Applications, Challenges, and Prospects of Artificial Intelligence in Crop Production

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Although the manuscript addresses an important and timely topic, the current literature coverage is not sufficiently comprehensive for a review article. The authors state that the review is based on 40 core references and derived secondary references; however, the scope of the manuscript covers multiple broad areas, including biotic stress monitoring, soil health management, precision operation, supply chain optimization, climate-resilient agriculture, deep learning, sensor fusion, data-driven methods, and hybrid modeling. For such a broad review, the reference base should be substantially expanded to include more recent and representative studies. In particular, the authors may consider incorporating relevant recent works on image-based high-throughput plant phenotyping, transfer-learning-based pest image recognition, and large-scale crop disease and pest datasets, such as studies related to From sensors to insights: Technological trends in image-based high-throughput plant phenotyping, Pest image recognition algorithm based on joint adversarial transfer learning, and DLCPD-25: A Large-Scale and Diverse Dataset for Crop Disease and Pest Recognition. These works could help strengthen the discussion on visual perception, dataset diversity, model generalization, and practical AI applications in crop production.

The manuscript presents a broad narrative overview, but the review methodology is not sufficiently transparent. The authors should clearly explain how the literature was searched, screened, categorized, and synthesized. For example, the manuscript should specify the databases used, search keywords, inclusion and exclusion criteria, publication time range, and how the final set of studies was selected. Without a clear methodological description, it is difficult for readers to judge the completeness, objectivity, and reproducibility of the review. Adding a search strategy diagram or a PRISMA-style flowchart would greatly improve the rigor of the manuscript.

The manuscript summarizes many application scenarios of AI in agriculture, but much of the discussion remains descriptive. The authors should move beyond listing previous studies and provide deeper synthesis. For example, the review should compare the strengths and limitations of different AI techniques across tasks, discuss why certain models perform better in specific agricultural scenarios, and identify unresolved bottlenecks such as domain shift, annotation cost, small-object detection, data imbalance, interpretability, and edge deployment. More comparative tables summarizing model types, data sources, application scenarios, performance indicators, and limitations would help readers better understand the current state of the field.

There is an obvious inconsistency in the formatting of section titles. The title “1. Introduction” uses normal title capitalization, whereas subsequent major section headings, such as “2. CORE APPLICATION SCENARIOS OF ARTIFICIAL INTELLIGENCE IN AGRICULTURE,” are written in all capital letters. Similar capitalization inconsistencies appear in other main sections. The authors should carefully standardize all section and subsection headings according to the journal style. This issue may seem minor, but it affects the professionalism and readability of the manuscript.

The future prospects section covers several important directions, including advanced AI models, multimodal data fusion, lightweight deployment, interpretability, technological inclusion, and sustainable development. However, many statements remain broad and general. The authors should provide more concrete research gaps and actionable future directions. For example, the manuscript could discuss how foundation models, large-scale agricultural datasets, multimodal phenotyping platforms, domain adaptation, and lightweight edge-AI systems can be practically integrated into crop production. The discussion would be stronger if the authors could distinguish short-term technical improvements from long-term research opportunities and implementation challenges.

Author Response

1. Although the manuscript addresses an important and timely topic, the current literature coverage is not sufficiently comprehensive for a review article.

Responses: We greatly appreciate this constructive comment. The reference list has been substantially expanded and updated from the original 40 to 68 peer-reviewed high-quality publications. The newly included literature covers state-of-the-art advances in AI-driven crop phenotyping, biotic stress detection, multi-sensor fusion, UAV-based monitoring, and lightweight edge AI applications. Key landmark studies, including Wang et al. (2026), Maraveas (2024), Zhang et al. (2025), Yu et al. (2025), Liu et al. (2024), and Chen et al. (2024), have been fully integrated into the text. The updated literature foundation now provides comprehensive and up-to-date coverage of the field, significantly enhancing the depth and reliability of the review.

2. The authors state that the review is based on 40 core references and derived secondary references; however, the scope of the manuscript covers multiple broad areas, including biotic stress monitoring, soil health management, precision operation, supply chain optimization, climate-resilient agriculture, deep learning, sensor fusion, data-driven methods, and hybrid modeling.

For such a broad review, the reference base should be substantially expanded to include more recent and representative studies. In particular, the authors may consider incorporating relevant recent works on image-based high-throughput plant phenotyping, transfer-learning-based pest image recognition, and large-scale crop disease and pest datasets, such as studies related to From sensors to insights: 1) Technological trends in image-based high-throughput plant phenotyping, Pest image recognition algorithm based on joint adversarial transfer learning. 2) DLCPD-25: A Large-Scale and Diverse Dataset for Crop Disease and Pest Recognition.

These works could help strengthen the discussion on visual perception, dataset diversity, model generalization, and practical AI applications in crop production.

Response: We thank the reviewer for pointing out this critical issue. A dedicated subsection titled “MATERIAL AND METHOD” has been added to the manuscript. This section comprehensively describes the systematic search strategy, including databases (Google Scholar, ScienceDirect), search keywords, inclusion/exclusion criteria, and publication time frame (2018–May 2026). In addition, the below studies were added into the resubmitted version. We also added more than 20 related papers. We believe these additions strengthen the review’s authority and timeliness.

• Maraveas, C. Technological trends in image-based high-throughput plant phenotyping. AgriEngineering 2024, 6, 3375–3407.https://doi.org/10.3390/agriengineering6030193
• Zhang, H.W.; Wang, R.F.; Wang, Z.; Su, W.H. DLCPD-25: A Large-Scale and Diverse Dataset for Crop Disease and Pest Recognition. Sensors 2025, 25, 7098. https://doi.org/10.3390/s25227098

3. The manuscript presents a broad narrative overview, but the review methodology is not sufficiently transparent. The authors should clearly explain how the literature was searched, screened, categorized, and synthesized. For example, the manuscript should specify the databases used, search keywords, inclusion and exclusion criteria, publication time range, and how the final set of studies was selected. Without a clear methodological description, it is difficult for readers to judge the completeness, objectivity, and reproducibility of the review. Adding a search strategy diagram or a PRISMA-style flowchart would greatly improve the rigor of the manuscript.

Response: A standard PRISMA 2020 flowchart is now included as Figure 2 in the supplementary materials, clearly illustrating the literature identification, screening, and eligibility assessment process. The total number of identified, excluded, and included records is explicitly stated (196 initially identified, final 69 included). We hope this revision fully addresses the transparency and reproducibility requirements of systematic reviews.

4. The manuscript summarizes many application scenarios of AI in agriculture, but much of the discussion remains descriptive. The authors should move beyond listing previous studies and provide deeper synthesis. For example, the review should compare the strengths and limitations of different AI techniques across tasks, discuss why certain models perform better in specific agricultural scenarios, and identify unresolved bottlenecks such as domain shift, annotation cost, small-object detection, data imbalance, interpretability, and edge deployment. More comparative tables summarizing model types, data sources, application scenarios, performance indicators, and limitations would help readers better understand the current state of the field.

Response:Thank you for your valuable and constructive comment. We have thoroughly addressed your concerns by both supplementing comparative tables and deepening the analytical discussion based on the original content of our manuscript. As suggested, we have added Table 1in Section 3. This table systematically summarizes the strengths, limitations, application scenarios, and representative references of mainstream AI techniques (CNN, YOLO, ViT, transfer learning, hybrid models, self-supervised learning) in crop biotic stress monitoring and phenotyping. It directly addresses the lack of cross-technique comparison and helps readers clearly understand the technical differences and scenario adaptability of different models.

5. There is an obvious inconsistency in the formatting of section titles. The title “1. Introduction” uses normal title capitalization, whereas subsequent major section headings, such as “2. CORE APPLICATION SCENARIOS OF ARTIFICIAL INTELLIGENCE IN AGRICULTURE,” are written in all capital letters. Similar capitalization inconsistencies appear in other main sections. The authors should carefully standardize all section and subsection headings according to the journal style. This issue may seem minor, but it affects the professionalism and readability of the manuscript.

Response: We apologize for this inconsistency. All section and subsection headings have been fully standardized to journal-style sentence case, and inconsistent uppercase or mixed-case formatting has been corrected throughout the manuscript. The formatting is now consistent and compliant with typical agricultural and computational journal requirements.

6. The future prospects section covers several important directions, including advanced AI models, multimodal data fusion, lightweight deployment, interpretability, technological inclusion, and sustainable development. However, many statements remain broad and general. The authors should provide more concrete research gaps and actionable future directions. For example, the manuscript could discuss how foundation models, large-scale agricultural datasets, multimodal phenotyping platforms, domain adaptation, and lightweight edge-AI systems can be practically integrated into crop production. The discussion would be stronger if the authors could distinguish short-term technical improvements from long-term research opportunities and implementation challenges.

Response: Thank you for this valuable and constructive comment. Thank you for your constructive and detailed feedback. We have completely rewritten the entire Section 5 (Future Prospects). The revised section fully retains and integrates all original content from the manuscript, while significantly enhancing the depth and specificity of the discussion.

We explicitly address your suggestions by elaborating on the practical integration of foundation models, large-scale agricultural datasets, multimodal phenotyping platforms, domain adaptation, and lightweight edge-AI systems into crop production. Additionally, we clearly distinguish short-term technical improvements (1–2 years) from long-term research opportunities (3–5 years) and further discuss key implementation challenges, making the future directions concrete, actionable, and research-oriented.

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript claims that it is “based on 40 core literatures and derived secondary references,” but it does not provide databases searched and inclusion/exclusion criteria. Without a PRISMA-style workflow or any reproducible selection protocol, the paper reads like a broad narrative essay rather than a systematic review.

The paper lists many deep learning models, but it mostly describes where they are used rather than explaining why particular architecture works, what data conditions they require, or where they fail.

Summary tables are needed to compare datasets, crop types, sensor modalities, model architecture etc.

Important evaluation issues are missing or only briefly mentioned.

The discussion of LLMs, interpretability, and policy is directionally reasonable, but it stays at a high level and does not identify concrete research gaps.

Highly recommend citing the following paper:

A Field-Adaptive Mechanical Weeding System Coupling Oscillating Pneumatic Mechanism With Deep Learning for Intra-Row Weed Control in Lettuce

A review of visual navigation for agricultural robots in open fields and controlled environments

Author Response

1.The manuscript claims that it is “based on 40 core literatures and derived secondary references,” but it does not provide databases searched and inclusion/exclusion criteria. Without a PRISMA-style workflow or any reproducible selection protocol, the paper reads like a broad narrative essay rather than a systematic review.

Response: We thank the reviewer for pointing out this critical issue. A dedicated subsection titled “MATERIAL AND METHOD” has been added to the manuscript. This section comprehensively describes the systematic search strategy, including databases (Google Scholar, ScienceDirect), search keywords, inclusion/exclusion criteria, and publication time frame (2018–May 2026). In addition, the below studies were added into the resubmitted version.

A standard PRISMA 2020 flowchart is now included as Figure 2 in the supplementary materials, clearly illustrating the literature identification, screening, and eligibility assessment process. The total number of identified, excluded, and included records is explicitly stated (196 initially identified, final 69 included). We hope this revision fully addresses the transparency and reproducibility requirements of systematic reviews.

2.The paper lists many deep learning models, but it mostly describes where they are used rather than explaining why particular architecture works, what data conditions they require, or where they fail. Summary tables are needed to compare datasets, crop types, sensor modalities, model architecture etc. Important evaluation issues are missing or only briefly mentioned.

Response: Thank you for your valuable and constructive comment. We have thoroughly addressed your concerns by both supplementing comparative tables and deepening the analytical discussion based on the original content of our manuscript. As suggested, we have added Table 1in Section 3. This table systematically summarizes the strengths, limitations, application scenarios, and representative references of mainstream AI techniques (CNN, YOLO, ViT, transfer learning, hybrid models, self-supervised learning) in crop biotic stress monitoring and phenotyping. It directly addresses the lack of cross-technique comparison and helps readers clearly understand the technical differences and scenario adaptability of different models.

3.The discussion of LLMs, interpretability, and policy is directionally reasonable, but it stays at a high level and does not identify concrete research gaps.

Response: Thank you for your comment. We have expanded and detailed the discussion of LLMs, interpretability, and policy in Section 4.2, identifying concrete research gaps instead of general statements. The revised content covers crop-specific LLMs, multimodal integration, interpretability frameworks, and agricultural AI policy standards, making the analysis specific and actionable.

4.Highly recommend citing the following paper:

A Field-Adaptive Mechanical Weeding System Coupling Oscillating Pneumatic Mechanism With Deep Learning for Intra-Row Weed Control in Lettuce

A review of visual navigation for agricultural robots in open fields and controlled environments

Response: All recommended key papers have been fully cited, discussed, and integrated into the relevant sections. In addition we also added more than 20 related papers. We believe these additions strengthen the review’s authority and timeliness.

• Wang, R.F, Zhao, C.T, Tu, Y.H, Chen, Z.Q, Su, W.H. A Field-Adaptive Mechanical Weeding System Coupling Oscillating Pneumatic Mechanism With Deep Learning for Intra-Row Weed Control in Lettuce. Journal of Field Robotics. 2026, Volume 278, 121620. https://doi.org/10.1002/rob.70230
• Wang, R.F.; Xu, R.; Chee, P.W.; Wang, H.; Li, C. A review of visual navigation for agricultural robots in open fields and controlled environments. Comput. Electron. Agric. 2026, 248, 111754. https://doi.org/10.1016/j.compag.2026.111754.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Good job! The authors successfully addressed all my questions.

Reviewer 2 Report

Comments and Suggestions for Authors

The author answered all my questions. It's good to be published.