AgriSense AI: A Multimodal Prompt-Engineered Framework for Intelligent Soil Analysis, Disease Detection and Crop Recommendation

This study introduced AgriSense AI, a web-based intelligent agriculture framework developed to provide integrated support for soil assessment, plant disease identification, crop selection, and farm management. The proposed framework combines image-based analysis, pretrained multimodal artificial intelligence models, real-time weather information, structured prompt engineering, and multilingual advisory services within a single application. Unlike agriculture systems that depend on dedicated sensors and locally trained models, AgriSense AI allows users to access AI-assisted recommendations through an internet-connected device by uploading soil or plant images.   The evaluation confirmed that the developed system can generate structured agricultural responses and support the successful interaction of multiple AI and API-based modules. During the selected test evaluation, the soil analysis module obtained an observed category-level accuracy of 71.4%, while the plant disease detection module obtained 70.0%. In addition, all ten Soil Analysis API requests were completed successfully, resulting in a 100% API success rate. The required output information was available in every tested response, and the average response time was approximately 3.68 seconds. These findings demonstrate the functional reliability and practical feasibility of the proposed web-based framework under the evaluated test conditions.   The main contribution of AgriSense AI is the integration of multiple agricultural decision-support functions into an accessible and hardware-independent platform. The use of structured JSON outputs improves the consistency and usability of AI-generated information, while the integration of weather data supports more context-aware crop and farm management recommendations. Multilingual support further increases the potential usability of the system for users with different language preferences.   Although the obtained results demonstrate the feasibility of the proposed approach, the current evaluation was conducted using a limited number of test samples. Therefore, the reported category-level accuracies represent the observed performance of the integrated application and should not be considered general performance values for the underlying pretrained models. Future work can include testing with larger and more diverse real-field datasets, validating soil estimates using laboratory measurements, obtaining agricultural expert feedback, and improving the evaluation of crop, fertilizer, and disease recommendations. The conceptual federated learning component may also be extended into a fully implemented privacy-preserving learning framework. With these improvements, AgriSense AI can be further developed into a more reliable, scalable, and practically useful decision-support system for smart agriculture.

[1]     A. Morchid et al., “High-technology agriculture system to enhance food security: A concept of smart irrigation system using Internet of Things and cloud computing,” J. Saudi Soc. Agricult. Sci., 2024, doi: 10.1016/j.jssas.2024.02.001. [2]       A. Corallo, M. E. Latino, M. Menegoli, and F. Signore, “Digital technologies for sustainable development of agri-food: Implementation guidelines toward Industry 5.0,” IEEE Trans. Eng. Manag., vol. 71, pp. 10699–10715, 2024, doi: 10.1109/TEM.2024.3403251. [3]       R. Kumar Kasera, S. Gour, and T. Acharjee, “A comprehensive survey on IoT and AI based applications in different pre-harvest, during-harvest and post-harvest activities of smart agriculture,” Comput. Electron. Agricult., vol. 216, 2024, doi: 10.1016/j.compag.2023.108522. [4]     M. Aldossary, H. A. Alharbi, and C. Anwar Ul Hassan, “Internet of Things (IoT)-enabled machine learning models for efficient monitoring of smart agriculture,” IEEE Access, vol. 12, pp. 75718–75734, 2024, doi: 10.1109/ACCESS.2024.3404651. [5]       M. Hasan et al., “Ensemble machine learning-based recommendation system for effective prediction of suitable agricultural crop cultivation,” Front. Plant Sci., vol. 14, 2023, doi: 10.3389/fpls.2023.1234555. [6]     R. K. Ray, S. K. Das, and S. Chakravarty, “Smart crop recommender system—A machine learning approach,” in Proc. 12th Int. Conf. Cloud Comput., Data Sci. Eng. (Confluence), 2022, pp. 494–499, doi: 10.1109/Confluence52989.2022.9734173. [7]     B. Motamedi and B. Villányi, “A predictive analytics model with Bayesian-optimized ensemble decision trees for enhanced crop recommendation,” Decis. Anal. J., vol. 12, 2024, doi: 10.1016/j.dajour.2024.100516. [8]     M. R. Mahmood, M. A. Matin, S. K. Goudos, and G. Karagiannidis, “Machine learning for smart agriculture: A comprehensive survey,” IEEE Trans. Artif. Intell., vol. 5, no. 6, pp. 2568–2588, 2024, doi: 10.1109/TAI.2023.3345278. [9]       P. Garg et al., “Environmental and soil parameters for germination of leaf spot disease in the groundnut plant using IoT-enabled sensor system,” IEEE Sensors Lett., vol. 7, no. 12, 2023, doi: 10.1109/LSENS.2023.3330923. [10]  A. A. Khan, M. Faheem, R. N. Bashir, C. Wechtaisong, and M. Z. Abbas, “Internet of Things (IoT) assisted context aware fertilizer recommendation,” IEEE Access, vol. 10, pp. 129505–129519, 2022, doi: 10.1109/ACCESS.2022.3228160.