IoT-Enabled Pest Identification and Classification with New Meta-Heuristic-Based Deep Learning Framework

Abstract The pest and insect affected crop is an important concern to cause damage to the agricultural sector. While identifying the pest in the crop, the camera placement is not supported in an inconsistent manner to capture the pest images. Hence, certain Internet of Things (IoT) devices are used to catch the pest images with its corresponding agriculture based sensor, yet it also faces some limitations to provide the accurate results. In order to alleviate the problem, an IoT-assisted pest identification and classification method is proposed. Initially, the IoT sensors are used to collect the required images. Subsequently, the input images are used to perform the object detection phase that is accomplished by the Yolov3, where the pest is detected significantly. Further, the detected images are fed into the model of “Convolutional Neural Network (CNN),” in which the deep features are fetched and finally given as input to the classifier model of “Convolution Neural Long Short-Term Memory (CNLSTM),” in turn some hyper parameters are optimally tuned by “Adaptive Honey Badger Algorithm (AHBA).” Hence, the experimental results prove that the recommended method achieves the better performance in terms of diverse metrics.


Introduction
Identifying and classifying crop pest is one of the vital challenges in the field of agriculture (Khanramaki, Asli-Ardeh, and Kozegar 2021).Insects are the most important factor for causing damages to the crops and reducing the crop productivity.Insect classification is known as the complex task owing to its complex structure and also with the high similarity in appearance among the distinct species (Turkoglu et al. 2022a).It is very essential for recognizing and classifying the insects presents in the crops at earlier stage by providing highly effective pesticides and also with biological control methods for preventing the spread of insects that causes crop diseases (Wang et al. 2020a;Kasinathan and Uyyala 2021).The traditional way of identifying insects is considered to be inefficient, consumption of time, and labor-intensive.The vision-aided computerized technique of image processing has been implemented based on machine learning for accurately performing the classification and identification of insects for overcoming the conventional problems in the field of agriculture research.
IoT is the well-known revolutionary technology for future communications and computing.The people in the world are highly dependent on agriculture for income and food resources.Therefore, the intelligent Information Technology (IT) technologies are necessary for overcoming the challenges of traditional agricultural approaches.IoT makes easier for the farmers in the agriculture by providing a lot of techniques to achieve sustainable and precise agriculture production for facing the challenges in the agricultural field (Ayaz et al. 2019).IoT technology supports the farmers to collect the information regarding the natural scenarios like soil fertility, temperature, moisture, and weather.Crop online monitoring system enhances the agricultural productivity and crop growth, finds the animal intrusion into field and useful for detecting the weed, pest, and water level (Chen et al. 2021a).IoT enables the farmers to monitor their agricultural field anywhere in the world at any time.The main intention of the IoT is to extend the network by concatenating various kinds of connected devices (Ai et al. 2020;Chen et al. 2021b).IoT mostly focused on three perspectives like cost saving, automation, and communication in the system.IoT encourages the people for pursuing their routine activities based on the internet and saving the cost and time for making the productive (Cheng et al. 2017).
An automated insect identification system is implemented to interpret seven geometrical features and also utilized deep learning and machine learning algorithms for obtaining the better results by considering less number of insect classes (Ayan, Erbay, and Varc ¸ın 2020;Xie et al. 2018).When involving the machine learning algorithms, the classification accuracy is mainly relied on the structure of the extracted features.Here, the optimal features are chosen for the machine learning, which maximizes the complexity in computation (Esgario, Tassis, and Krohling 2021).In addition, the accuracy needs to be improved by incorporating the deep learning algorithms for categorizing the huge image datasets (Karar et al. 2021).Deep learning algorithm is used for performing the automated feature extraction by using the raw data that decreases the challenges of the hand-crafted features and also for addressing the highly complex issues related to image classification (Rahman et al. 2020).Recently, the deep learning approaches are investigated with the help of Convolutional Neural Networks (CNNs) that ensure promising solutions for the existing challenges (Tetila et al. 2020).When compared to machine learning techniques, deep learning techniques are helpful in automatically obtaining the representative features (Kong et al. 2022) from the training dataset, which avoids complex image processing steps and labor-intensive feature engineering for satisfying diverse outdoor conditions (Rustia et al. 2021;Wang et al. 2020b).An efficient deep CNN network is developed for classifying the insect species of field-crop insect images, which provides higher accuracy in classification accuracy.Thus, deep learning techniques have tendency to detect the pests in practical applications.Recently, farmland resources are declined faster and sudden surges of unpredictable natural calamities like flooding, global warming and so on, which makes food scarcity issue to derive the smart agriculture.Owing to the evolution of technology and science, the recent prerequisite for breakthrough outcomes achieves the high productivity and more efficiency that tends to IoT adoption in agriculture sector (Quy et al. 2022).It leads to the new advent of IoT-based smart agriculture.Hence, it is significant for developing a new IoT-enabled pest identification and classification model with the help of deep learning approach.
The core offerings of the research works are given as follows: To develop a new IoT-based pest identification and classification model for accurately detecting the pest in the crop field and reducing their effects at the earlier stage for improving the crop production in the agricultural areas.
To integrate an enhanced deep architecture named CNLSTM for finding the deep features from the YOLOv3-based detected images and classifying the extracted features along with the parameter optimization using the suggested AHBA to identify the type of pest present in the images.
To introduce an improved meta-heuristic algorithm named AHBA for parameter optimization of hidden neurons in LSTM to elevate the performance of the optimal pest classification in the proposed model.
To examine the suggested pest identification and classification model with existing meta-heuristic algorithms and classifiers with diverse performance measures.
The rest of the research paper is elucidated as follows.The Section 2 explains the existing studies of pest classification and its challenging issues.In Section 3, the proposed pest identification and classification model is depicted.In Section 4, dataset collection and YOLOv3-based pest detection is carried out.In Section 5, the extracted deep features for CNLSTM-based classification and the proposed algorithm are explored.In Section 6, the experimental results and comparative analysis are given.In Section 8, the developed pest identification and classification model is summarized.

Literature Survey
2.1.Related Works Bhoi et al. (2021) have introduced an enhanced model for identifying the pests that were affecting the rice at the time crop productivity.Here, the IoT-based mechanism was used for passing the rice pest images to the cloud storage and has provided the pest information.When the pest was identified, the information regarding the presence of pest was sent to the farmers or owners for taking respective actions.The simulated results have illustrated that the enhanced approach has minimized the rice wastage at the productivity field through the continuous monitoring of the pests in the rice field.Chen et al. (2020) have implemented a deep learning-based model for obtaining the insect locations and also for analyzing the environmental information from the weather stations for getting the pests information in the field with the help of enhanced deep learning approach.The experimental results have revelaed that the proposed approach have secured better identification accuracy.Precise identification of the insects and pest has decreased the amount of pesticides usage that has also minimized the pesticide damage over soil.Turkoglu et al. (2022b) have presented two types of classification approaches with the help of deep feature extraction that were obtained from the pretrained CNN.The proposed model was validated with the help of diverse diseases and pest images.It was observed that the accuracy scores were better with the majority of the ensemble model and has provided improved performance than the existing models.In 2019, Liu et al. (2019) have developed an end-to-end method for classifying and detecting the huge multi-class pests with the help of deep learning.The three major part of proposed framework were novel module with attention-based approach, the developed neural network for ensuring the region proposals, and a score map for classifying the pest and bounding box regression.The performance analysis was done for demonstrating the effectiveness of the multi-class pest detection through the proposed model.Li et al. (2021) have involved the novel technique for enhancing the accuracy of the small pest detection.The suggested framework was trained with the help of transfer learning methodology using the tiny pest training set.Here, the developed deep learning architecture has provided better performance than other approaches and the analysis have demonstrated that the proposed method has ensured the robust performance toward detecting the tiny pests at varied light reflections and pest densities.In 2018, Yue et al. (2018) have proposed an enhanced residual-based network for detecting the pest in the crop field.The proposed method was correlated with the traditional approaches and has demonstrated the high powerful capacity of the developed model toward the image reconstruction.The analysis results have explored that the proposed approach has revealed an enhanced recall rate for the pest detection.Wang et al. (2021) have developed an efficient deep learning model in the pest monitoring system for automatically detecting and counting the pest in the rice planthoppers.Here, the proposed approach was developed to extract the high-quality regions in the pest images even extracts the tiny regions.The analysis results have shown that the suggested approach has shown improved recognition recall when compared with the stat-of-the-arts approaches.Thenmozhi and Reddy (2019) have implemented an elevated deep learning network for classifying the insect species through the three available datasets.The suggested approach was validated with the other deep learning architectures under the insect classification.Further, this model has included the transfer learning for tuning the pr-trained models.The final outcome has elucidated that the suggested model was effective in classifying the different types of insects and also for applicable in the agricultural sector for crop protection.

Problem Statement
Plant pests are the most important factor for causing the huge loss in the agricultural production along with the social, ecological and economical implications.It is essential for recognizing and classifying the insects present in the crops at the initial stage.This is to avoid the insect spread into the crop that results in crop diseases by choosing the efficient biological control and pesticides approaches.Numerous features and challenges of Agriculture pest detection are reviewed in Table 1.Artificial Intelligence (Bhoi et al. 2021) decreases the rice wastage at the time of production through monitoring the pests at the regular interval of time.However, the advanced technique needs for achieving the higher performance.YOLOv3, LSTM (Chen et al. 2020) reduces the damages that are caused in the environment by involving enormous usage of pesticides and also increases the crop quality.Yet, there is a requirement for enhancing the perspectives in the images to solve the issues related to insufficient training samples.Ensemble learning (Turkoglu et al. 2022b) ensures high robustness.But, it cannot handle the imbalance problem of training data.CNN, Channel-spatial attention (Liu et al. 2019) are more robust for detecting the tiny pests on image.On the other hand, the size of the sample requires to be maximized in diverse external scenarios for gaining better results.TPest-RCNN (Li et al. 2021) helps to improve the performance by maximizing the replacement frequency of traps.However, this model technically critical as it confines with computer-vision.Deep CNN (Yue et al. 2018) is used for The remote monitoring mechanism with IoT devices are the vital technique involved in diverse applications like object tracking in smart cities, healthcare, human surveillance, modern farming and etc.By assisting the IoT, the insect control can be administered and controlled from anywhere in the world.In Rustia et al. (2019), the IoT network is used along with the wireless imaging system for developing the remote greenhouse pest monitoring system.Here, the blob counting and k-means clustering approach are used in the imaging system for automatically counting the insects and pests present in the trap sheet.Similarly, in another research, the IoT-based smart farm field management methodology is implemented for monitoring the crop growth, detecting (Tabjula et al. 2021a) the insects in the crop field and also for determining the appropriate pesticide to manage the crop pests.Yet, the automated identification of the insects and pests are the major challenges in the pest monitoring systems.Hence, the machine and deep learning algorithms are used for the decision-making and object detection in diverse insect control mechanisms.In Gutierrez et al. (2019), the effective performance of the machine learning, deep learning, and computer vision algorithms are utilized for the pest detection especially in the tomato farms.This study has shown that the deep learning architecture provides enhanced performance when compared among three considerations.From the literature works, it is clear that IoT is more essential for the pest monitoring systems, where the deep learning techniques are known to be optimal approach for detecting and classifying the insects and pests from the crop images.Thus, IoT and deep learning are combined into the pest identification and classification model for offering more benefits to the farmers and the architecture has been diagrammatically represented in Figure 1.
A novel classification model for pest identification with IoT by deep learning architecture is developed for identifying and classifying the pests in the crops for reducing the usage of fertilizers and increasing the crop production by preventing the pests at earlier stage of the crop growth.The IoT technology is used for collecting the required crop images from the agricultural fields through the sensors.These garnered images are considered for the object detection phase, where the YOLOv3 detector is used to detect the pest regions in the given input images.The detected images from the YOLOv3 are given to the CNLSTM network, where the CNN framework is used for retrieving the most essential features of the pest detected images.The pest features are passed toward the developed LSTM network, where the LSTM is deployed for classifying the feature into distinct pest classes in the agricultural field.The performance of classification model is further enhanced by the proposed AHBA on conducting the optimization in the hidden neurons of LSTM network, which intends to achieve the maximization of accuracy in the classification phase in the proposed optimal pest classification model.

IoT-Enabled Pest Detection and Smart Agriculture
Agriculture is defined to be a science of crop production, animal husbandry and soil cultivation, where the resultant products need to be marketed in effective manner.The food demand has been increasing abundantly both in the qualitative and quantitative aspects, which can be satisfied by incorporating the computer technology into the agricultural practices.Owing to the population growth of world, it is also mandatory to achieve the high productivity regarding the requirements of food in terms of nutrition.The crop production can be mainly affected by various diseases (Bojja et al. 2020) that are caused due to certain factors like the presence of insects and pests into the crop field.It needs to be prevented for enhancing the crop production by reducing the disease-affected crop in the agricultural areas.Hence, the IoT technology is required for automatically detecting the insects and pests through the IoT sensor devices, which helps the farmers to monitor and remove the pest that affects the crops at the earlier stage of the crop growth.In the traditional way, the insects and pests are determined through the manual way by the medical experts, which take high time consumption and may results in inaccurate results.To alleviate the issues like providing the imprecise output and time complexity, an automated model of pest identification (Ayan, Erbay, and Varc ¸ın 2020;Chen et al. 2020) can be applied through IoT.IoT-based agricultural systems provide accurate results when determining the pest regions in the crop field.
Here, the sensors need to be properly installed and maintained in the agricultural areas.The sensors act as the assistance of the farmers to find the target locations of the fields that are affected by the insects and pathogens.Initially, the sensors collect the data that are transferred into the centralized platforms through the wireless medium.This helps the farmers from the distant locations to monitor and protect their crops from the insects and pest attacks, which also reduces the possibility of environmental contamination and crop intoxication and also minimizes the usage of pesticides into the crop field.Thus, it is very essential to incorporate IoT technology with agriculture for detecting the insects and pests in the crops.The IoT-based pest detection in smart agriculture is shown in Figure 2.

Pest Detection from IoT Gathered Data Using Yolov3 with Deep
Feature Extraction

Dataset Description
The proposed pest detection model with IoT requires the pest images that are collected from (https://github.com/xpwu95/IP102:access date: , where m ¼ 1, 2, :::, Mand Mrefer the total number of pest images for 102 classes of pests.The sample images with different classes of pest from the dataset are shown in Figure 3.

Pest Detection Using Yolov3
The proposed pest detection model with IoT utilizes the YOLOv3 (Aly et al. 2021) for detecting the pest region from the input images mg input m : YOLO detection technique is considered as the deep learning scheme for performing the classification and identification of objects by using the input pest images.The object detection is performed with the determination of object placement with respect to their bounding boxes with the given images.This proposed model considers the YOLO-V3 classifier to attain the more accurate results when compared to earlier versions of this classifier.The training phase of YOLO-V3 is done since it has capacity to acquire the supplementary semantic information regarding the input images.However, it needs high time for processing the training network and also provides better information through the help of previous feature map.Finally, the YOLOv3 detects the pest from input images, and the detected images mg det m are then employed for the deep feature extraction phase.

Deep Features Extraction from Detected Pest
The image features are extracted from the detected images mg det m using CNN (Chen et al. 2020) in the proposed pest identification and classification model.CNN is one among the deep learning approach utilized for acquiring the learning features using the detected image mg det m and further, the sorting the features based on the feature inside the hidden neurons in the convolutional layer.In this process, the feature map distributes equal weights to entire neurons that are presented within the same feature map with respect to same time, and various weights are provided to neurons that are within distinct feature maps.Hence, the c th output of feature is mathematically expressed in Eq. (1).
Term, mg det m and Fe c is indicated as the input pest detected images and the convolution filter at c th output feature map, respectively.Here, 2D convolution filters are incorporated for measuring the filter model at every location present over the input images.The activation function of non-linear features is depicted by af ðÁÞ: In addition, the pooling layer helps to reduce the spatial resolution and spatial inconsistency in the input noise along with the transformation of feature map is performed.The more features are gathered via the pooling layer that gains higher value for the receptive field as in Eq. (2).
The pooling layer output is declared as ðOPÞ cst , which is placed at the c th feature map and the element of the pooling layer is annotated as ðmg det m Þ cxy in the location of ðx, yÞ: In addition, the corresponding loss function is considered as mean square.Since many optimizers are used in network model, the CNN considers Stochastic Gradient Descent (SGD) that is processed by the gradient multiplication of learning rate with respect to weights.The output of the pooling layer is given into the LSTM for pest detection.The extracted features from the DCNN are represented as FT ext f : 5. Modified CNN-LSTM Network for Optimal Pest Classification with IoT

CNN-LSTM Model
The extracted features FT ext f are taken as input for the developed CNLSTM approach for classifying the pest images to avoid the crop diseases through the suggested optimal pest classification model with IoT.The classification phases are positively enhanced with the recurrent structures belongs to the deep learning scheme and hence, the external memories are independent of preserving the output.This recurrent architecture of LSTM is ensured with the advantages like less computation complexity.In this LSTM (Chen et al. 2020) network, the four components are presented that are "cells, input gate, output gate and forget gate."The cell is processed with the data that is transmitted to input gate and then into output gate.Thus, the forget gate is used to find the data transmitted inside the network as in Eq. (3).
The weight matrices are described as C d , C h , C i , C r and the input variable is denoted as FT ext f : Then, "the cell output, output gate and forget gate" are represented as h u , r u and H u , respectively.The bias value of three gates is signified by x d , x g , x i , x r : Here, the term rannotates the sigmoid activation function and output of the hidden respect is given in l u , accordingly.The input gate is formulated in Eq. ( 4).
With the help of sigmoid function, a new cell states is updated to generate the new vector G _ t , which is estimated by Eq. ( 5).
In order to upgrade the old cell with new cell, the forget gate is required to add up some parameters with its previous state, which is mentioned in Eq. ( 6).
Finally, the sigmoid function of output gates are used to provide the outcome with respect to cell state, which is formulated using Eqs.( 7) and ( 8).
The sigmoid activation function is noted as r with tan has hyperbolic tangent.Finally, the classified outcomes for the proposed pest identification and classification model are acquired from the developed CNLSTM approach.

Proposed AHBA
The proposed pest identification and classification model is performed using the suggested AHBA for optimizing the hidden neurons in LSTM to improve the accuracy of the classification process.HBA (Hashim et al. 2022;Tabjula et al. 2021b) is chosen in the proposed pest identification and classification model as it has high convergence rate with the minimum time consumption.It also contains the improved exploration process owing to the diversity of ample population over the search process.On the other hand, it is necessary to resolve other practical scale optimization issues.Therefore, the proposed AHBA algorithm is improved for overcoming the mentioned challenges in existing algorithm.Thus, in the proposed AHBA, the random variable rr is computed by the new fitness-based concept for increasing the convergence performance of the improved algorithm.
Here, the term a is denoted as the variables determined based on fitness concept in the proposed algorithm, which is used for determining the random variables rr that is established in the random way in the conventional algorithm.Term bestfit and worstfit denotes the best fitness value and worst fitness value.HBA is the conventional algorithm that is motivated by the foraging (food searching) behavior of the honey badger.This animal finds their prey by smelling and digging or through following the honey guide bird.This algorithm runs under two digging mode and honey mode.It includes the "exploration and exploitation phase."The population of the solutions present in the algorithm is computed through Eq. ( 11).Here, the j th position of solution is indicated by y j ¼ y 1 j , y 2 j , :::, y d j h i : The position of the honey badgers is determined by Eq. ( 12).
Here, the upper and lower bounds are marked by Ub j and Lb j , accordingly and the position of j th honey badger is indicated by y j and the random number is expressed as rr 1 that has the range of ð0, 1Þ: Further, the inverse square law is involved for determining the intensity of the smell of the prey.When the high smell intensity is computed, then the movement of the honey badger will be fast.The computation of the smell intensity of the prey is given in Eq. ( 13).
Here, the term S in indicates the smell intensity of the prey; D in shows the distance among the j th solution and prey and ST represents the concentration strength or source strength.Further, the density factor a is the constraint for the transition of exploration to exploitation phase.The density factor is computed through Eq. ( 16).
Here, the term cexpresses the constant value and the total iteration count is declared by itr mx : Then, the new position y new for the search agents (honey badgers) is determined through the digging and honey phase.When considering the digging phase, the propagation of the honey badger is based on the Cardioid motion that is represented in Eq. ( 17).
Here, the random parameters are indicated by rr 3 , rr 4 and rr 5 : The position of the prey (best position) is shown by y pry and the distance between in th honey badger and prey is represented by D in : The badger may get certain disturbance f for altering the search directions that is shown in Eq. ( 18).
Term a indicates the influence factor of time varying search.When considering the honey phase, the honey badger is guided by the honey guide bird for reaching the new position y new that is shown in Eq. ( 19).
Here, the prey location and new position are indicated by y pry and y new , respectively.The random number is indicated by rr 7 that lies in the interval of ð0, 1Þ: Setting the population and its parameters Fitness is computed for all honey badger with the objective function While ðitr itr mx Þ do Upgrade the decreasing factor with Eq. ( 16).For ðin ¼ 1 to pPÞ Compute the intensity S in using Eq. ( 12).Determine the random variables rr using Eq. ( 9) If ðrr < 0:5Þ Use Eq. ( 17) to revise the position Else Use Eq. ( 19 The improved algorithm named AHBA enhances the overall performance of the proposed pest classification model with better efficiency in detection accuracy.The flowchart of the suggested AHBA is given in Figure 4.

Modified CNN-LSTM Model
The proposed optimal pest classification model with IoT builds a hybrid model of deep learning technique that termed as CNLSTM by optimally tuned the hidden neuron counts of LSTM with the help of AHBA algorithm for acquiring the better classification performance with less time requirement.The advantage of using CNN is to determine the informative features that relied on the statistical way of training even if it contains large scale dimension of data for rendering the higher predictive results.On the second hand, LSTM is majorly employed for diminishing the gradient vanishing issue and also proves the relative measure of insensitivity regarding gap length that is better rather than classical approaches of sequence learning and applicable for distinct applications.Hence, an improved and hybrid type of LSTM and CNN is constructed named as CNLSTM for increasing the performance of extraction techniques of feature and classification in the enhanced optimal pest classification model.The extracted features from CNN are considered into the developed optimized LSTM for identifying the pest types.This would prevent the crop disease and enhance the crop production.The objective function FF of the suggested optimal pest classification model is to improve the accuracy, which is determined via Eq.( 20).
Here, the term HN lstm b is indicating the hidden neuron count in LSTM.The developed AHBA tunes the hidden neurons that lie between 5 and 255.Accuracy accy is measured as the "closeness of the measurements to a specific value" as given the Eq. ( 21).
Here, "the true positive and true negative values" are indicated by PT and NT, correspondingly and "false positive and false negative values" are noted as PF and NF, accordingly.The developed CNLSTM classifier for optimal pest classification is diagrammatically illustrated in Figure 5.

Experimental Setup
Python tool was utilized for implementing the pest identification and classification model and it was assessed via the performance analysis.Here, the simulated analysis was accomplished that depends on the some quantitative measures and comparative analysis was made with several classical algorithms and classifiers for validating the performance of the proposed model.This experimental analysis was evaluated by having 10 population count and total iteration as 25 for the proposed pest identification and classification model.The proposed AHBA-CNLSTM was compared with other meta-heuristic algorithms like "Particle Swarm Optimization (PSO; Chaudhary et al. 2020), Tunicate Swarm Algorithm (TSA; Wankhede, Sambandam, and Kumar 2021), Deer Hunting Optimization Algorithm (DHOA; Divya and Ganeshbabu 2020), HBA (Hashim et al. 2022) and deep learning algorithms like CNN (Ambati et al. 2020;Wang et al. 2021), deep-CNN (Yue et al. 2018), RCNN (Li et al. 2021) and LSTM (Chen et al. 2020)."

Performance Metrics
The implemented pest identification and classification model is evaluated using various quantitative measures that are given as follows.
(a) FPR F pr is defined as "the ratio between the numbers of negative events wrongly categorized as positive (false positives) and the total number of actual negative events" as in Eq. ( 22).
(b) Specificity S spcty is "the proportion of negatives that are correctly identified" as in Eq. ( 23) (c) Sensitivity S senstvy is "the proportion of positives that are correctly identified" as in Eq. ( 24) (d) F1-score F score is referred as "the measurement of the accuracy in the conducted test" as in Eq. ( 25) (e) FNR F nr is "the proportion of positives which yield negative test outcomes with the test" as in Eq. ( 26) (f) NPV N pv is described as "the sum of all persons without disease in testing" as in Eq. ( 27) (g) MCC M cc is "a measure of the quality of binary classifications of testing" as given in the Eq. ( 28) (h) FDR is "a method of conceptualizing the rate of errors in testing when conducting multiple comparisons" as in Eq. ( 29) Precision prsn is explained as "the fraction of relevant instances among the retrieved instances" as in Eq. ( 30).
The imaging results of the pest detection using YOLOv3 approach in the proposed pest identification model are depicted in the Figure 6.

Algorithmic Analysis on Proposed Model
The efficacy of the implemented pest identification and classification model is evaluated in terms of different learning percentage based on developed FS-SSO with traditional heuristic development as represented in the Figure 7.The proposed AHBA-CNLSTM shows higher accuracy rather than the value of 0.3% for PSO-CNLSTM, 0.3% for TSA-CNLSTM, 0.35% for HOA-CNLSTM and  0.41% for HBA-CNLSTM, respectively at 70 th learning rate.When FNR is used to analyze the proposed AHBA-CNLSTM, it acquires low error with respect to all learning percentage.While increasing the learning percentage to 65, which is better than the other existing algorithms.Therefore, the proposed pest identification and classification model with implemented AHBA-CNLSTM is superior to former implemented meta-heuristic approaches.The proposed pest identification and classification model that is estimated against different optimization algorithms as in the Table 2 for evaluating its performance based on dataset.The outcome of the proposed AHBA-CNLSTM is 0.06%, 0.03%, 0.12% and 0.13% correspondingly improved by F1-score than the PSO-CNLSTM, TSA-CNLSTM, HOA-CNLSTM and HBA-CNLSTM.On observing the comparative analysis on optimization algorithms, the proposed AHBA-CNLSTM indicates the superior performance by decreasing the errors that affects the pest detection.Therefore, the proposed pest identification and classification model has improved its performance rather than other existing methods.

Overall Performance Analysis Based on Different Classifiers
The overall performance analysis is made on the proposed pest identification and classification model for estimating its performance with the inference of distinct classifiers that is portrayed in the  proposed pest identification and classification model enhances its performance than the existing methods.

Ablation Results Analysis of Proposed Pest Identification and Classification Model
The ablation experimental results of the recommended model is given in  Hence, it proves that it effectively increases the classification rate for rapid identification of pests.

Conclusion
This research has developed a novel pest identification and classification model with implemented AHBA for achieving the accurate detection of the pest in the crop field.Initially, the pest images were collected through the IoT technology with the sensors.The collected images were subjected into the object detection phase, where the YOLOv3 detector was utilized for detecting the pest regions in the given input images.The detected images were obtained from the YOLOv3 that were further given into the CNN framework to extract the significant features from the pest images.The extracted features were considered for developed CNLSTM network, where the optimal feature classification with proposed AHBA was performed that has classified into different classes of pests in the agricultural field.Depends on the experimental results, it was illustrated that the performance regarding accuracy of proposed AHBA-CNLSTM was 2.08% higher than CNN, 4.2% higher than Deep-CNN, 4.2% higher than RCNN and 5.3% higher than LSTM classifier.Therefore, the overall performance of the proposed pest identification and classification model with implemented AHBA-CNLSTM was superior to the other conventional techniques.

Figure 1 .
Figure 1.Proposed pest identification and classification architecture with IoT and deep learning approach.

Figure 3 .
Figure 3. Sample images for the pest detection from the dataset.

Figure 5 .
Figure 5. Developed CNLSTM classifier of the proposed optimal pest classification model.

Figure 6 .
Figure 6.Detected Images of the Proposed Pest Identification Model.

Table 1 .
(Wang et al. 2021lenges of Agriculture pest detection.thedensity of the monitoring cameras that are employed for the surveillance.But, there is a need for improving the model that resist to present the noisy or bluer representation of images through the process of DnCNN, Deblur GAN and other image enhancement approaches.CNN(Wang et al. 2021) performs well on small size pest detection.However, this model achieves poor accuracy.CNN(Thenmozhi andReddy 2019) has more potential toward the pest detection especially in the outdoor applications.But, it require more time for processing a large number of data.Therefore, a new pest detection model using deep learning and Iot is required to develop on considering these abovementioned drawbacks. reducing ) to upgrade the position End if Validate new position and set to the fitness fn new If ðfn new fn in Þ Declare ðy in y new Þ and ðfn in fn new Þ End if If ðfn new fn pry Þ Declare ðy pry y new Þ and ðfn pry fn new Þ 6.5.Classifier analysis on Proposed Model Based on Optimal Pest ClassificationThe suggested pest identification and classification model is tested at varying learning percentage using the developed AHBA-CNLSTM with distinct classifiers as shown in Figure8.The proposed AHBA-CNLSTM renders 34.6%, 27.2%, 29.6% and 48.9% improved MCC value than the "CNN, Deep-CNN, RCNN and LSTM," respectively at the learning rate of 65.The proposed AHBA-CNLSTM secures less false value, which is compared against various classifier in all quantitative measures.Therefore, the proposed pest identification and classification model with implemented AHBA-CNLSTM exploits effective performance compared over conventional classifier techniques.

Table 3
The proposed AHBA-CNLSTM provides efficient performance in classifying the pest diseases rather than conventional algorithms.Therefore, the Table 4 and compared against other techniques.The ablation analysis is mainly used for classification technique.It uses to analyze the accurate value of proposed work and compared with some existing classifier model.As given in the Table 4, the suggested hybrid CNLSTM exploits more accuracy percentage when compared the previously implemented methods.

Table 2 .
Comparative analysis on proposed pest identification and classification model using existing meta-heuristic algorithms.

Table 3 .
Comparative analysis on proposed pest identification and classification model with existing classifiers.

Table 4 .
Ablation experimental analysis on proposed pest identification and classification model with classical classifiers.