Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 10s (2025) 1604 https://internationalpubls.com Deep Learning-Based Crop Disease Detection using IoT and Image Processing 1 Mr. Khilendra Tumareki, 2Prof(Dr) Asha Ambhaikar, 3Mr. Suraj Kumar Sahu 1 M.Tech. Scholar , 2Director, 3Associate Professor 2, 3 Department of Computer Science & Engineering 1, 2,3 MATS School of Engineering & IT, MATS University Aarang, 1 Khilendratumareki@gmail.com ,2drasha@matsuniversity.ac.in, 3sahu79487@gmail.com Article History: Received: 12-01-2025 Revised: 15-02-2025 Accepted: 01-03-2025 Abstract: Crop diseases pose a significant threat to global agricultural productivity, with the potential to cause severe economic losses, food shortages, and a negative environmental impact. In traditional agriculture, the early detection and timely management of plant diseases have been challenging tasks, often relying on manual inspection, expert knowledge, and reactive measures. However, as agriculture becomes more reliant on technology, innovative solutions have emerged to address these issues, offering more efficient, automated, and accurate methods for disease detection and management. Keywords: agricultural, environmental, automated, productivity. 1 Introduction Crop diseases pose a significant threat to global agricultural productivity, with the potential to cause severe economic losses, food shortages, and a negative environmental impact. In traditional agriculture, the early detection and timely management of plant diseases have been challenging tasks, often relying on manual inspection, expert knowledge, and reactive measures. However, as agriculture becomes more reliant on technology, innovative solutions have emerged to address these issues, offering more efficient, automated, and accurate methods for disease detection and management. One such advancement is the integration of Internet of Things (IoT) sensors with deep learning techniques, enabling the collection and analysis of both environmental and visual data. IoT sensors can provide real-time information about environmental conditions such as temperature, humidity, and soil pH, which play a crucial role in the development and spread of crop diseases. Meanwhile, image processing technologies, particularly those leveraging deep learning models like Convolutional Neural Networks (CNNs), allow for the automatic extraction of visual features from crop images, enabling the identification of disease symptoms on leaves or other parts of the plant. This research aims to explore and develop an advanced crop disease detection system that combines IoT-based environmental sensing with deep learning-based image analysis. The proposed system uses high-resolution images of crop leaves, captured through field-deployed cameras, and environmental data collected by IoT sensors installed in the crop field. By employing a self-attention mechanism mailto:Khilendratumareki@gmail.com Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 10s (2025) 1605 https://internationalpubls.com within a deep neural network, the system can dynamically focus on disease-specific patterns in the data, improving classification accuracy. The significance of this study lies in its potential to provide farmers with a powerful, automated tool for the early detection of crop diseases. Early intervention can significantly reduce the economic burden of diseases and minimize the use of pesticides, leading to more sustainable agricultural practices. Additionally, this research contributes to the broader field of precision agriculture, where data-driven solutions are transforming the way crops are monitored, managed, and protected. This paper outlines the methodology for crop disease detection, explains the deep learning models and techniques employed, and discusses the experimental results obtained from the proposed system. It also highlights the challenges faced during development and the potential future directions for improving the system's accuracy, scalability, and applicability in real-world agricultural settings. 2 Literature Review The field of crop disease detection has gained significant attention in recent years, driven by the need for more efficient and sustainable agricultural practices. Traditional methods of disease detection, primarily based on visual inspection by experts, have limitations in terms of speed, accuracy, and scalability. As a result, there has been a growing interest in utilizing advanced technologies such as image processing, Internet of Things (IoT), and machine learning to address these challenges. In particular, deep learning-based models have shown great promise in automating the process of disease detection through the analysis of high-resolution images of crops. The integration of IoT sensors, which monitor environmental factors such as humidity, temperature, and soil conditions, has further enhanced the accuracy of these systems by providing critical contextual information. Recent studies have explored various machine learning techniques, including convolutional neural networks (CNNs), transfer learning, and attention mechanisms, to improve the precision and reliability of crop disease classification. This chapter reviews key research contributions in the area of crop disease detection, highlighting the methodologies, innovations, and challenges encountered in developing robust systems for real-time disease monitoring and intervention. Paper Title Key Findings Remarks Lightweight Meta- Ensemble Approach • 94-98% accuracy across datasets • Reduced parameters (< 1M) • IoT device compatibility • Meta-ensemble of MLP- Mixer and LSTM Basic feature concatenation without selection mechanism; Simple SVM classifier limiting feature interaction Transfer Learning with C-GAN • Synthetic data generation • 97% classification accuracy • DenseNet121 backbone Innovative data synthesis but computationally expensive; Limited real-world validation Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 10s (2025) 1606 https://internationalpubls.com • Improved data augmentation EfficientNet for Classification • Multi-crop disease detection • 95% accuracy achieved • Efficient feature extraction • Lightweight architecture Good efficiency but lacks feature selection; No attention mechanism implementation Vision Transformer Approach • 93% average accuracy • Explainable AI approach • Good feature visualization • Multi-head attention Strong attention framework but resource-intensive; Complex deployment requirements Novel CNN Architecture • Real-time processing capability • 96% classification accuracy • Multi-disease classification • Optimized architecture Efficient real-time processing but basic feature handling; No advanced feature selection Attention Dense CNN • Smartphone compatibility • 97% accuracy achieved • Real-world image processing • Resource-efficient design Mobile-friendly but limited feature interaction; Basic attention implementation 3 Methodology 3.1 Proposed Methodology The proposed methodology aims to accurately detect crop diseases by leveraging advanced deep learning techniques integrated with IoT-based environmental sensing. The methodology comprises the following core components: image acquisition, IoT data collection, preprocessing, feature extraction, feature selection, fusion, and final classification through a self-attention-based deep neural network. Let the dataset be denoted by 𝒟 = {(𝐼𝑖, 𝑆𝑖, 𝑦𝑖)}𝑖=1 𝑁 where: Ii is the ith image of a crop leaf captured via camera-based monitoring, Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 10s (2025) 1607 https://internationalpubls.com Si ∈ Rm represents the IoT sensor data vector corresponding to the environmental condition (e.g., humidity, temperature, soil pH, light intensity), yi ∈ {1,2, … , C} is the label indicating one of C known crop disease classes or healthy condition. 3.1.1 Image Acquisition and IoT Sensing High-resolution leaf images are captured using field-deployed imaging devices, while environmental context is gathered via IoT sensors installed in the crop field. The IoT sensors continuously monitor ambient parameters, which are critical for disease correlation analysis. 3.1.2 Preprocessing • Image data undergoes several preprocessing steps: • Resizing all images to fixed resolution (𝐻 ×𝑊). • Normalization of pixel values to standardize input distribution. • Data augmentation (rotation, flipping, zooming, etc.) to artificially increase dataset diversity and robustness. Sensor data is standardized via feature scaling techniques (e.g., min-max normalization or z-score normalization) to ensure uniformity across different physical units. 3.1.3 Deep Feature Extraction A pretrained deep convolutional neural network (CNN), such as ResNet-18, is used to extract semantic features from leaf images: 𝑥𝑖 = 𝑓CNN(𝐼𝑖), where 𝑓CNN: 𝑅 𝐻×𝑊×𝟛 → 𝑅𝑑 This generates a d-dimensional feature vector for each input image Ii. These deep features capture edges, textures, and high-level patterns indicative of various crop diseases. 3.1.4 Feature Selection using mRMR From the high-dimensional space of extracted features, a subset of features ℱselected is chosen using the Minimum Redundancy Maximum Relevance (mRMR) strategy. The objective is to maximize relevance to the target variable $y$ while minimizing redundancy among selected features: ℱselected = argmax 𝑆⊆ℱ [ 1 |𝑆| ∑ 𝐼(𝑓𝑖; 𝑦) 𝑓𝑖∈𝑆 − 1 |𝑆|2 ∑ 𝐼(𝑓𝑖; 𝑓𝑗) 𝑓𝑖,𝑓𝑗∈𝑆 ] where I(fi; y) is the mutual information between feature fi and the label y. 3.1.5 Feature Fusion (Image + IoT) If sensor data Si is available, it is concatenated with the selected deep features: 𝑧𝑖 = [𝑥𝑖 selected|𝑆𝑖], 𝑧𝑖 ∈ 𝑅𝑘+𝑚 where xi selected is the reduced image feature vector, and $k$ is the number of selected features. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 10s (2025) 1608 https://internationalpubls.com 3.1.6 Self-Attention-Based Classification A self-attention mechanism is incorporated within a deep neural network to enable dynamic weighting of features: Attention(𝑄, 𝐾, 𝑉) = softmax( 𝑄𝐾𝑇 √𝑑𝑘 )𝑉 where the query, key, and value matrices Q, K, V are derived from linear transformations of the fused feature input zi. This mechanism allows the model to focus on disease-specific patterns. The output of attention is passed through fully connected layers to produce class probabilities: 𝑦�̂� = argmax(softmax(𝑊out ⋅ Attention(𝑄, 𝐾, 𝑉) + 𝑏out)) 3.2 Algorithm: Self-Attention Based Crop Disease Classification Crop Disease Detection using Self-Attention Deep Neural Network Input: Image dataset 𝒟, pretrained CNN fCNN, number of features k Output: Predicted class labels {yî} for each test image foreach sample (Ii, Si) in 𝒟 do Preprocess image Ii; Normalize and scale sensor vector Si; Extract deep features: xi ← fCNN(Ii); Select top-k features: xi selected ← mRMR(xi); Fuse with sensor data: zi ← [xi selected|Si]; Compute attention matrices Q, K, V; Calculate attention output ai ← Attention(Q, K, V); Predict disease label: yî ← argmax (softmax(fFC(ai))); return yî 3.3 Methodology Flowchart The flowchart illustrates a process for classifying plant species using deep learning techniques. It starts with a dataset of plant images, likely focusing on different plant leaves. These images serve as the raw input for the model, where the first step involves extracting relevant features from them using a deep neural network (DNN). In this stage, the model is trained to recognize patterns within the images, converting the raw pixel data into more abstract representations that capture the important features necessary for distinguishing between plant species. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 10s (2025) 1609 https://internationalpubls.com Following feature extraction, the next step is feature selection. This is a crucial process where the most relevant and significant features are chosen while discarding those that are less important or redundant. This step helps improve the model's efficiency by reducing the dimensionality of the data, ensuring that the neural network works with only the most informative features. It also contributes to the overall accuracy of the classification by minimizing noise from irrelevant data. Figure 1 Methodology Flowchart After feature selection, the model incorporates a self-attention mechanism in its deep neural network. Self-attention enables the model to focus more on important aspects of the extracted features that are crucial for classification, allowing the model to weigh different features differently based on their relevance. This step improves the model's ability to understand complex relationships between features, which enhances its performance in classifying the plant species more accurately. Finally, the processed data passes through the model to produce the final prediction, which is the classification of the plant species based on the learned features and the self-attention mechanism. The flowchart encapsulates a sophisticated process that combines deep learning, feature selection, and attention mechanisms to accurately classify plant images, providing a highly efficient and effective classification pipeline. 4 Result and Discussion 4.1 Results The proposed methodology for crop disease detection using IoT sensors and image processing integrated with deep learning techniques has shown promising results. The dataset used for this study included high-resolution crop leaf images, paired with IoT sensor data that provides critical environmental context such as humidity, temperature, soil pH, and light intensity. These images and Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 10s (2025) 1610 https://internationalpubls.com sensor data were processed using a deep convolutional neural network (CNN), specifically ResNet-18, to extract meaningful features indicative of crop diseases. After preprocessing the image data, including resizing, normalization, and data augmentation, the features were extracted, and a subset of relevant features was selected using the Minimum Redundancy Maximum Relevance (mRMR) strategy. This feature selection method ensured that the most important features for predicting crop diseases were retained while minimizing redundancy. Figure 2 Confusion Matrix The model incorporated a self-attention mechanism to dynamically focus on important features during the classification process. This mechanism allowed the network to weigh the extracted features based on their relevance, significantly improving the model's accuracy in identifying disease patterns specific to different crop diseases. The confusion matrix presented in the results demonstrates the efficacy of the classification process. The model achieved an impressive accuracy of 99.34%, which indicates the robustness and reliability of the deep learning-based approach for crop disease classification. The matrix shows that the model correctly predicted the majority of the labels with few misclassifications, as evident from the near- perfect diagonal line in the confusion matrix, signifying correct predictions across most of the disease classes. 4.2 Discussion The high classification accuracy achieved by the proposed method can be attributed to several key factors. First, the integration of IoT sensor data with image-based deep learning allows the model to consider both visual cues from the crop leaves and the environmental conditions that might influence disease development. This fusion of image and sensor data ensures that the model has a more holistic Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 10s (2025) 1611 https://internationalpubls.com understanding of the crop's condition. The use of deep learning, specifically the CNN, allows the model to automatically learn hierarchical features from the images, reducing the need for manual feature engineering. ResNet-18, as a pretrained model, further enhanced feature extraction by leveraging its ability to capture high-level patterns in the images. The mRMR feature selection technique played a crucial role in improving the model's performance by focusing on the most relevant features. This helped in reducing the dimensionality of the data while ensuring that the most important information for disease classification was retained. By removing redundant features, the model was able to make faster and more accurate predictions. The self-attention mechanism incorporated in the deep neural network architecture allowed the model to dynamically assign different weights to features based on their importance, further improving its ability to focus on disease-specific patterns. This feature is particularly beneficial for crop disease detection, where subtle variations in leaf images may signify the onset of a disease. 5 Conclusion and Future Scope 5.1 Conclusion This study presents an advanced methodology for crop disease detection by integrating Internet of Things (IoT) sensor data with deep learning techniques, specifically leveraging self-attention-based deep neural networks. The approach effectively combines image processing of high-resolution leaf images and environmental data from IoT sensors to identify crop diseases with high accuracy. The deep convolutional neural network (CNN), specifically ResNet-18, was employed for feature extraction, followed by a Minimum Redundancy Maximum Relevance (mRMR) feature selection technique that ensured the most relevant features were used for classification. The inclusion of a self-attention mechanism allowed the model to dynamically focus on the most important features, enhancing its ability to classify crop diseases accurately. The model achieved an impressive classification accuracy of 99.34%, as demonstrated by the confusion matrix, which highlights the robustness of the system in correctly predicting disease labels across multiple classes. This level of accuracy indicates that the proposed methodology is not only effective but also highly reliable for real-world applications in agricultural settings. The integration of IoT data and deep learning presents a novel and powerful solution for precision agriculture. By leveraging environmental context along with visual data, the model offers a more comprehensive understanding of the crop's health, which can help in early disease detection and intervention. This work paves the way for the development of autonomous and intelligent systems capable of supporting farmers in managing crop diseases efficiently and effectively. 5.2 Future Scope The future scope of this research lies in expanding the system’s capabilities by integrating a larger and more diverse dataset, incorporating real-time field deployment for dynamic environmental conditions, and optimizing for mobile or drone-based platforms. Further advancements could include enhancing the model’s robustness through regularization techniques, exploring multi-modal data fusion for more accurate disease classification, and implementing edge computing for faster decision-making. Additionally, the system could be integrated with precision agriculture tools for automated disease Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 10s (2025) 1612 https://internationalpubls.com management, while ensuring scalability and cost-effectiveness for widespread adoption. Addressing the explainability of deep learning models will also be crucial in ensuring transparency and trust in the predictions made by the system, ultimately supporting smarter and more efficient agricultural practices. References [1] Ritesh Maurya, Satyajit Mahapatra, and Lucky Rajput. A lightweight meta-ensemble approach for plant disease detection suitable for iot-based environments. IEEE Access, 12:28096–28108, 2024. [2] A Abbas, S Jain, M Gour, and S Vankudothu. Tomato plant disease detection using transfer learning with c-gan synthetic images. Computers and Electronics in Agriculture, 187:106279, 2021. [3] Umit Atila, Murat U¸car, Kemal Akyol, and Ercan U¸car. Plant leaf disease classification using efficientnet deep learning model. Ecological Informatics, 61:101182, 2021. [4] Pradeep Singh Thakur, Preeti Khanna, Tanuja Sheorey, and Ayushi Ojha. Explainable vision transformer enabled convolutional neural networks for plant disease identification: Plantxvit. arXiv preprint arXiv:2207.07919, 2022. [5] Sunil Maurya Hassan and Amit Kumar Maji. Plant disease identification using a novel convolutional neural network. IEEE Access, 10:5390–5401, 2022. [6] Atul Pandey and Kamal Jain. A robust deep attention dense convolutional neural network for plant leaf disease identification and classification from smart phone captured real world images. Ecological Informatics, 70:101725, 2022 [7] Loddo, M. Loddo, and C. Di Ruberto, ‘‘A novel deep learning based approach for seed image classification and retrieval,’’ Comput. Electron. Agricult., vol. 187, Aug. 2021, Art. no. 106269, doi:10.1016/j.compag.2021.106269. [8] Qian, M. Tong, X. Yu, and S. Zhuang, ‘‘CNN-based visual processing approach for biological sample microinjection systems,’’ Neurocomputing, vol. 459, pp. 70–80, Oct. 2021, doi: 10.1016/j.neucom.2021.06.085. [9] R. Maurya, V. K. Pathak, and M. K. Dutta, ‘‘Deep learning based microscopic cell images classification framework using multi-level ensemble,’’ Comput. Methods Programs Biomed., vol. 211, Nov. 2021, Art. no. 106445, doi: 10.1016/j.cmpb.2021.106445. [10] S. Yadav, N. Sengar, A. Singh, A. Singh, and M. K. Dutta, ‘‘Identification of disease using deep learning and evaluation of bacteriosis in peach leaf,’’ Ecolog. Informat., vol. 61, Mar. 2021, Art. no. 101247. [11] Y. Zhao, C. Sun, X. Xu, and J. Chen, ‘‘RIC-Net: A plant disease classification model based on the fusion of inception and residual structure and embedded attention mechanism,’’ Comput. Electron. Agricult., vol. 193, Feb. 2022, Art. no. 106644, doi: 10.1016/j.compag.2021.106644. [12] R. Maurya, R. Burget, R. Shaurya, M. Kiac, and M. K. Dutta, ‘‘Multi-head attention-based transfer learning approach for porato disease detection,’’ in Proc. 15th Int. Congr. Ultra Modern Telecommun. Control Syst. Workshops (ICUMT), Oct. 2023, pp. 165–169, doi:10.1109/ICUMT61075.2023.10333272. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 10s (2025) 1613 https://internationalpubls.com [13] X. Chen, G. Zhou, A. Chen, J. Yi, W. Zhang, and Y. Hu, ‘‘Identification of tomato leaf diseases based on combination of ABCK-BWTR and B-ARNet,’’ Comput. Electron. Agricult., vol. 178, Nov. 2020, Art. no. 105730. [14] Tolstikhin, N. Houlsby, A. Kolesnikov, L. Beyer, X. Zhai, T. Unterthiner, J. Yung, A. Steiner, D. Keysers, J. Uszkoreit, M. Lucic, and A. Dosovitskiy, ‘‘MLP-mixer: An all-MLP architecture for vision,’’ 2021, arXiv:2105.01601. [15] Pandey and K. Jain, ‘‘A robust deep attention dense convolutional neural network for plant leaf disease identification and classification from smart phone captured real world images,’’ Ecolog. Informat., vol. 70, Sep. 2022, Art. no. 101725, doi: 10.1016/j.ecoinf.2022.101725. [16] P. Bedi and P. Gole, ‘‘Plant disease detection using hybrid model based on convolutional autoencoder and convolutional neural network,’’ Artif. Intell. Agricult., vol. 5, pp. 90–101, Jan. 2021. [17] Identity Based Authentication Scheme (IAS) for Securing WSN Based internet of Things (Journal of Electrical Systems) pp . Vol 20 N0. 2s(2024) (April 2024) ISSN 1112-5209 https://journal.esrgroups.org/jes/article/view/1794 [18] The Impact of QoS parameters on the Performance of IoT Application. International Journal of Intelligent System and Application in Engineering. (IJISAE) Volume -12 no3 pp (15 March 2024) ISSN 2147-6799 https://ijisae.org/index.php/IJISAE/article/view/6155 [19] The Impact of Quality of Service (QoS) Parameters on IoT Application Performance Volume 3, Issue 3, 2024 , PP 1-20 , ISSN 2583-6196 https://journal.inence.org/index.php/ijfiahm/article/view/363/263 https://journal.esrgroups.org/jes/article/view/1794 https://ijisae.org/index.php/IJISAE/article/view/6155 https://journal.inence.org/index.php/ijfiahm/article/view/363/263