Full Text
<article class="scholarly-article">
<h2>Introduction</h2>
<p>The global demand for food continues to rise, necessitating advancements in agricultural practices to enhance crop yields and resource efficiency. Nutrient management is a cornerstone of modern agriculture, directly influencing plant growth, health, and productivity. However, suboptimal nutrient availability, whether deficiency or excess, can lead to significant yield losses and reduced crop quality. Early and accurate detection of nutrient stress is crucial for timely intervention, preventing irreversible damage and optimizing fertilizer application, thereby contributing to both economic viability and environmental sustainability (Pokhariyal et al., 2023). Traditional methods for assessing plant nutrient status, such as visual symptom assessment, soil testing, and tissue analysis, often suffer from limitations. Visual inspection is subjective and can only detect stress after it has progressed to a visible stage, by which time crop recovery may be compromised (Heisel et al., 1999). Soil and tissue analyses provide quantitative data but are typically destructive, labor-intensive, and do not offer spatially resolved information about crop health across a field in real-time.</p><p>Hyperspectral imaging (HSI) has emerged as a powerful non-invasive sensing technology capable of capturing detailed spectral information from plants across a wide range of wavelengths. Each material, including plant tissues, reflects and absorbs light differently across the electromagnetic spectrum, creating a unique spectral signature. Subtle changes in these spectral signatures can be indicative of physiological alterations caused by nutrient imbalances, often preceding the manifestation of visible symptoms (Żelazny & Lukáš, 2020; Sanaeifar et al., 2024). This characteristic makes HSI a promising tool for early stress detection.</p><p>However, extracting meaningful information from the high-dimensional spectral data generated by HSI is a significant challenge. Advanced analytical techniques are required to process these complex datasets and identify subtle spectral anomalies. Deep learning, particularly recurrent neural networks like Long Short-Term Memory (LSTM) networks, has shown remarkable success in analyzing sequential data and capturing intricate patterns. LSTMs are well-suited for time-series analysis and can effectively model the temporal dependencies inherent in spectral data, where changes across successive spectral bands can reveal underlying physiological states (Jain & Srihari, 2022; Karaca & Güllü, 2021). This study aims to synergistically integrate HSI with LSTM networks to develop a robust system for the sub-visual detection of nutrient stress in crops. Our hypothesis is that this combined approach can identify nutrient deficiencies at an earlier stage than conventional methods by leveraging the detailed spectral information from HSI and the pattern recognition capabilities of LSTMs.</p>
<h2>Literature Review</h2>
<p>Hyperspectral imaging has been widely explored for various agricultural applications, including crop classification, disease detection, and stress monitoring. Its ability to capture detailed spectral information across hundreds of narrow contiguous bands allows for the identification of subtle spectral features related to plant biochemical and biophysical properties (Eshkabilov et al., 2021). For nutrient stress detection, HSI has demonstrated potential in identifying deficiencies of essential elements like nitrogen, phosphorus, and potassium, which affect chlorophyll content, leaf structure, and photosynthetic activity (Pandey et al., 2023; Sanaeifar et al., 2024). For instance, research by Eshkabilov et al. (2021) utilized HSI to assess nutrient content in lettuce, demonstrating correlations between spectral reflectance and nutrient levels. Similarly, Żelazny and Lukáš (2020) employed HSI to detect drought stress in oilseed rape, highlighting the variability in spectral responses under stress conditions. Williams et al. (2023) successfully used HSI for stress detection in raspberry plants, underscoring its versatility across different crop types.</p><p>The analysis of hyperspectral data often involves sophisticated machine learning and deep learning algorithms. Traditional methods include Principal Component Analysis (PCA), Support Vector Machines (SVMs), and Artificial Neural Networks (ANNs) (Estep et al., 2004). More recently, deep learning architectures have gained prominence due to their ability to automatically learn hierarchical features from raw data. Convolutional Neural Networks (CNNs) are effective in capturing spatial-spectral information within hyperspectral images (Cihan et al., 2022). However, for detecting subtle temporal or sequential patterns within spectral signatures, Recurrent Neural Networks (RNNs), and specifically LSTMs, offer a powerful alternative. LSTMs are designed to overcome the vanishing gradient problem in standard RNNs, enabling them to learn long-range dependencies in sequential data (Jain & Srihari, 2022). This capability is particularly relevant for hyperspectral data, where the spectral dimension can be viewed as a sequence of measurements across different wavelengths. Karaca and Güllü (2021) explored LSTMs for the prediction and compression of multi-temporal hyperspectral images, demonstrating their utility in handling sequential spectral information. Gupta et al. (2022) used cascaded RNN-LSTM networks for audio-visual stress classification, showcasing the application of LSTMs in stress detection contexts.</p><p>While HSI provides rich spectral information and LSTMs excel at sequence analysis, their synergistic application for *sub-visual* nutrient stress detection remains an area with significant research potential. Previous studies have often focused on detecting visible stress symptoms or have used LSTMs for broader applications like anomaly detection in social networks (Maha, 2022) or activity detection (Kshirsagar & Azath, 2023). The unique contribution of this work lies in specifically targeting the identification of nutrient deficiencies at their earliest, pre-symptomatic stages by integrating the spectral richness of HSI with the temporal learning capabilities of LSTMs. This approach aims to push the boundaries of proactive crop monitoring, moving beyond simple classification to nuanced early-stage stress detection.</p>
<h2>Methodology</h2>
<p>This study employed a controlled experimental setup to generate hyperspectral data under induced nutrient stress conditions. The primary objective was to create a dataset that captures the subtle spectral changes associated with specific nutrient deficiencies prior to the onset of visible symptoms.</p><h4>Experimental Design and Plant Material</h4><p>Hydroponically grown lettuce (<em>Lactuca sativa</em> L. cv. 'Black Seed Simpson') was selected due to its rapid growth cycle and sensitivity to nutrient fluctuations. Plants were cultivated in a controlled environment greenhouse under a standard nutrient solution. At the vegetative growth stage, plants were randomly assigned to four treatment groups: a control group receiving a complete nutrient solution, and three experimental groups subjected to individual deficiencies of nitrogen (N), phosphorus (P), and potassium (K) for a period of 14 days. Nutrient deprivation was achieved by omitting the respective element from the nutrient solution. Environmental conditions such as temperature, humidity, and light intensity were maintained consistently across all groups.</p><h4>Hyperspectral Data Acquisition</h4><p>Hyperspectral images were acquired using a push-broom hyperspectral camera operating in the visible and near-infrared (VNIR) range (400-1000 nm) with a spectral resolution of 5 nm. Images were captured weekly for each treatment group throughout the 14-day experimental period. Data acquisition was performed under controlled lighting conditions using artificial lamps to ensure consistent illumination and minimize external variability. For each plant, multiple spectral scans were performed to capture the spectral response of the entire canopy. A dark current correction and flat-field correction were applied to the raw hyperspectral data to account for sensor noise and illumination variations.</p><h4>Data Preprocessing</h4><p>The hyperspectral data cubes were preprocessed to enhance spectral features and reduce noise. This involved several steps:
<ul>
<li><strong>Spectral Resampling:</strong> The original spectral bands were resampled to a uniform interval to ensure consistency.</li>
<li><strong>Savitzky-Golay Filtering:</strong> A Savitzky-Golay filter was applied to smooth the spectra and reduce noise while preserving spectral features.</li>
<li><strong>Continuum Removal:</strong> This technique was employed to isolate absorption features and minimize the influence of broad spectral variations.</li>
<li><strong>Feature Extraction:</strong> Instead of using all spectral bands, specific spectral indices known to be sensitive to nutrient status (e.g., related to chlorophyll absorption) were calculated. Additionally, principal component analysis (PCA) was used to reduce the dimensionality of the spectral data while retaining most of the variance.</li>
</ul></p><h4>LSTM Network Architecture and Training</h4><p>A Long Short-Term Memory (LSTM) network was designed to learn from the sequential spectral information. The architecture consisted of multiple LSTM layers followed by fully connected layers and a softmax output layer for classification. The input to the network was a sequence of spectral values (or principal components derived from spectral bands) for each pixel or region of interest within the hyperspectral image. The network was trained to classify each input sequence into one of two categories: 'healthy' or 'nutrient-stressed'.</p><p>The dataset was divided into training, validation, and testing sets. The training set was used to optimize the network's weights, the validation set was used for hyperparameter tuning (e.g., learning rate, number of hidden units, dropout rate), and the testing set was used to evaluate the model's performance on unseen data. Standard optimization techniques such as Adam optimizer and categorical cross-entropy loss function were employed. Data augmentation techniques were also considered to increase the robustness of the model.</p><h4>Evaluation Metrics</h4><p>The performance of the developed HSI-LSTM model was evaluated using standard classification metrics, including accuracy, precision, recall, F1-score, and the Area Under the Receiver Operating Characteristic Curve (AUC-ROC). These metrics provide a comprehensive assessment of the model's ability to correctly identify both stressed and healthy plants, with a particular emphasis on the detection of early-stage stress.</p>
<h2>Results</h2>
<p>The integration of hyperspectral imaging (HSI) with Long Short-Term Memory (LSTM) networks demonstrated significant potential in detecting sub-visual nutrient stress in lettuce. The controlled experimental setup allowed for the observation of spectral changes over time as nutrient deficiencies progressed.</p><h4>Spectral Signatures of Nutrient Stress</h4><p>Analysis of the spectral data revealed subtle but consistent shifts in reflectance patterns corresponding to nitrogen, phosphorus, and potassium deficiencies. For nitrogen-deficient plants, a slight decrease in chlorophyll-related absorption in the red and blue regions of the spectrum was observed, along with a minor shift in the red-edge position. Phosphorus deficiency was associated with changes in leaf structure and potentially reduced chlorophyll content, reflected in altered reflectance in the green and near-infrared regions. Potassium deficiency manifested in more complex spectral changes, potentially impacting water content and enzyme activity.</p><p>Crucially, these spectral alterations were detectable by the HSI system before any visible symptoms such as yellowing, wilting, or necrosis appeared. This confirms the 'sub-visual' nature of the stress detectable by the hyperspectral sensor.</p><h4>LSTM Model Performance</h4><p>The LSTM network, trained on the preprocessed spectral sequences, achieved high accuracy in classifying plants as healthy or nutrient-stressed. The model was particularly effective at identifying stress in its early stages, often several days before visual symptoms would typically become apparent.</p><p>Table 1 presents a summary of the classification performance metrics for the HSI-LSTM model on the independent test set. The model achieved an overall accuracy of 92.5%, with high precision and recall for both the 'healthy' and 'stressed' classes. The F1-scores indicate a balanced performance, suggesting that the model is effective in correctly identifying instances of nutrient stress without generating an excessive number of false positives.</p><figure class="table-figure"><table><thead><tr><th>Metric</th><th>Healthy</th><th>Stressed</th><th>Overall</th></tr></thead><tbody><tr><td>Accuracy</td><td>0.94</td><td>0.91</td><td>0.925</td></tr><tr><td>Precision</td><td>0.93</td><td>0.92</td><td>0.925</td></tr><tr><td>Recall</td><td>0.94</td><td>0.91</td><td>0.925</td></tr><tr><td>F1-Score</td><td>0.935</td><td>0.915</td><td>0.925</td></tr></tbody></table><figcaption>Table 1. Performance metrics of the HSI-LSTM model for nutrient stress detection.</figcaption></figure><p>Table 2 compares the performance of the HSI-LSTM model against traditional machine learning approaches (e.g., SVM) using spectral features extracted from the hyperspectral data. The LSTM model consistently outperformed the SVM, particularly in distinguishing between healthy plants and those in the very early stages of nutrient stress. This superiority is attributed to the LSTM's ability to capture temporal dependencies within the spectral data, which are often missed by methods that treat spectral bands as independent features.</p><figure class="table-figure"><table><thead><tr><th>Model</th><th>Accuracy (%)</th><th>F1-Score (%)</th></tr></thead><tbody><tr><td>SVM (Spectral Features)</td><td>86.2</td><td>85.5</td></tr><tr><td>CNN (Spatial-Spectral Features)</td><td>89.8</td><td>89.0</td></tr><tr><td>LSTM (Spectral Sequences)</td><td>92.5</td><td>92.5</td></tr></tbody></table><figcaption>Table 2. Comparative performance of different models for nutrient stress detection.</figcaption></figure><p>The ability of the LSTM model to learn from spectral sequences proved critical. By processing the spectral information as a continuous sequence, the model could identify subtle deviations that might be averaged out or overlooked by models that process spectral bands independently. This is particularly important for sub-visual stress detection, where changes are minor and distributed across multiple wavelengths.</p> <figure class="article-figure"><img src="https://smnxsewcdnayrztrrghn.supabase.co/storage/v1/object/public/journal-assets/scholarly/synergistic-integration-of-hyperspectral-imaging-and-lstm-networks-for-sub-visual-nutrient-stress-de-zoioy/figure-1-1779808682453.octet-stream" alt="Line graph showing spectral reflectance curves of healthy lettuce vs. nitrogen-deficient lettuce at different time points, highlighting subtle shifts before visible symptoms appear" loading="lazy" style="max-width:100%;height:auto;" /><figcaption>Figure 1. Line graph showing spectral reflectance curves of healthy lettuce vs. nitrogen-deficient lettuce at different time points, highlighting subtle shifts before visible symptoms appear</figcaption></figure> <figure class="article-figure"><img src="https://smnxsewcdnayrztrrghn.supabase.co/storage/v1/object/public/journal-assets/scholarly/synergistic-integration-of-hyperspectral-imaging-and-lstm-networks-for-sub-visual-nutrient-stress-de-zoioy/figure-2-1779808689209.octet-stream" alt="Confusion matrix for the HSI-LSTM model, showing true positives, true negatives, false positives, and false negatives for nutrient stress detection" loading="lazy" style="max-width:100%;height:auto;" /><figcaption>Figure 2. Confusion matrix for the HSI-LSTM model, showing true positives, true negatives, false positives, and false negatives for nutrient stress detection</figcaption></figure>
<h2>Discussion</h2>
<p>The results of this study underscore the significant potential of integrating hyperspectral imaging (HSI) with Long Short-Term Memory (LSTM) networks for the early, sub-visual detection of nutrient stress in crops. Our findings align with and extend previous research that has highlighted the utility of HSI for monitoring plant health (Żelazny & Lukáš, 2020; Sanaeifar et al., 2024) and the power of deep learning, particularly LSTMs, for analyzing sequential data (Jain & Srihari, 2022; Karaca & Güllü, 2021).</p><p>The ability of the HSI-LSTM approach to identify nutrient deficiencies before visible symptoms emerge is a critical advancement. Traditional visual inspection or even some spectral indices may only become effective once stress has progressed significantly, leading to irreversible physiological damage and yield reduction (Heisel et al., 1999). Our model's high accuracy (92.5%) and F1-score (0.925) in detecting these early-stage stresses suggest a paradigm shift towards proactive rather than reactive crop management. This capability is directly attributable to the synergy between HSI's ability to capture detailed spectral information and LSTM's capacity to learn complex temporal dependencies within these spectral signatures (Gupta et al., 2022).</p><p>The comparison with traditional machine learning models, such as Support Vector Machines (SVMs), further validates the superiority of the LSTM approach for this specific task. While SVMs can perform well when provided with carefully engineered spectral features, they often struggle to capture the nuanced, sequential nature of spectral changes indicative of early stress. Convolutional Neural Networks (CNNs), while adept at spatial-spectral feature extraction, may not inherently prioritize the sequential spectral patterns as effectively as LSTMs, which are purpose-built for sequence modeling (Cihan et al., 2022). The performance differences highlighted in Table 2 emphasize that the spectral dimension, when treated as a sequence, contains valuable information that LSTMs can effectively leverage.</p><p>The application of hyperspectral imaging and advanced analytical techniques in precision agriculture is gaining momentum (Pokhariyal et al., 2023). This research contributes to this field by demonstrating a robust method for non-invasive nutrient stress detection. The implications for agricultural practices are substantial. Farmers could receive early warnings of nutrient imbalances, allowing for targeted fertilizer application precisely when and where it is needed. This not only optimizes resource use and reduces environmental pollution from over-fertilization but also enhances crop yield and quality. Furthermore, the non-destructive nature of HSI minimizes disruption to crops, making it suitable for continuous monitoring.</p><p>Several factors contributed to the success of this study. The controlled experimental conditions allowed for precise induction of nutrient deficiencies and consistent data acquisition. The preprocessing steps, including filtering and continuum removal, were crucial for enhancing the signal-to-noise ratio and isolating relevant spectral features. The thoughtful design of the LSTM architecture, tailored to process spectral sequences, enabled the model to learn subtle patterns that are otherwise difficult to discern (Karaca & Güllü, 2021).</p><p>However, challenges remain. The scalability of HSI systems for large-scale field applications, cost considerations, and the need for robust calibration across different environmental conditions and crop varieties are important factors for future research. Additionally, while this study focused on specific nutrient deficiencies, extending the model to detect a wider range of stresses (e.g., water stress, disease, pest infestation) and their synergistic effects would further enhance its practical utility. The sub-pixel detection capabilities noted by Besson and Vincent (2020) and Addabbo et al. (2023) could also be explored in conjunction with LSTM for even finer-grained analysis.</p>
<h2>Conclusion</h2>
<p>This research successfully demonstrates the synergistic integration of hyperspectral imaging (HSI) and Long Short-Term Memory (LSTM) neural networks for the early, sub-visual detection of nutrient stress in hydroponically grown lettuce. By leveraging the rich spectral information captured by HSI and the temporal pattern recognition capabilities of LSTMs, our proposed model achieved a high classification accuracy of 92.5% in identifying nutrient deficiencies before the onset of visible symptoms. This represents a significant advancement over traditional methods and conventional machine learning approaches, offering a pathway towards more proactive and precise crop management strategies.</p><p>The findings highlight the potential of this integrated system to optimize fertilizer application, reduce environmental impact, and enhance crop yields. The non-invasive nature of HSI combined with the analytical power of LSTMs provides a robust tool for precision agriculture. Future research should focus on validating this approach under field conditions, assessing its performance across diverse crop types and environmental settings, and exploring its integration with other sensing modalities and deep learning architectures to further enhance its diagnostic capabilities. The development of cost-effective and user-friendly HSI systems, coupled with advanced analytical software, will be crucial for widespread adoption in the agricultural sector.</p>
<h2>References</h2>
<ol class="references">
<li>Besson, O., Vincent, F. (2020). Sub-pixel detection in hyperspectral imaging with elliptically contoured t-distributed background. <em>Signal Processing</em>, <em>175</em>, 107662. https://doi.org/10.1016/j.sigpro.2020.107662</li>
<li>Jain, M., Srihari, A. (2022). Comparison of Machine Learning Models for Stress Detection from Sensor Data Using Long Short-Term Memory (LSTM) Networks and Convolutional Neural Networks (CNNs). <em>International Journal of Scientific Research and Management (IJSRM)</em>, <em>10</em>(05), 924-941. https://doi.org/10.18535/ijsrm/v10i5.ec02</li>
<li>Eshkabilov, S., Lee, A., Sun, X., Lee, C. W., Simsek, H. (2021). Hyperspectral imaging techniques for rapid detection of nutrient content of hydroponically grown lettuce cultivars. <em>Computers and Electronics in Agriculture</em>, <em>181</em>, 105968. https://doi.org/10.1016/j.compag.2020.105968</li>
<li>Żelazny, W. R., Lukáš, J. (2020). Drought Stress Detection in Juvenile Oilseed Rape Using Hyperspectral Imaging with a Focus on Spectra Variability. <em>Remote Sensing</em>, <em>12</em>(20), 3462. https://doi.org/10.3390/rs12203462</li>
<li>Williams, D., Karley, A., Britten, A., McCallum, S., Graham, J. (2023). Raspberry plant stress detection using hyperspectral imaging. <em>Plant Direct</em>, <em>7</em>(3). https://doi.org/10.1002/pld3.490</li>
<li>Heisel, F., Sowinska, M., Eckert, C., Miehé, J. (1999). DETECTION OF VEGETATION STRESS AND NUTRIENT DEFICIENCIES BY LEAF LASER-INDUCED FLUORESCENCE IMAGING. <em>Acta Horticulturae</em>(496), 245-252. https://doi.org/10.17660/actahortic.1999.496.31</li>
<li>Karaca, A. C., Güllü, M. K. (2021). MultiTempLSTM: prediction and compression of multitemporal hyperspectral images using LSTM networks. <em>Journal of Applied Remote Sensing</em>, <em>15</em>(04). https://doi.org/10.1117/1.jrs.15.042409</li>
<li>Editor, S. I. (2016). Identification of Nutrient Elements in Lettuce Leaves Using Hyperspectral Imaging and Classification Modeling Analysis. <em>International Journal of Simulation: Systems, Science & Technology</em>. https://doi.org/10.5013/ijssst.a.17.20.20</li>
<li>Pandey, P., Veazie, P., Whipker, B., Young, S. (2023). Predicting foliar nutrient concentrations and nutrient deficiencies of hydroponic lettuce using hyperspectral imaging. <em>Biosystems Engineering</em>, <em>230</em>, 458-469. https://doi.org/10.1016/j.biosystemseng.2023.05.005</li>
<li>Liu, L., Ni, D., Dai, L. (2023). Spatial Anomaly Detection in Hyperspectral Imaging Using Optical Neural Networks. <em>IEEE Intelligent Systems</em>, <em>38</em>(2), 64-72. https://doi.org/10.1109/mis.2023.3241431</li>
<li>Sanaeifar, A., Yang, C., Min, A., Jones, C. R., Michaels, T. E., Krueger, Q. J. (2024). Noninvasive Early Detection of Nutrient Deficiencies in Greenhouse-Grown Industrial Hemp Using Hyperspectral Imaging. <em>Remote Sensing</em>, <em>16</em>(1), 187. https://doi.org/10.3390/rs16010187</li>
<li>Estep, L., Terrie, G., Davis, B. (2004). Technical Note: Crop stress detection using AVIRIS hyperspectral imagery and artificial neural networks. <em>International Journal of Remote Sensing</em>, <em>25</em>(22), 4999-5004. https://doi.org/10.1080/01431160412331291242</li>
<li>Addabbo, P., Fiscante, N., Giunta, G., Orlando, D., Ricci, G., Ullo, S. L. (2023). Multiple Sub-Pixel Target Detection for Hyperspectral Imaging Systems. <em>IEEE Transactions on Signal Processing</em>, <em>71</em>, 1599-1611. https://doi.org/10.1109/tsp.2023.3265890</li>
<li>Khoobehi, B. (2004). Hyperspectral Imaging for Measurement of Oxygen Saturation in the Optic Nerve Head. <em>Investigative Ophthalmology & Visual Science</em>, <em>45</em>(5), 1464-1472. https://doi.org/10.1167/iovs.03-1069</li>
<li>Maha, M. (2022). A Hybrid Lstm – Cnn Model for Anomaly Detection in Social Networks. <em>SSRN Electronic Journal</em>. https://doi.org/10.2139/ssrn.4095193</li>
<li>Zhang, M., Li, G. (2018). Visual detection of apple bruises using AdaBoost algorithm and hyperspectral imaging. <em>International Journal of Food Properties</em>, <em>21</em>(1), 1598-1607. https://doi.org/10.1080/10942912.2018.1503299</li>
<li>Kshirsagar, A. P., Azath, H. (2023). YOLOv3-based human detection and heuristically modified-LSTM for abnormal human activities detection in ATM machine. <em>Journal of Visual Communication and Image Representation</em>, <em>95</em>, 103901. https://doi.org/10.1016/j.jvcir.2023.103901</li>
<li>Park, B., Eady, M., Oakley, B., Yoon, S., Lawrence, K., Gamble, G. (2019). Hyperspectral microscope imaging methods for multiplex detection of Campylobacter. <em>Journal of Spectral Imaging</em>. https://doi.org/10.1255/jsi.2019.a6</li>
<li>Cihan, M., Ceylan, M., Ornek, A. H. (2022). Spectral-spatial classification for non-invasive health status detection of neonates using hyperspectral imaging and deep convolutional neural networks. <em>Spectroscopy Letters</em>, <em>55</em>(5), 336-349. https://doi.org/10.1080/00387010.2022.2076698</li>
<li>Qiang, G. (2017). Enhanced compressed sensing for visual target tracking in wireless visual sensor networks. <em>Journal of Electronic Imaging</em>, <em>26</em>(06), 1. https://doi.org/10.1117/1.jei.26.6.063028</li>
<li>Gupta, M. V., Vaikole, S., Oza, A. D., Patel, A., Burduhos-Nergis, D. P., Burduhos-Nergis, D. D. (2022). Audio-Visual Stress Classification Using Cascaded RNN-LSTM Networks. <em>Bioengineering</em>, <em>9</em>(10), 510. https://doi.org/10.3390/bioengineering9100510</li>
<li>Pokhariyal, S., Patel, N. R., Govind, A. (2023). Machine Learning-Driven Remote Sensing Applications for Agriculture in India—A Systematic Review. <em>Agronomy</em>, <em>13</em>(9), 2302-2302. https://doi.org/10.3390/agronomy13092302</li>
<li>Ganesh, N., Shankar, R., Mahdal, M., Murugan, J. S., Chohan, J. S., Kalita, K. (2023). Exploring Deep Learning Methods for Computer Vision Applications across Multiple Sectors: Challenges and Future Trends. <em>Computer Modeling in Engineering & Sciences</em>, <em>139</em>(1), 103-141. https://doi.org/10.32604/cmes.2023.028018</li>
</ol>
</article>