AgroSense: An Integrated Deep Learning System for Crop Recommendation via Soil Image Analysis and Nutrient Profiling
Vishal Pandey, Ranjita Das, Debasmita Biswas · 2025
Read AgroSense: An Integrated Deep Learning System for Crop Recommendation via Soil Image Analysis and Nutrient Profiling by Vishal Pandey, Ranjita Das, Debasmita Biswas (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-01344.txt.
Introduction
Choosing the right crop for a field usually needs slow lab tests of soil, which delays planting choices. Vishal Pandey, Ranjita Das, and Debasmita Biswas present AgroSense, a deep learning system that recommends crops from soil photos plus nutrient data, in real time. It aims to support food security and more sustainable farming.
One module classifies soil type from images, comparing three vision models. A second module reads structured data such as nutrient levels, soil acidity, and rainfall, testing four learning methods. The team built a combined data set of ten thousand paired samples, drawn from public sources with many more images and profiles. Their fused model gave accurate, context-aware recommendations in tests, pointing toward field-ready tools that cut the wait for soil results.
Catalog record
| CREATOR | Vishal Pandey, Ranjita Das, Debasmita Biswas |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 53 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-01344
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
AgroSense: An Integrated Deep Learning System for Crop Recommendation via Soil Image Analysis and Nutrient Profiling Vishal Pandey, Ranjita Das, Debasmita Biswas Abstract Meeting the increasing global demand for food security and sustainable farming requires intelligent crop recommendation systems that operate in real time. Traditional soil analysis techniques are often slow, labor-intensive, and not suitable for on-field decision-making. To address these limitations, we introduce AgroSense , a deep-learning framework that integrates soil image classification and nutrient profiling to produce accurate and contextually relevant crop recommendations. AgroSense comprises two main components: a Soil Classification Module, which leverages ResNet-18, EfficientNet-B0, and Vision Transformer architectures to categorize soil types from images; and a Crop Recommendation Module, which employs a Multi-Layer Perceptron, XGBoost, LightGBM, and TabNet to analyze structured soil data including nutrient levels, pH, and rainfall. We curated a multimodal dataset of 10,000 paired samples drawn from publicly available Kaggle repositories—approximately 50,000 soil images across seven classes and 25,000 nutrient profiles—for experimental evaluation. The fused model achieves 98.0 % accuracy, with precision 97.8 %, recall 97.7 %, and F1-score 96.75 %, while RMSE and MAE drop to 0.32 and 0.27, respectively. Ablation studies underscore the critical role of multimodal coupling, and statistical validation via t-tests and ANOVA confirms the significance of our improvements. AgroSense offers a practical, scalable solution for real-time decision support in precision agriculture and paves the way for future lightweight multimodal AI systems in resource-constrained environments. keywords Deep Learning; Multimodal Data Fusion; Soil Classification; Crop Recommendation; Computer Vision; Precision Agriculture 1 Introduction Modern agriculture faces the dual challenge of increasing crop productivity while maintaining environmental sustainability. Central to this effort is the accurate analysis of soil, as key properties such as pH, nutrient levels, and texture significantly influence plant growth and yield outcomes. However, conventional soil testing methods are often costly, time-consuming, and largely confined to laboratory environments—making them impractical for real-time, in-field decision-making motwani2022soil ; manjula2022efficient . With the evolution of precision agriculture, there has been a surge of interest in applying artificial intelligence (AI), particularly machine learning and deep learning, to enhance soil assessment. Recent advances demonstrate that models like Convolutional Neural Networks (CNNs) and Transformers are capable of extracting valuable information from soil images to support crop prediction tasks rajak2017crop . Despite these advancements, many existing systems continue to treat soil texture analysis and nutrient profiling as independent processes. This fragmented approach limits the effectiveness of predictions, as it fails to capture the nuanced relationship between a soil’s physical structure and its chemical composition both of which are vital for making informed crop recommendations. Most current crop recommendation systems operate on this disconnected paradigm, treating visual and chemical characteristics of soil separately. While deep learning has advanced the analysis of soil imagery and machine learning has proven effective for nutrient-based prediction, few studies have successfully combined the two into a cohesive framework kulkarni2018improving ; rajak2017crop . This separation leads to incomplete soil characterization and less reliable recommendations. To address this gap, AgroSense integrates image-based soil classification with structured nutrient data, enabling a more holistic, real-time, and context-aware approach to crop recommendation in the field of precision agriculture. Many existing crop recommendation systems continue to treat soil texture and nutrient composition as independent variables, which often leads to fragmented decision-making and less reliable crop suggestions. This study seeks to address that limitation through the introduction of AgroSense an integrated deep learning framework that simultaneously considers both visual and chemical properties of soil to enhance accuracy and enable real-time recommendations within a precision agriculture context. AgroSense is designed as a comprehensive solution that fuses image-based soil classification with nutrient profiling to produce data-driven crop recommendations. It employs advanced deep learning architectures, including a custom Convolutional Neural Network (CNN), ResNet18, EfficientNet-B0, and Vision Transformer, to extract detailed visual features from soil images. These visual insights are then combined with structured soil data comprising attributes such as pH, nitrogen, phosphorus, potassium, humidity, temperature, and rainfall and fed into machine learning models such as Multi-Layer Perceptron (MLP), XGBoost, LightGBM, and TabNet to predict suitable crops. Empirical evaluations indicate that this multimodal approach achieves a test accuracy of 98.0%, significantly outperforming traditional methods based on tabular data alone. The framework was developed using a combination of publicly available datasets. For soil classification, two Kaggle image datasets were utilized: one containing over 1,400 images spanning four soil types Alluvial, Black, Clay, and Red and another featuring five additional types including Cinnamon, Laterite, Peat, and Yellow. For the crop recommendation module, three datasets were used: the Crop Recommendation Dataset, the Crop Recommendation Using Soil Properties and Weather Data Dataset, and a third dataset containing nutrient profiles and associated crop labels. Together, these resources provide a diverse and representative foundation, facilitating robust model training and generalization across varied soil and environmental conditions. In essence, AgroSense brings together the strengths of modern computer vision and tabular learning into a unified platform tailored for real-world agricultural applications. By leveraging multimodal data fusion and deep learning, the framework sets the stage for a new level of accuracy and practicality in smart farming systems. 2 Literature Review Recent advancements in deep learning have made notable contributions to the development of soil classification and crop recommendation systems. Numerous studies have employed machine learning and neural network techniques to support agricultural decision-making, leading to greater efficiency and sustainability in farming practices. However, the majority of these systems tend to concentrate on only one aspect of the problem. Some focus solely on analyzing soil texture using image-based classification, while others rely entirely on soil nutrient levels derived from tabular data. The lack of a unified approach that combines both visual and numerical information has limited the accuracy and broader applicability of these models in real-world agricultural settings. Motwani et al. motwani2022soil proposed a machine learning-based system for soil analysis and crop recommendation using only tabular data. While their work demonstrated the potential of artificial intelligence in agriculture, it did not account for soil texture, a crucial factor influencing crop selection. In another study, Manjula and Djodiltachoumy manjula2022efficient utilized deep learning techniques to predict crop suitability based on soil nutrient data. However, their model lacked a mechanism for analyzing soil imagery, which is essential for accurately identifying soil type and structure. Sunandini et al. sunandini2023smart explored an IoT-driven system that used sensors for soil monitoring in conjunction with machine learning algorithms. Despite the system’s innovation, it relied primarily on sensor data and did not incorporate deep learning for image-based soil classification, thereby limiting its generalizability across varied soil conditions. Broadly, research in soil classification and crop recommendation falls into three major categories: machine learning-based crop recommendation systems, deep learning methods for soil analysis, and hybrid models that integrate multiple data sources. Traditional machine learning approaches typically use structured data to recommend crops, but they often overlook soil texture, which restricts the models’ ability to fully capture field conditions. On the other hand, deep learning models have shown strong capabilities in classifying soil textures through convolutional neural networks (CNNs), yet they frequently ignore nutrient profiling. Hybrid approaches, including those that employ IoT technologies, attempt to unify various data inputs, but they often fail to utilize deep learning architectures effectively resulting in limitations in scalability and