Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Scientific Reports volume 15, Article number: 1456 (2025)
Metrics details
Abstract The Internet of Things (IoT) has recently attracted substantial interest because of its diverse applications. In the agriculture sector, automated methods for detecting plant diseases offer numerous advantages over traditional methods. In the current study, a new model is developed to categorize plant diseases within an IoT network. The IoT network is simulated for monitoring crop diseases. Routing is performed with Henry Gas Chicken Swarm Optimization (HGCSO), which is designed by integrating Henry Gas Solubility Optimization (HGSO) and Chicken Swarm Optimization (CSO). The fitness parameters of the model include delay, energy, distance, and link lifetime (LLT). At the Base Station (BS), plant disease categorization is performed by collecting plant leaf images. Preprocessing is done on the input images using median filtering. Various features, such as Histogram of Oriented Gradient (HoG), statistical features, Spider Local Image Features (SLIF), and Local Ternary Patterns (LTP) are extracted. Plant disease categorization is carried out using a Deep Residual Network (DRN), which is trained using the developed Caviar Henry Gas Chicken Swarm Optimization (CHGCSO) that combines the CAViaR model with HGCSO. Comparative results show an accuracy of 94.3%, a maximum sensitivity of 93.3%, a maximum specificity of 92%, and an F1-score of 93%, indicating that the CHGCSO-based DRN outperforms existing methods. Graphic Abstract
The IoT is an emerging technology that is rapidly developing in today’s world. To fully leverage IoT for specific applications, an effective sensing and monitoring model is essential. IoT enables the creation of a smarter, more interconnected world25,36. IoT in farming has recently inspired various innovative designs. It enables remote access, allowing for energy savings and more efficient resource management20,22. An IoT platform for disease detection can significantly boost crop production by alerting farmers early, allowing them to take action before the disease causes damage17,32. The type of disease detected by an IoT device is often not immediately apparent to the farmer. Additionally, modelling plant disease infection can be time-consuming and costly. Automating plant disease detection improves accuracy while reducing costs. Integrating image processing with IoT is crucial for precise assessment and timely intervention34.
The detection and classification of diseases in rice plants15,16are based on digital images and have gained significant attention in the research community. The automated model can assist farmers by guiding them in selecting the right pesticides at the right time37. Some researchers have focused on optimization algorithms19,30,38,43,44and deep learning, particularly Convolutional Neural Networks (CNNs)5, to address crop disease detection. These approaches have demonstrated the ability to outperform previous methods in various automated identification tasks, such as natural language processing and visual recognition. Deep learning is a well-known technique for recognition tasks21. Despite the efficiency of CNNs in general object recognition, deep learning offers an effective model for identifying diseases in crops45. Automated plant disease detection technology offers advantages for plant monitoring and reduces the reliance on manual inspections27,40. It also aids in early disease identification, decreases pesticide use, and supports precision agriculture7,26,33,35.
The main contributions are highlighted as follows:
HGCSO for routing: The HGCSO is utilized for routing, which is obtained by combining HGSO and CSO.
CHGCSO-based DRN: The CHGCSO is adapted for classifying plant disease. Here, the CHGCSO is adapted for training DRN. The CHGCSO is obtained by combining HGSO and CSO and the CAViaR model.
The upcoming sections are organized as follows: Section 2 provides an overview of traditional plant disease classification methods. Section 3 explains the IoT framework. Section 4 presents the CHGCSO-based DRN for classifying plant diseases. Section 5 assesses the efficiency of classical classification approaches, followed by the conclusion in Section 6.
In 2018, S. Ramesh and Bharghava Rajaram31 developed a model for earlier discovery of disease in rice crops from visual symptoms. The rice crops were utilized for the processing. It provided on-field disease discovery and prediction. However, this scheme did not perform earlier discovery of crop disease. Artzai Picon et al.29 utilized a CNN model for classifying plant disease in 2019. Here, the information based on images were incorporated. It combined the benefits of concurrently learning while minimizing the complication of classifying the disease. However, misclassification occurs whenever the quality of the image is low. In 2019, Artzai Picon et al.28 utilized a Deep convolution Neural Network (DCNN)-based algorithm for dealing with plant disease. Here, the discovery of their relevant disease was done including Septoria, Tan Spot and Rust. However, this method acquired huge generalization power. Monalisa Mishra et al.24 developed an automatic technique for detecting plant disease in an IoT. The method placed the nodes with a simulation platform to capture plant leaves. The model maintained the sink node that accumulated data and assisted in monitoring IoT in 2020. Here, the Sine Cosine Algorithm-based Rider Neural Network (SCA-based RideNN) was introduced for classifying diseases in plants. However, the method consumed more time for processing. In 2020, Yushan Zhao et al.45 devised a Multi-Context Fusion Network (MCFN) for classifying the plant disease. Here, it adapted with CNN for the visual features extraction from crop samples. The contextual features accumulated from sensors help to the classification of crop disease. Here, the fusion of contextual features and visual features were done by a deep fully connected network,but this model poses unbalanced data, which affects the performance. Kumari Shibani et al.36 developed a technique, namely Random Forest and Support Vector Machine and Logistic Regression (RF+SVM+LR) for classifying plant disease in 2020. Here, the first part aimed to model a system wherein the tracking of plant needs was done with sensors. Here, the automated water controller tasks with moisture content fall below a specific threshold. However, this method suffered from high computational costs. In 2020, Juan Wen et al.42 developed Enhanced Super-Resolution Generative Adversarial Networks (ESRGAN) for classifying plant diseases. Here, the ESRGAN was utilized for recovering the super-resolution from low-resolution images. In addition, the Transfer learning was adapted for compensating training samples. It was unable to process other databases. David Argüeso et al.3 established a plant disease identification model using distance metric Few-Shot Learning (FSL) using Triplet loss in 2020. Here, the CNN was utilized for extracting the general plant leaf features. This approach shortened the time taken for production and minimized the cost of plant disease. However, this method generated an error value, which affected the classification performance. In 2021, Olusola Oluwakemi Abayomi-Alli et al.1 devised cassava disease recognition. Here, an image colour histogram alteration scheme was implemented for the creation of synthetic images to perform data augmentation. It detected and recognized the disease in cassava leaf effectively with less images. This system failed to identify a variety of diseases in plants. Ahmad Almadhor et al2. devised an Artificial Intelligence (AI)-driven approach for Disease Recognition of Guava Plant in 2021. It helped the farmers to avoid the production loss due to the early and effective finding of plant disease. However, this model gained less accuracy for some specific classes of disease. In 2021, Nidhi Kundu et al 17, devised an Automatic and Intelligent Data Collector and Classifier. The data used in this analysis was gethered from the pearl millet farmland. It was located in ICAR, Mysore, India and the parametric and farmland data were collected. These gathered information were processed in Raspberry Pi and cloud server with the help of ‘Custom-Net’ approach in the disease identification of pearl millet. It was an automated scheme with low-cost and user-friendly tool for farmers. However, the safety of data was the major drawback. Zia ur Rehman et al.32, devised novel parallel real-time processing in 2021. Here for improving the visual image impact a hybrid contrast stretching scheme was utilized and the infected areas detection was done by MASK RCNN. At the same time, CNN was trained for extracting the features. The Ensemble Subspace Discriminant Analysis (ESDA) was utilized to obtain good accuracy. However, it was trained by a small number of ground truth images. Table 1 shows the literature survey of the previous works.
IoT comprises several kinds of objects that include smart devices, and are linked amongst each other to exchange collected data over the network. Various smartest devices are devised on the basis of resources that comprise communication and processing ability for exchanging data. Figure 1 displays the IoT model in which various nodes are splitted for sending data to BS with effective paths. The smartest objects are energy-constrained, and hence gateway must be interested for supervision. Consider IoT network posesl nodes given as, (H=left{ H_1,H_2,dots ,H_l,dots ,H_kright}) that relies in coverage area D. Each IoT node k attains data based on the plant leaves. IoT tends to be imperative in enhancing production in agriculture and is utilized to monitor soil temperature. Furthermore, IoT strengthens efficiency and minimizes physical work and time and helps to make farming more effective.
IoT model.
The energy model9 in IoT is examined using data transmission using various nodes. Whenever a normal sensor node sendsdbits data then the energy of the IoT is expressed as,
where, (E_a(d)) expresses transmitted energy, (E_c) symbolizes consumed energy when 1-bit data send by the the node, (E_b) signifies utilized energy whenever a node transmits and receives 1-bit data, f symbolizes communication area, and d signifies data bits.
where, (E_b) symbolizes consumed energy in the free space approach, (E_g) signifies energy consumed in the multipath fading scheme.
The energy helps to receive d bits of data and formulated as,
where, (E_m(d)) express received energy
where, (E_m) denotes the energy utilization when sending and receiving 1-bit data, (epsilon (d)) implies energy utilization of sensor nodes, d denotes the data bits.
LLT4 signifies network lifetime, which associates two sensor nodes in IoT. Consider (C_1)and (C_2) signifies two mobile nodes located at (A_{C1},B_{C1}) and (A_{C2},B_{C2}) for which the LLT is formulated as,
such that,
where, l signifies range of transmission, (G_{C1}) symbolizes speed of mobility in the sensor node (C_1), (G_{C2}) signifies the sensor node’s mobility speed (C_2), (A_{C2},B_{C2}) signifies node coordinate, (C_2), (A_{C1},B_{C1}) symbolize node coordinate (C_1), ({theta }_{C1}) refers motion direction of node (C_1), and ({theta }_{C2}) signifies direction of motion in node (C_2).
The goal is to devise a CHGCSO-based DRN for classifying plant disease in IoT. At first, the IoT nodes are simulated HGCSO performs routing based on fitness function. The HGCSO is devised by combining HGSO13and CSO23. After the routing, the rice plant disease classification is performed at the BS. In the BS, the input image is fed into the preprocessing. Then, the features like HoG18, statistical features, SLIF11and LTP39are extracted. Finally, the plant disease classification is done with DRN8, which will be trained using CHGCSO. The CHGCSO is designed by combining the CAViaR model10, HGSO and CSO. Figure 2 depicts the architecture of the HGCSO-based DRN.
Block diagram of HGCSO-based Deep Residual Network for Plant Disease Classification in IoT.
The routing is carried out with HGCSO using fitness function.
Determining the best solution to optimization problems requires solution encoding. Here, the HGCSO is adapted for choosing a suitable solution wherein the optimum node is chosen for routing. On the basis of fitness, the solution set attains optimum routing. This routing is performed using fitness function. Figure 3 presents solution modelling for optimum routing with HGCSO. Here, esignifies the node index.
Solution representation for routing with HGCSO.
The evaluation of fitness for optimized routing comprises certain parameters that include energy, distance, LLT and delay.
where, (E_i) signifies energy consumption in (i^{th}) node, (F_{i,j}) is Euclidean distance between (i^{th}) and (j^{th}) node, (L_{ij}) symbolizes LLT, and t symbolizes delay.
Here, the distance is given as,
where, (n_i) and (n_j) signifies distance between (i^{th}) and (j^{th}) node.
The delay is formulated as,
where, l symbolizes the entire nodes in a path, and h signifies the entire nodes in a network. The LLT is described in section 3.2.
The routing is done using the proposed HGCSO, and is produced by integrating HGSO13and CSO23. Algorithm 1 displays the pseudo-code of HGCSO and the flowchart of the HGCSO is given in Figure 4. The steps of HGCSO are specified below:
Step 1: Initiation The count and spot of gases are initialized as,
where, u signifies random numeral in [0, 1], (H_{max}) and (H_{min}) symbolize bounds of the problem, and e refers iteration time.
Step 2: Discovery of Fitness The fitness is already described in section 4.1.2.
Step 3: Clustering Divided into equal parts by gas type, each cluster of agents consists of the same gases and shares the same Henry’s constant value.
Step 4: Evaluation Evaluation in each cluster seeks out the gas that reaches the highest equilibrium state, after which a ranking process identifies the optimal gas from the entire swarm.
Step 5: Update Henry coefficient The Henry coefficient of cluster w at redundancy (u+1) is given as,
where, ({lambda }_w(u)) signifies Henry coefficient for cluster w at iteration u, (alpha) symbolize temperature, ({alpha }^{theta })refers constant in such a way that ({alpha }^{theta }=298.15).
Step 6: Update Solubility
The solubility is specified as,
where, (K_{o,w}(u)) signifies biased force on gas o in cluster w, and x is constant, and ({lambda }_w(u+1)) refers Henry coefficient to the cluster w at redundancy (u+1).
where, (K_{o,w}(u))denotes limited pressure on gas oin clusterw, and xis constant, and ({lambda }_w(u+1))denotes to the cluster w Henry coefficient at iteration (u+1).
Step 7: Update Position
The HGSO is adapted to strengthen algorithm performance and resolve optimization issues. According to HGSO13 , the update equation is,
where, (M_{o,w}(u)) refers position of gas oin cluster w, Y refers constant, u indicates iteration time, (M_{o,best}(u)) indicates best gas o in cluster w, (M_{best}(u)) signifies optimum gas in swarm, (beta) implies the gas capability o, and (gamma) denotes influence in gas o by other gases in cluster w, and X denotes flag that alters the direction of the search agent and (R_{o,w}(u)) refers solubility of gas o in cluster w.
Here, the gas o ability is,
where, (X_{best}(u)) signifies fitness of best gas in the whole system, and (mu) is constant where (mu =0.05).
Here, the temperature is given as,
where, (u_{max})refers total number of iterations.
The aforementioned equation can be rewritten as,
According to CSO23, the group-mate roosters are followed for searching food. In addition, they also arbitrary steal the good food discovered by other chickens and this process is given by,
where, Rand signifies a uniform arbitrary number amongst 0 to 1, (r_1) refers rooster index, (r_2) symbolizes chicken index, (M_{o,w}(u)) indicates current position of hen, (K_1) and (K_2) represent the random numbers, (M_{r_1,w}(u)) represent the position of rooster and (M_{r_2,w}(u)) indicate the position of chicken.
Substituting equation (19) in equation (17),
The final update of HGCSO is shown as,
Step 8: Local optima of escape:
Here, the step is used to escape from local optima. The rank and choose count of worst agents as,
where, (ell) symbolize count of search agents.
Step 9: Update worst search agents position
where, ({rho }_{(o,w)}) refers position of gas oin cluster w, e signifies random number, and ({rho }_{min}) and (rho _{max}) are limits of difficulty.
Step 10:Fitness re-computation: The fitness is re-evaluated wherein the solution linked to the highest fitness is employed in routing.
Step 11: Terminate: The optimum weights are generated in a repeated manner until the highest iteration is accomplished.
Pseudo code HGCSO
Flowchart of HGCSO.
The classification of plant disease is done with CHGCSO-based DRN. Consider a dataset J having í rice plant images, and is expressed as,
where, (I_{ell }) signifies ({ell }^{th}) input image, and i symbolize total images.
Here, the group of input images J is adapted as input for the median filter. Median filter (https://towardsdatascience.com/image-processing-class-egbe443-4-filters-aa1037676130) is a smoothing method that reduces the edge blurring in which the notion is to replace the current point of the image with median. Furthermore, the median filter assists to discard the impulse noise. It is represented as,
where, (I(d+k,e+j)) symbolize corresponding image element,(k, j) signifies position of image, and K refers moving region. The pre-processed output is expressed as P.
Once the pre-processing is done, the extraction of features is done for effective plant disease classification. P is adapted as input for extracting features. Here, the features, like HoG, LTP, SLIF, and statistical features are adapted for further analysis. The feature vector is expressed as,
where, (C_1) symbolizes HoG feature, (C_2) represents a mean feature, (C_3) signifies variance feature, (C_4) refers standard deviation feature, (C_5) is skewness, (C_6) is kurtosis feature, (C_7) is SLIF, (C_8) is LTP feature.
The classification of plant disease is done with CHGCSO-based DRN.
a) Training of DRN with CHGCSO algorithm
The CAViaR10 model indicates the evolution of quantile in contrast to time considering an autoregressive procedure, and evaluates the attributes with respect to regression quantiles. This statistical method helps compute the potential loss over a certain time period, based on historical records. The HGCSO is formulated by combining HGSO and CSO, leveraging the benefits of both algorithms.
Step 1: Initialization: The solution initialization is formulated as,
where,(eta) expresses total solution and (T_{mu }) symbolizes (mu ^{th})solution.
Step 2 Discovery of Error: The optimal solution is found using the error function, which is formulated as a minimization problem, as shown:
where, (K_h) signifies expected output, and O denotes output of classifier, g symbolizes entire data samples, such that (1 < hle g).
Step 3: Find update solutions:
To generate global optimum solution, the proposed HGCSO is adapted, which is given as,
Subtract (M_{o,w}(u)) on both sides,
As per CaViaR10, the update equation is given by the current and past observation and is expressed as,
where,(delta) represents a vector of unknown parameter, (M_{o,w}(u)) represents current autoregressive quantile, and (M_{o,w}(u-1)) is the previous autoregressive quantile, f[.]represents fitness of solution, and w is constant.
Assume (chi =rho =2),
After substituting equation (33) in equation (30) in RHS, the final update of CHGCSO is given as,
where,
where Rand signifies an arbitrary number between 0 to 1, (K_1) and (K_2) represent the random numbers, Yrefers constant, u indicates iteration time, (M_{o,best}(u)) indicates best gas o in cluster w, (M_{r_1,w}(u)) represent the position of the rooster and (M_{r_2,w}(u)) indicate the position of chicken, (M_{best}(u)) denotes optimum gas in swarm, (beta) implies the capability of gas o, and (gamma) denote impact of other gases on gas o in cluster w, and X denote flag that alters search agent direction and (R_{o,w}(u)) refers solubility of gas o in cluster w, (delta) implies an unknown parameter vector, (M_{o,w}(u)) denotes present autoregressive quantile, and (M_{o,w}(u-1)) is the preceding autoregressive quantile, f[.] represents solution fitness.
Step 4 Evaluation of Best Solution: After recomputing the error for each solution, the one showing the least error is selected as the optimal solution.
Step 5 Termination: The procedure is repeated until the optimal disease detection solution is obtained, and the process is terminated after reaching the maximal iteration (u_{max}) .
The efficiency of CHGCSO+DRN is evaluated by varying training data using 100 and 200 nodes.
The CHGCSO+DRN are implemented on Matlab and Windows 10 OS having 2GB RAM and Intel Core processor.
The rice leaf diseases dataset (https://archive.ics.uci.edu/ml/datasets/Rice+Leaf+Diseases) comprises three classes of diseases that include Brown spot, bacterial leaf blight, and Leaf smut wherein each class contains 40 images. The dataset is multivariate, featuring integer attributes, with 120 instances and a total of 57,566 web hits. The dataset was created by manually dividing infected leaves into distinct classes based on disease type.
Figure 5 displays the experimental outcomes of CHGCSO+DRN. Here, the input rice leaf image is revealed in Figure 5a). The pre-processed image is revealed in Figure 5b). The HoG feature extracted image is depicted in Figure 5c). The LTP feature extracted image is displayed in Figure 5d).
Experimental outcomes of developed CHGCSO+DRN with (a) Input Image (b) Pre-processed Image (c) HoG feature Extracted Image (d) LTP Feature Extracted Image.
Sensitivity, energy, accuracy, F1-score, specificity and ROC are the considered evaluation metrics in this model.
The techniques taken for the assessment include SCA based RideNN24, RF+SVM+LR36, DCNN28, Residual network (ResNet)14, MCFN45, ESRGAN42, and CHGCSO+DRN.
a) Evaluation utilizing 100 nodes
Comparative of methods with 100 nodes using (a) Testing accuracy (b) Sensitivity (c) Specificity (d) F1-score (e) ROC.
Figure 6 exposes comparative analysis with 100 nodes. The accuracy results are shown in Figure 6(a). For 90% of data, the testing accuracy measured by SCA based RideNN, RF+SVM+LR, DCNN, ResNet, MCFN, ESRGAN, and CHGCSO+DRN are 0.784, 0.767, 0.776, 0.826,0.784, 0.767, and 0.868. The sensitivity result is presented in Figure 6(b). For 90% of training data, the sensitivity analyzed by SCA based RideNN, RF+SVM+LR, DCNN, ResNet, MCFN, ESRGAN, and CHGCSO+DRN are 0.790, 0.775, 0.735, 0.769, 0.741, 0.756, and 0.846. The specificity result is shown in Figure 6(c). For 90% of training data, the specificity analyzed by SCA based RideNN, RF+SVM+LR, DCNN, ResNet, MCFN, ESRGAN, and CHGCSO+DRN is 0.796, 0.786, 0.814, 0.819, 0.776, 0.826, and 0.888. The review with f1-score is displayed in Figure 6(d). For 90% of training data, the f1-score evaluated by SCA based RideNN, RF+SVM+LR, DCNN, ResNet, MCFN, ESRGAN, and CHGCSO+DRN are 0.773,0.757, 0.765, 0.815, 0.773, 0.757, and 0.855. The ROC analysis is displayed in Figure 6(e). When FPR is 10, the TPR of SCA based RideNN, RF+SVM+LR, DCNN, ResNet, MCFN, ESRGAN, and CHGCSO+DRN are 1, 1, 1, 1, 1, and 1. The values of AUC are 0.5098, 0.5207, 0.5274, 0.5396, 0.541, and 0.5575.
b) Evaluation utilizing 150 nodes
The estimation of techniques with 150 nodes is exposed in Figure 7. The accuracy results are provided in Figure 7(a). When training data is 50%, the accuracy measured by SCA based RideNN, RF+SVM+LR, DCNN, ResNet, MCFN, ESRGAN, are 0.598, 0.606, 0.649, 0.694, 0.774, 0.796, and the CHGCSO+DRN is 0.774. The sensitivity result is given in Figure 7(b). For 50% of training data, the sensitivity created by SCA based RideNN, RF+SVM+LR, DCNN, ResNetMCFN, ESRGAN, are 0.600, 0.612, 0.633, 0.665,0.796, 0.815, and the CHGCSO+DRN is 0.750. The specificity result is shown in Figure 7(c). For 50% of training data, specificity obtained by SCA based RideNN, RF+SVM+LR, DCNN, ResNet, MCFN, ESRGAN, are 0.655, 0.660, 0.686, 0.704, 0.816, 0.825, whereas the CHGCSO+DRN is 0.752. The f1-score analysis is shown in Figure 7(d). For 90% of training data, the f1-score evaluated by SCA based RideNN, RF+SVM+LR, DCNN, ResNet, MCFN, ESRGAN, and CHGCSO+DRN are 0.815, 0.798, 0.806, 0.859, 0.763, 0.785, and 0.902. The ROC analysis is displayed in Figure 7(e). When FPR is 10, the TPR of SCA based RideNN, RF+SVM+LR, DCNN, ResNet, MCFN, ESRGAN, and CHGCSO+DRN are 1,1, 1, 1, 1, and 1. The values of AUC are 0.5128, 0.5287, 0.5354, 0.5446, 0.5653, and 0.5885.
Comparative with 150 nodes using (a) Accuracy (b) Sensitivity (c) Specificity (d) F1-score (e) ROC.
c) Evaluation utilizing 200 nodes
Assessment with 200 nodes using (a) Accuracy (b) Sensitivity (c) Specificity (d) f1-score (e) ROC.
Figure 8 shows estimate of methods accompanying 200 nodes. The accuracy result is revealed in Figure 8(a). For 90% of training data, the accuracy generated by SCA based RideNN is 0.933, RF+SVM+LR is 0.831, DCNN is 0.854, ResNet is 0.867, MCFN is 0.825, ESRGAN is 0.816, and CHGCSO+DRN is 0.943. The sensitivity result is conferred in Figure 8(b). For 90% of training data, the sensitivity obtained by SCA based RideNN is 0.889, RF+SVM+LR is 0.883, DCNN is 0.904, ResNet is 0.860, MCFN is 0.816, ESRGAN is 0.825, and CHGCSO+DRN is 0.933. The evaluation of specificity is presented in Figure 8(c). For 90% of training data, the specificity generated by SCA located RideNN is 0.831, RF+SVM+LR is 0.747, DCNN is 0.749, ResNet is 0.912, MCFN is 0.825, ESRGAN is 0.816, and CHGCSO+DRN is 0.920. The f1-score estimation is given in Figure 8(d). For 90% of training data, the f1-score judged by SCA located RideNN, RF+SVM+LR, DCNN, ResNet, MCFN, ESRGAN, and CHGCSO+DRN are 0.920, 0.820, 0.842, 0.855, 0.814, 0.805, and 0.930. The ROC analysis is displayed in Figure 8(e). When FPR is 10, the TPR of SCA based RideNN, RF+SVM+LR, DCNN, ResNet, MCFN, ESRGAN, and proposed CHGCSO+DRN are 1,1, 1, 1, 1, and 1. The values of AUC are 0.5245, 0.5347, 0.5524, 0.5696, 0.5815, and 0.6035.
Table 2 represents the analysis based on k-fold cross validation by considering 100, 150, and 200 nodes based on all evaluation metrics. Here, the performance is increased when increasing the k-fold and the maximum results are obtained at k-fold=9.
Figure 9shows the analysis based on routing. In this research, the routing performance of the devised model is compared with other routing techniques, such as Competitive Shuffled Shepherd Optimization (CSSO)12, Sunflower EarthWorm (S-EWA)41, and Novel Intelligent Robust Optimized Dynamic and Efficient Cluster (NIRODEC) protocol6. The residual energy metric is considered for the evaluation and Figures 9a), 9b), and 9c) denote the assessment based on 100, 150, and 200 nodes respectively. When the number of rounds 1000, the residual energy obtained by the CSSO, S-EWA, NIRODEC protocol, and the devised scheme is 0, 0, 0, and 0.033 for 100 nodes, 0.001, 0.001, 0.001, and 0.046 for 150 nodes, and 0.006, 0.009, 0.011, and 0.059 for 200 nodes, respectively.
Assessment based on routing (a) 100 nodes, (b) 150 nodes, and (c) 200 nodes.
Table 3 shows the ablation study of the CHGCSO+DRN. Here, the performance of the CHGCSO+DRN is compared with each individual component and the performance is high for CHGCSO+DRN for all evaluation metrics. The evaluation is done by considering different nodes.
Table 4 depicts the evaluation accompanying 100, 150 and 200 nodes. Here, we discuss the best results. Utilizing 200 nodes, the very high evaluating accuracy of 0.943, specificity of 0.920, sensitivity of 0.933, and F1-score of 0.930 is calculated by the CHGCSO+DRN. The reasons for the better performance of the devised approach are discussed below: In the devised approach, the HGCSO is devised by combining HGSO and CSO, which inherit the advantages of both algorithms. Here, the HGSO is effective in addressing the engineering design issues whereas the CSO can acquire enhanced optimization outcomes in both robustness and accuracy. The proposed CHGCSO is developed by combining the CAViaR model and HGCSO is used to train DRN. The CAViaR model indicates the evolution of quantile in contrast to time considering an autoregressive procedure and evaluates the attributes with respect to regression quantiles. It is a statistical method, which is utilized to compute the quantity of potential loss that could occur over a particular time period using past records. Thus, the effectiveness of the devised approach is optimum than the conventional approaches.
Table 5 displays the analysis based on computation time. Here, when comparing the existing methods, like, SCA based RideNN, RF+SVM+LR, DCNN, MCFN, ESRGAN, DRN, ResNet, and the CHGCSO+DRN attains minimum computation time.
A novel model has been developed for plant disease classification in an IoT network. The process begins with the simulation of the IoT network to monitor crop diseases. Routing is performed using the HGCSO algorithm, where the fitness function incorporates energy, delay, distance, and LLT. At the BS, plant disease classification is conducted by collecting images of rice crops. Each rice crop image undergoes preprocessing, with median filtering applied to eliminate noise. Then, feature extraction is performed to identify key characteristics. Finally, plant disease classification is achieved using a Deep Recurrent Network (DRN), where the DRN is trained with the CHGCSO algorithm. The CHGCSO is a hybrid model combining the CAViaR model and HGCSO. The CHGCSO-based DRN outperforms existing methods, achieving a maximum accuracy of 94.3%, sensitivity of 93.3%, specificity of 92%, and an F1-score of 93%. In future, the applicability of the model will be extended beyond rice crops to support disease detection in a variety of plant types, enabling broader usage in agriculture. Also, the test will be conducted under different environmental conditions and IoT network configurations to evaluate and enhance the reliability of the model and accuracy across diverse real-world scenarios.
The data and code supporting this study’s findings are available from the corresponding author upon reasonable request.
Abayomi-Alli, O. O., Damaševičius, R., Misra, S. & Maskeliūnas, R. Cassava disease recognition from low-quality images using enhanced data augmentation model and deep learning. Expert Systems 38, e12746 (2021).
Article MATH Google Scholar
Almadhor, A. et al. AI-driven framework for recognition of guava plant diseases through machine learning from DSLR camera sensor based high resolution imagery. Sensors 21, 3830 (2021).
Article ADS PubMed PubMed Central MATH Google Scholar
Argüeso, D. et al. Few-Shot Learning approach for plant disease classification using images taken in the field. Computers and Electronics in Agriculture 175, 105542 (2020).
Article MATH Google Scholar
Balachandra, M., Prema, K. & Makkithaya, K. Multiconstrained and multipath QoS aware routing protocol for MANETs. Wireless networks 20, 2395–2408 (2014).
Article Google Scholar
Chandanapalli, S. B., Reddy, E. S. & Lakshmi, D. R. Convolutional neural network for water quality prediction in WSN. Journal of Networking and Communication Systems 2, 40–47 (2019).
MATH Google Scholar
Chavan, V. G., Patil, P. R., Potgantwar, A. D. & Bhaladhare, P. R. A Novel Intelligent Efficient and Dynamic Cluster Based Routing Protocol for IoT Assisted Precision Agriculture. Journal of Algebraic Statistics 13, 4025–4037 (2022).
MATH Google Scholar
Chen, M.-R., Yang, L.-Q., Zeng, G.-Q., Lu, K.-D. & Huang, Y.-Y. IFA-EO: An improved firefly algorithm hybridized with extremal optimization for continuous unconstrained optimization problems. Soft Computing 27, 2943–2964 (2023).
Article MATH Google Scholar
Chen, Z., Chen, Y., Wu, L., Cheng, S. & Lin, P. Deep residual network based fault detection and diagnosis of photovoltaic arrays using current-voltage curves and ambient conditions. Energy Conversion and Management 198, 111793 (2019).
Article MATH Google Scholar
Chen, Z., He, M., Liang, W. & Chen, K. Trust-aware and low energy consumption security topology protocol of wireless sensor network. Journal of Sensors 2015, 716468 (2015).
Article MATH Google Scholar
Engle, R. F. & Manganelli, S. CAViaR: Conditional autoregressive value at risk by regression quantiles. Journal of business and economic statistics 22, 367–381 (2004).
Article MathSciNet MATH Google Scholar
Fausto, F., Cuevas, E. & Gonzales, A. A new descriptor for image matching based on bionic principles. Pattern Analysis and Applications 20, 1245–1259 (2017).
Article MathSciNet MATH Google Scholar
Gunisetti, L., Koduri, S. B. & Jagannathan, V. Optimized deep learning system for smart maize leaf disease detection in IoT platform via routing algorithm. Multimedia Tools and Applications 82, 13533–13555 (2023).
Article MATH Google Scholar
Hashim, F. A., Houssein, E. H., Mabrouk, M. S., Al-Atabany, W. & Mirjalili, S. Henry gas solubility optimization: A novel physics-based algorithm. Future Generation Computer Systems 101, 646–667 (2019).
Article MATH Google Scholar
Hu, W.-J. et al. MDFC-ResNet: An agricultural IoT system to accurately recognize crop diseases. IEEE Access 8, 115287–115298 (2020).
Article MATH Google Scholar
Khirade, S. D., & Patil, A. B. In 2015 International conference on computing communication control and automation. (IEEE, 2015), pp. 768-771.
Kulkarni, O. in 2018 fourth international conference on computing communication control and automation (ICCUBEA). (IEEE, 2018), pp. 1-4.
Kundu, N. et al. IoT and interpretable machine learning based framework for disease prediction in pearl millet. Sensors 21, 5386 (2021).
Article ADS PubMed PubMed Central MATH Google Scholar
Lakshmi, N. D., Latha, Y. M., & Damodaram, A. Silhouette extraction of a human body based on fusion of HOG and graph-cut segmentation in dynamic backgrounds. (2013).
Lian, J. et al. The educational competition optimizer. International Journal of Systems Science 55, 3185–3222 (2024).
Article MATH Google Scholar
Maratha, P. & Gupta, K. HFLBSC: Heuristic and fuzzy based load balanced, scalable clustering algorithm for wireless sensor network. Wireless Personal Communications 125, 281–304 (2022).
Article Google Scholar
Maratha, P. & Gupta, K. Linear optimization and fuzzy-based clustering for WSNs assisted internet of things. Multimedia Tools and Applications 82, 5161–5185 (2023).
Article MATH Google Scholar
Maratha, P., Gupta, K. & Luhach, A. K. Improved fault-tolerant optimal route reconstruction approach for energy consumed areas in wireless sensor networks. IET Wireless Sensor Systems 10, 112–116 (2020).
Article Google Scholar
Meng, X., Liu, Y., Gao, X., & Zhang, H. In Advances in Swarm Intelligence: 5th International Conference, ICSI 2014, Hefei, China, October 17-20, 2014, Proceedings, Part I 5. (Springer, 2014), pp. 86-94.
Mishra, M., Choudhury, P. & Pati, B. Modified ride-NN optimizer for the IoT based plant disease detection. Journal of Ambient Intelligence and Humanized Computing 12, 691–703 (2021).
Article Google Scholar
Moraru, A., Pesko, M., Porcius, M., Fortuna, C. & Mladenic, D. Using machine learning on sensor data. Journal of computing and information technology 18, 341–347 (2010).
Article Google Scholar
Ni, L. et al. Pi puf: a processor-intrinsic puf for iot. Computers and Electrical Engineering 105, 108540 (2023).
Article MATH Google Scholar
Peter, A. P. et al. Artificial intelligence model for monitoring biomass growth in semi-batch Chlorella vulgaris cultivation. Fuel 333, 126438 (2023).
Article CAS MATH Google Scholar
Picon, A. et al. Deep convolutional neural networks for mobile capture device-based crop disease classification in the wild. Computers and Electronics in Agriculture 161, 280–290 (2019).
Article MATH Google Scholar
Picon, A. et al. Crop conditional Convolutional Neural Networks for massive multi-crop plant disease classification over cell phone acquired images taken on real field conditions. Computers and Electronics in Agriculture 167, 105093 (2019).
Article MATH Google Scholar
Qi, A. et al. FATA: an efficient optimization method based on geophysics. Neurocomputing 607, 128289 (2024).
Article MATH Google Scholar
Ramesh, S. & Rajaram, B. Iot based crop disease identification system using optimization techniques. ARPN J. Eng. Appl. Sci 13, 1392–1395 (2018).
MATH Google Scholar
Rehman, Z. U. et al. Recognizing apple leaf diseases using a novel parallel real-time processing framework based on MASK RCNN and transfer learning: An application for smart agriculture. IET Image Processing 15, 2157–2168 (2021).
Article MATH Google Scholar
Ruby, R. et al. Enhancing secrecy performance of cooperative NOMA-based IoT networks via multiantenna-aided artificial noise. IEEE Internet of Things Journal 9, 5108–5127 (2021).
Article MATH Google Scholar
Sabrol, H., & Satish, K. In 2016 international conference on communication and signal processing (ICCSP). (IEEE, 2016), pp. 1242-1246.
Shi, B. et al. Prediction of recurrent spontaneous abortion using evolutionary machine learning with joint self-adaptive sime mould algorithm. Computers in Biology and Medicine 148, 105885 (2022).
Article PubMed MATH Google Scholar
Shibani, K., Kumar, K. Sendhil, & Shanmugam, G. Siva. In Artificial Intelligence and Evolutionary Computations in Engineering Systems. (Springer, 2020), pp. 143-154.
Shrivastava, V. K. & Pradhan, M. K. Rice plant disease classification using color features: a machine learning paradigm. Journal of Plant Pathology 103, 17–26 (2021).
Article MATH Google Scholar
Su, H. et al. RIME: A physics-based optimization. Neurocomputing 532, 183–214 (2023).
Article MATH Google Scholar
Tan, X. & Triggs, B. Enhanced local texture feature sets for face recognition under difficult lighting conditions. IEEE transactions on image processing 19, 1635–1650 (2010).
Article ADS MathSciNet PubMed MATH Google Scholar
Unni, R. et al. Enhancement of mechanical and thermal properties of Ixora coccinea L. plant root derived nanocellulose using polyethylene glycol-glutaraldehyde system. Chemosphere 298, 134324 (2022).
Article CAS PubMed MATH Google Scholar
Vimala, S., Merlin, N. Gladiss, Ramanathan, L., & Cristin, R. Optimal routing and deep regression neural network for rice leaf disease prediction in IoT. International Journal of Computational Methods 18, 2150014 (2021).
Wen, J., Shi, Y., Zhou, X. & Xue, Y. Crop disease classification on inadequate low-resolution target images. Sensors 20, 4601 (2020).
Article ADS PubMed PubMed Central MATH Google Scholar
Yuan, C. et al. Polar lights optimizer: Algorithm and applications in image segmentation and feature selection. Neurocomputing 607, 128427 (2024).
Article MATH Google Scholar
Yuan, C. et al. Artemisinin optimization based on malaria therapy: Algorithm and applications to medical image segmentation. Displays 84, 102740 (2024).
Article CAS MATH Google Scholar
Zhao, Y. et al. An effective automatic system deployed in agricultural Internet of Things using Multi-Context Fusion Network towards crop disease recognition in the wild. Applied Soft Computing 89, 106128 (2020).
Article MATH Google Scholar
Download references
P. Maratha, M. A. Shah: these authors contributed equally to this work.
Department of Computer Science & Applications, Maharshi Dayanand University, Rohtak, 124001, India
Ashish Saini, Nasib Singh Gill & Preeti Gulia
Department of Computer Science & Information Technology, Central University of Haryana, Mahendragarh, 123031, India
Anoop Kumar Tiwari & Priti Maratha
Department of Economics, Kardan University, Kabul, Afghanistan
Mohd Asif Shah
Centre for Research Impact & Outcome, Chitkara University Institute of Engineering and Technology, Chitkara University, 140401, Rajpura, Punjab, India
Mohd Asif Shah
Chitkara Centre for Research and Development, Chitkara University, Baddi, Himachal Pradesh, 174103, India
Mohd Asif Shah
Division of Research and Development, Lovely Professional University, Phagwara, Punjab, India
Mohd Asif Shah
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
Ashish Saini: Conceptualization, Problem formulation, Methodology, Original draft preparation, Reviewing and Editing, and Final drafting. Nasib Singh Gill: Data curation, Programming, Simulation, Validation. Preeti Gulia: Numerical analysis, Visualization, and System set-up. Anoop Kumar Tiwari: Mathematical Modelling, Visualization, and Investigation. Priti Maratha: Supervision, Problem formulation, Programming, Validation, Writing, Reviewing, and Editing. Mohd Asif Shah: Conceptualization, Supervision, Validation, Reviewing, Ensuring Ethical Standards.
Correspondence to Priti Maratha or Mohd Asif Shah.
The authors declare no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
Reprints and permissions
Saini, A., Gill, N.S., Gulia, P. et al. Smart crop disease monitoring system in IoT using optimization enabled deep residual network. Sci Rep 15, 1456 (2025). https://doi.org/10.1038/s41598-025-85486-1
Download citation
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-025-85486-1
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
Advertisement
© 2025 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.
About The Author
More Stories
Zombie IXPs: The Four Types of Exchanges That Refuse to Die, but Fail to Live
The Shift in Peering Threatening the Internet’s Foundations
Remembering Alan Barrett: A Builder of the African Internet