique challenges encountered in the agricultural sector with network variability and limited bandwidth in heterogenous environments, particularly in the context of combine tractors equipped with advanced sensors. A key innovation of our system is its ability to effectively manage Dynamic Network Conditions in Heterogenous Federated Learning Environments, which are common in rural farming environments. We recognize that inconsistent connectivity can impede timely data transmission and model updates, and our architecture incorporates adaptive strategies designed to ensure robust performance under such variable conditions.

Moreover, we also placed an emphasis on Resource Utilization Optimization by implementing mechanisms that assess and classify devices based on their computational power, energy efficiency, and latency. This approach allows for effective task allocation among edge devices, thereby enhancing overall system performance in resource-constrained agricultural contexts. Additionally, our commitment to Decentralized Data Processing ensures that sensitive agricultural data remains localized, significantly enhancing data privacy and minimizing the risks associated with centralized data management. Lastly, we explore the concept of Heterogeneous Federated Learning, accommodating the diverse capabilities of different devices, which allows for a more customized approach to model training that meets the specific needs of various agricultural applications. We believe these contributions represent a meaningful advancement over existing FL applications in agriculture and provide a solid foundation for future research in this critical area.

we have made revisions to both the motivation and abstract sections to clarify our scope and highlight the novelty of our work for the readers. We appreciate your guidance in helping us enhance the clarity of our manuscript.

 

2. The proposed work seems to be puzzling in terms of different issues like data privacy, data security, network-related issues, operational efficiency, and resource utilization. A very clear and concise distinction is required among them, and each should be defined clearly with the scope of work presented in the paper.

Response:

We sincerely appreciate your insight, as it allows us to enhance the overall presentation of our research. We recognize the importance of clearly distinguishing between the concepts of data privacy, data security, network-related issues, operational efficiency, and resource utilization, as each plays a critical role in the context of our heterogenous federated learning architecture for precision agriculture in dynamic network conditions.

In response to your valuable feedback, we have added a new section titled ‘Proposed Research Focused’ in our manuscript. This section provides a detailed explanation of each key term related to our work, ensuring that we clarify their meanings and significance. By doing so, we aim to eliminate any potential confusion among readers and enhance their understanding of the concepts discussed in the paper. We appreciate your guidance in helping us improve the clarity of our manuscript.

 

3. The work provides the literature review related to the proposed work. However, it is very hard to find any research gap or limitation that needs to be focused on this paper.

Response:

We have revised the relevant sections of the manuscript to ensure that these gaps are explicitly stated, allowing readers to better understand the significance of our research. We believe that these revisions will enhance the overall quality of our manuscript and more effectively communicate the contributions of our work.

In our related works section, we endeavored to provide a comprehensive overview of existing research surrounding the application of federated learning in agriculture, and we acknowledge that identifying specific research gaps and limitations is crucial for framing the context of our contributions. To enhance clarity, we have explicitly articulated these gaps in the revised version of the manuscript. Specifically, we identified the challenge of dynamic network conditions characteristic of rural agricultural environments, which is often overlooked in existing frameworks that do not adequately address the implications of unreliable connectivity on performance. Additionally, we noted a significant limitation in the lack of focus on heterogeneous federated learning, as many current studies tend to apply uniform models without considering the diverse capabilities of various devices used in agriculture. We have also highlighted the need for optimization of resource utilization and operational efficiency within federated learning systems, emphasizing that these areas are essential for practical applicability in real-world scenarios. Furthermore, we recognized that the implementation of checkpointing mechanisms is not widely discussed in the context of federated learning applied to agriculture, despite its potential to enhance robustness in environments with fluctuating network conditions. By addressing these research gaps and limitations, we aim to provide a clear rationale for our proposed work and its contributions to the field.

 

4. The proposed architecture is very ambiguous and not clearly explained. It requires a clear explanation for each of its components.

Response:
In response to your concerns, we have taken the initiative to enhance the manuscript by adding a dedicated subsection that clearly outlines the various components of the architecture. This section now offers detailed descriptions of each element, including their functions and interrelationships within the system. We believe this addition will significantly improve the clarity and understanding of our architectural framework.

 

5. The work is based on federated learning. However, there is lack of FL architecture and its working explanation.

Response:
In response to your valuable feedback regarding the explanation of the federated learning architecture, we have taken the initiative to add a new figure within the federated learning subsection of the system architecture. This figure provides a visual representation of the architecture, illustrating how each component is employed within our proposed system. Additionally, we have included a detailed explanation accompanying the figure to clarify the roles and interactions of these components, thereby enhancing the reader's understanding of their significance in the overall architecture. We believe that this addition will address your concerns effectively and improve the clarity of our presentation. Thank you for your constructive suggestions, which have helped us refine our work.

 

6. The proposed work talks about the clustering of the devices (integrated into tractors). However, it is very hard to understand why clustering is an important part of the proposed work. If this is the case, what is the basis of the clustering? Can’t the proposed work not implemented without clustering the devices?

Response:
Thank you for your thoughtful feedback regarding the significance of clustering within our proposed work. Clustering is a fundamental component of our approach as it directly enhances the efficiency and effectiveness of the federated learning process in heterogenous environments. By organizing the tractors into clusters, we enable localized processing and communication, which significantly reduces the overall communication load to the global server. This is particularly crucial in agricultural settings where connectivity may be intermittent or unreliable. Clustering allows us to leverage the computational resources of multiple tractors, enabling them to collaboratively train models while minimizing the risks associated with data transmission over potentially insecure networks.

While it is possible to implement the proposed work without clustering, such an approach would likely lead to increased communication overhead and potential bottlenecks in data transmission, especially in real-world scenarios characterized by limited bandwidth and variable connectivity. Clustering not only optimizes resource utilization but also ensures that the system remains resilient and responsive to the dynamic conditions often encountered in agricultural environments.

We have also made revisions to the manuscript to further emphasize the rationale behind the clustering approach and its critical role in enhancing the overall performance and reliability of the system.

7. The work requires a clear explanation of the alpha score and its significance in the proposed work.

Response:
We truly appreciate your insights, as they help us enhance the clarity and comprehensibility of our manuscript. In response to your comment, we have revised the relevant sections to provide a more comprehensive explanation of the Alpha Score, detailing its purpose in evaluating and distinguishing between the underlying datasets associated with different machine learning models. We have also emphasized its critical role in informing the aggregation process at the global server, particularly in the context of our heterogeneous environment.

By clarifying the significance of the Alpha Score, we aim to ensure that readers fully understand its importance in maintaining the accuracy and effectiveness of our federated learning approach. Thank you once again for your valuable suggestions, which have significantly contributed to improving our work.

 

8. The proposed algorithms are not very clear. There is a tremendous amount of requirement to clearly explain and write the pseudo-code of the proposed algorithms.

Response:
We greatly appreciate your thorough review and the opportunity to enhance our work. To address your concerns, we have revised the pseudo-code sections to ensure that each algorithm is presented with a clear and logical structure. Each step of the algorithm now includes descriptive comments that explain the purpose and function of each operation, guiding readers through the algorithm's logic more intuitively. Additionally, we have included detailed explanations for key components and calculations within each algorithm to clarify the rationale behind specific choices, such as the selection of algorithms (e.g., Salsa20 encryption) and the methods for data transmission based on network type.

 

9. The work claims for the enhancement of data security in the proposed work. However, it has not proposed any novel work in terms of data security, instead it has employed an existing data encryption method. Hence, if the work is supposed to claim for any novel data security methods, it should be clearly explained.

Response:
We appreciate your observation that our implementation of the Salsa20 encryption algorithm is based on an existing method. While it is true that Salsa20 is an established encryption technique, we believe that its application within the specific context of federated learning for smart agriculture brings forth unique advantages. By employing this encryption method, we prioritize the security of model weights rather than transmitting raw data, thereby significantly reducing the risk of exposing sensitive agricultural information during communication.

our approach also considers the dynamic nature of agricultural environments, where network conditions can vary greatly. This adaptability not only enhances the overall efficiency of data transmission but also provides a layer of security that is crucial for protecting sensitive information in less secure settings. We acknowledge that our work does not introduce a novel encryption algorithm; however, we aim to demonstrate how existing methods can be effectively utilized in innovative ways to enhance data security for agricultural applications.

In light of your comments, we will clarify in our manuscript that while we utilize an established encryption method, the context and implementation uniquely contribute to a more secure federated learning framework. Additionally, we will discuss potential future research directions, such as exploring homomorphic encryption or developing new security protocols tailored for our application, to further enhance data protection in this field. Your insights have been instrumental in guiding our revisions, and we are grateful for your input.

 

10. The paper requires significant work to review for the English language and grammar used in the paper.
Response:
Thank you for your feedback about our articulation and grammar. We have utilized English language support tools and conducted a thorough review using Grammarly to identify and correct any potential errors related to language and grammar in our manuscript. We are committed to ensuring the highest quality in our writing. If you have any suggestions for corrections or rewording that could enhance the clarity and effectiveness of our paper, please do not hesitate to let us know. Your feedback is invaluable to us, and we are eager to make any necessary adjustments. Thank you for your understanding and support.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript “Towards Secure and Efficient Farming using Federated Learning with Decentralized Aggregation Protocols in Dynamic Network Conditions" is very interesting . However, the manuscript has various shortcomings . Below are detailed comments and suggestions for improving your work.

Comment 1 . In lines 1-16. The summary should explain the differences between this study and previous studies of the same type.

Comment 2 . In lines 101-116 ,It may be more convincing to add corresponding references or charts in the motivation chapter.

Comment 3 .The picture/table name in the article is too long. It may be best to simply describe the content of the picture and add a detailed description to the text.

Comment 4 . In lines 1165-1167,The content and citation format of references may need to be confirmed.

Author Response

Response Letter for Reviewer Comments

Comment 1 . In lines 1-16. The summary should explain the differences between this study and previous studies of the same type.

Response:

Thank you for your valuable feedback regarding the summary of our abstract. We appreciate your suggestion to clarify the differences between our study and previous research in the same field. In response, we have made revisions to the abstract to explicitly reflect these distinctions and highlight how our approach differs from existing studies. We believe that these changes will enhance the clarity and impact of our work.

 

Comment 2 . In lines 101-116 ,It may be more convincing to add corresponding references or charts in the motivation chapter.

Response:

Thank you for your thoughtful suggestion regarding the inclusion of corresponding references or charts in the motivation chapter. In response to your recommendation, we have made the necessary changes in the manuscript to incorporate relevant references and visual data representations. We believe these additions will strengthen the motivation section and provide clearer context for our research.

 

Comment 3 .The picture/table name in the article is too long. It may be best to simply describe the content of the picture and add a detailed description to the text.

Response:

 In response to your review, we revised the titles to make them more concise while ensuring that the content is clearly described within the manuscript. Your insights will help enhance the clarity and readability of our work. Thank you!

 

Comment 4 . In lines 1165-1167,The content and citation format of references may need to be confirmed.

Response:

We really appreciate your attention to detail, as it is crucial for maintaining the integrity of our work. We have thoroughly reviewed and made the necessary corrections to ensure that all references are accurate and formatted correctly in accordance with the required citation style.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The comments are as follows for the article entitled “Towards Secure and Efficient Farming using Federated Learning with Decentralized Aggregation Protocols in Dynamic Network Conditions”.
1.    Mention how does the proposed federated learning framework perform under extreme and persistent network outages in rural or remote regions.
2.    Discuss about the scalable possibility of the proposed work when the number of edge nodes increases to hundreds or thousands.
3.    Include a clear block diagram to represent the workflow of the each and every module involved in the work.
4.    Indicate the betterment of the Salsa20 encryption scheme over the traditional AES encryption scheme. Discuss about the latency and security.
5.    Explain about the failure recovery mechanisms included in the work during node failure conditions at extreme situations.
6.    Specify the expected performance deviations of the proposed method with respect to changes in seasonal and soil variations.
7.    The threshold conditions for similarity and dissimilarity in checkpointing are not defined clearly. Explain them.
8.    Compare the betterment of the proposed work with the existing FL-based agricultural solutions to prove its efficiency.
9.    Conclude the paper by mentioning the future research work of the proposed method. Discuss about the possibility of integration with hybrid FL and renewable power resources.
10.    Include or replace few more references from 2024 and 2025 with proper citations.

Author Response

Response Letter for Reviewer Comments

 

1. Mention how does the proposed federated learning framework perform under extreme and persistent network outages in rural or remote regions.

Response:

Thank you for raising a very insightful question regarding the performance of the system under extreme and persistent network outages in rural or remote regions. We really appreciate your interest in this critical aspect of our research. In our framework, we have employed strategies to ensure resilience during such challenging conditions. For instance, edge devices, such as tractors, can continue to operate by locally storing sensor data and model updates, allowing for uninterrupted functioning even when connectivity is lost. When the network is restored, the framework uses a differential data transmission approach, sending only the changes from the last known state instead of the entire model, which optimizes bandwidth usage and minimizes the risk of transmission failure. we also incorporate a checkpointing mechanism that regularly saves the model's state, enabling devices to revert to the most recent stable version during outages. Proximity-based communication is also utilized, allowing devices to connect to nearby edge nodes for local data aggregation, which reduces latency and improves reliability.

We have detailed these mechanisms and their impacts in the discussion section of the paper, where we thoroughly explore how our system effectively addresses network challenges.

 

2.  Discuss about the scalable possibility of the proposed work when the number of edge nodes increases to hundreds or thousands.

Response:

Thank you for your thoughtful question regarding the scalability of our proposed framework, particularly in the context of increasing the number of edge nodes to hundreds or thousands. For a more detailed exploration of how our framework handles scalability as the number of edge nodes increases, we have included an in-depth discussion in the discussion section of the paper. Our system is designed with scalability as a fundamental principle. It employs a hierarchical architecture that facilitates the formation of clusters among edge devices. This clustering approach optimizes communication by enabling local aggregation of model updates, which significantly reduces the bandwidth required and enhances overall efficiency. As the number of edge nodes increases, the framework can dynamically adjust by prioritizing essential updates and compressing data for transmission, ensuring effective communication even in larger deployments.

 

3.  Include a clear block diagram to represent the workflow of each module involved in the work.

Response:

We have included a clear workflow diagram in the manuscript that represents modules involved in our work. We hope this addition enhances the clarity and understanding of our proposed system.

 

4. Indicate the betterment of the Salsa20 encryption scheme over the traditional AES encryption scheme. Discuss about the latency and security.

Response:

We appreciate your interest in this important aspect of our work. In our paper, we have provided a detailed discussion of betterment of Salsa20 over AES encryption algorithms in the "Secure Data Transmission" section. This section outlines how Salsa20's design as a stream cipher contributes to lower latency and faster encryption and decryption times, especially in resource-constrained environments such as edge computing. we also mentioned the security features of Salsa20, highlighting its effectiveness in maintaining data confidentiality and integrity.

 

5. Explain about the failure recovery mechanisms included in the work during node failure conditions at extreme situations.

Response:

Thank you asking an important question. We have provided a detailed explanation of these mechanisms in the discussion section of the paper, where we outline how our system addresses node failures through local data storage, checkpointing, proximity-based recovery, dynamic node reassignment, and remote monitoring. This comprehensive approach ensures operational continuity and resilience in challenging agricultural environments.

 

6.  Specify the expected performance deviations of the proposed method with respect to changes in seasonal and soil variations.

Response:

In our study, we acknowledge that seasonal changes and variations in soil characteristics can significantly influence agricultural outcomes. As such, we have included a thorough analysis of these factors in the discussion section of our paper. We anticipate that the performance of our federated learning framework may exhibit deviations based on the specific conditions of the environment, such as differing moisture levels, nutrient availability, and crop growth stages associated with various seasons. Our framework is designed to adapt to these variations through continuous learning and data integration from multiple edge devices. By leveraging real-time data collected from sensors, the system can dynamically adjust its model to account for changes in soil conditions and seasonal impacts, thereby minimizing performance deviations. We invite you to refer to the discussion section for a more in-depth exploration of how our method addresses these challenges and the strategies we implement to ensure robust performance across diverse agricultural scenarios.

 

7. The threshold conditions for similarity and dissimilarity in checkpointing are not defined clearly. Explain them.

Response:

In our proposed method, the threshold conditions for determining similarity and dissimilarity during checkpointing are based on the degree of change in model weights between successive iterations. Specifically, we define a similarity threshold that quantifies the acceptable level of change in the model parameters; if the change is below this threshold, the model is considered similar, and no new checkpoint is created. Conversely, if the change exceeds the similarity threshold, the model is deemed significantly different, warranting the creation of a new checkpoint. We utilize metrics such as cosine similarity and Euclidean distance between the model weight vectors to quantify the degree of similarity or dissimilarity. By establishing these thresholds, we can effectively manage storage and computational resources, ensuring that only meaningful model updates are retained while avoiding unnecessary redundancy.

We also have clarified these threshold conditions further in the revised manuscript to ensure that they are clearly defined and easily understood.

 

8.  Compare the betterment of the proposed work with the existing FL-based agricultural solutions to prove its efficiency.

Response:

We have provided a comprehensive presentation of the comparison metrics in the experimental section of our paper. This detailed analysis is designed to facilitate a clearer understanding of the performance improvements achieved by our proposed federated learning framework when compared to traditional existing architectures used in agricultural solutions. By systematically outlining these metrics, we aim to highlight the advancements our approach offers, thereby enabling readers to appreciate the significant enhancements in efficiency, accuracy, and overall effectiveness that our framework brings to the field of agriculture. We believe this comparison is essential for contextualizing our findings and demonstrating the practical implications of our work in improving agricultural practices.

 

9.  Conclude the paper by mentioning the future research work of the proposed method. Discuss about the possibility of integration with hybrid FL and renewable power resources.

Response:

In our concluding remarks, we highlight the exciting potential of our federated learning framework for agricultural applications. We anticipate numerous opportunities for future research that could greatly enhance the capabilities of our system. For a comprehensive overview of these possibilities, we encourage readers to refer to the detailed descriptions provided in the Future Work section of our manuscript.

 

10.  Include or replace few more references from 2024 and 2025 with proper citations.

Response:

We appreciate your input, as it helps enhance the relevance and depth of our work. We have taken your recommendation into account and have added several recent citations in the manuscript to ensure that our paper reflects the latest advancements in the field.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The paper was carefully revised following the suggestion of the referee.

New sections were added and several aspect clarified.

Check figure 7, labels and colors.

 

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have carried out all suitable corrections suggested by me and they have improved the paper well. Hence the paper shall be accepted in present form.

Back to TopTop