Optimal Trained Deep LSTM Model for Detecting Plant Leave Diseases

Optimal Trained Deep LSTM Model for Detecting Plant Leave Diseases

Sarita Das(1)

Computer Science and Engineering KIIT University

Bhubaneswar, India

Bhabani Shankar Prasad Mishra(2) Computer Science and Engineering KIIT University

Bhubaneswar, India

Abhinandan Kumar Tiwar(3) Computer Science and Engineering KIIT University

Bhubaneswar, India

Bikramaditya Panda(4) Computer Science and Engineering KIIT University

Bhubaneswar, India

Abstract Automated plant leaf disease detection has the advantage of removing a lot of monitoring work from increased crop farms and spotting disease symptoms when they first appear on plant leaves, which is when they are most treatable. The proposed plant leaf disease detection has 3 steps: preprocessing, feature extraction and detection. In preprocessing phase, histogram equalization will takes place. A technique for adjusting contrast in image processing is called histogram equalization. In feature extraction, 3 features were extracted like statistical features, LGP and color features. In detection phase, we use an optimized LSTM to identify plant leaf diseases and classify them. SI-BMO algorithm is proposed to optimize the weight in the LSTM classifier. Finally, the models performance was evaluated and the result was analyzed successfully.

Keywordsplant leave disease detection system; feature extraction; Preprocessing; LSTM; SI-BMO

Abbreviation	Description
SVM	support vector machine
DL	deep learning
SI-BMO	Self Improved Blue Monkey Optimization
BM	blue monkey
ML	machine learning
OMNCNN	optimal mobile network-based convolutional neural network
BF	bilateral filtering
DCNN	deep convolutional neural networks
2D AADF	2D Adaptive Anisotropic Diffusion Filter
LGP	Local Gradient Pattern
AMA	Adaptive Mean Adjustment

NOMENCLATURE

ILSM	Intelligent Leaf Scanning Mechanism
LSTM	Long term short memory

I. INTRODUCTION

The benefit of automatic plant disease detection [9] [10] is that it eliminates a significant amount of monitoring work in large crop farms and identifies disease symptoms at an extremely early stage, when they first emerge on plant leaves. As a result, one of the key components of a plants pathologist's education is diagnosis. Illness control procedures can be a sunk cost and can result in additional plant losses if the disease and its cause are not correctly identified.

Therefore, accurate illness diagnosis is essential. Pathogenic fungi, on the other hand, are the culprits behind plant ailments like aphid, leaf blight, rust, wilt, blight, coiled, scab, gout, blight, cooling, root rot, mildew, and overgrowth. Major contributors to yield, commercial crop, and crop quality losses include systemic foliar infections. A plant disease symptom [12] [13] is a disease's outward manifestation in the plant. [8] A noticeable alteration in the plant's color, shape, or functionality as a result of the disease is one of the symptoms. Reduced economic and aesthetically detrimental effects of plant diseases are the main objective of plant protection [14]. This has historically been referred regarded as plant disease control [15], however modern social and environmental principles view "control" as being excessively rigid and uncompromising.

The farmland mass is used for more than just growing food in the modern world. The Indian economy is heavily reliant on agricultural output. [6] As a result, it is crucial to identify plant diseases in the sector of agriculture. Automatic plant disease diagnosis from raw photos is now achievable because to developments in artificial intelligence research [7]. Today, picture segmentation is most frequently performed using soft computing, which has the capacity to deal with uncertainty. To

identify the Semi-automatic leaf disease, SVM [11] classifier were developed. This research work's main contribution is:

• New plant leave disease detection is introduced which has 3 stages: Preprocessing, feature extraction and Detection.

• Optimized LSTM classifier is used to detect the plant leave disease.

• Optimal training of LSTM classifier is done by a new SI-BMO algorithm.

This report is categorized as follows:

Section I gives a description of the Introduction. The subject of the literature review is discussed in Section II. Architectural description of suggested plant leave disease detection is discussed in Section III. Section IV discusses preprocessing, feature extraction, and detection. Section V: Optimized LSTM classifier. In Section VI, SI-BMO algorithm is studied. The conclusion is the topic of Section VII.

1. LITERATURE REVIEW

   A. Literature Review

   In 2021, R. Sujatha et al. [1] established approaches for identifying plant diseases, including DL and (ML). In terms of detecting citrus plant diseases, the effectiveness of ML (SVM, Random Forest, Stochastic Gradient Descent, and VGG-16) and DL (Inception-v3, VGG-16, and VGG-19) was compared. In 2021, S. Ashwinkumar et al. [2] suggested a convolutional neural network based on an ideal mobile network for detecting and categorizing plant leaf diseases. The suggested OMNCNN system uses preprocessing,

   segmentation, feature extraction, and classification as its

   primary operational steps. Preprocessing based on BF was included.

   In 2021, M. Yogeshwari and G. Thailambal [3] developed a new method using DCNN for the identification of plant leaf diseases. The plant leaf images in the suggested scheme are first preprocessed utilizing filtering and enhancement methods. The 2D AADF is used in this work's image filtering to remove noise. AMA approach was used to enhance these denoised photos.

   In 2020, Navneet Kaur and V. Devendran [4] suggested the optimization-oriented segmentation and law mask framework as a solution to the classification of leaf disease with various sample classes and sizes. In the end, the leaves were classified mostly in learnt metric space using the SVM as a classifier.

   In 2021, U. Maheswaran et al. [5] unveiled a novel, portable smart device called SmartBox, which utilizes an ILSM to function. This ILSM technique was related to the SmartBox, which is built with a variety of intelligent parts like a camera and WiFi.

   Table 1 depicts the features and challenges of conventional approaches.

   TABLE I. FEATURES AND CHALLENGES OF CONVENTIONAL METHODS

   Author [citation]	Methods	Features	Challenges
   R. Sujatha et al. [1]	SVM	Performs similarly effectively when there has been a discernible margin of class separation	Plant diseases that harm the leaf can significantly reduce crop production and market value.
   S. Ashwinkumar et al. [2]	OMNCNN method	Automated prevention and detection methods are useful for easing the burdensome work of keeping track onlarge agricultural farms.	A large amount of trained data is needed.
   M. Yogeshwari and G. Thailambal [3]	DCNN	Identify plant leaf ailments	Lack of appropriate detection could seriously harm the amount and quality of the agricultural harvest.
   Navneet Kaur and V. Devendran [4]	law mask framework	Classify the issue of leaf disease using various sample sizes and types.	Need to spend a bit more time
   U. Maheswaran et al. [5]	ILSM	Sort the leaf photos into regions that are affected and those that are not.	Maintaining an agricultural field is difficult.

2. PROPOSED PLANT LEAVE DISEASE DETECTION: ARCHITECTURAL DESCRIPTION

   Automated plant leaf disease detection has the advantage of removing a lot of monitoring work from increased crop farms and spotting disease symptoms when they first appear on plant leaves, which is when they are most treatable. The proposed plant leaf disease detection has 3 steps: preprocessing, feature extraction and detection. Fig. 1 depicts the suggested plant leave disease detection architecture.

   • Preprocessing: In this phase, histogram equalization will takes place. A technique for adjusting contrast in image processing is called histogram equalization.

   • Feature extraction: In this phase, 3 features were extracted like statistical features, LGP and color features.

   • Detection: In this phase, we use an optimized LSTM to identify plant leaf diseases and classify them.

   Weight optimized by SI-BMO algorithm

   Input

   Preprocessing Feature extraction LSTM classifier

   Output

   Features

   • Statistical features

   • LGP

   • Color features

   Mean, Median, Standard deviation

   Fig. 1. Proposed plat leave disease detection system

3. DESCRIPTION ON PREPROCESSING AND FEATURE EXTRACTION

   1. Preprocessing

      Histogram equalization [16] will happen during this phase for image enhancement. When the image's useable data is considered to be adequate values, this strategy typically raises the overall contrast of numerous images. The intensities on the histogram can be more evenly spread by making this adjustment. This makes it possible for regions with less local contrast to acquire more contrast. This is accomplished by histogram equalization by evenly distributing the most prevalent intensity values.

      The technique works well in images where the foreground

      • Statistical features: Mean (Mean) , Median (Mid ) , Standard deviation (STD) were the statistical features extracted here.

      • LGP [17]: LGP is widely utilized in numerous applications for image classification. In order to generate the fluctuation of local intensity for a given pixel Ab , LGP first determines the gradient values, ei , of adjacent pixel Ai as ei | Ai Ab |

        . Later, the threshold is determined using the average of the

        gradient values for N neighboring pixels, or eam , which is

        1 N 1

        and background are both dark and both bright. When applied to

        computed using AM as

        eam

        ei

        N i0

        . The LGP then

        photos with poor color depth, histogram equalization can also have undesired effects (such a visible image gradient). The preprocessed image is depicted as P .

   2. Feature extraction

   From the preprocessed image P , 3 features like statistical features, LGP and color features were extracted.

   compares each surrounding pixel's gradient value to the

   threshold, eam , and sets a binary value of "1" if the gradient value is greater than or equal to eam and a binary number of "0" otherwise. Finally, the center pixel was given the binary number sequence in any direction that was encoded

   • as the equal decimal value. The concluding equations of LGP are Equations 1 and 2.

     N 1

     i

     LGPN ,r xb , yb c(ei eam ).2

     the output gate. Consider that the input data r1 , r

     time t . The processing method for each

     2 ,…., r t is the

     1, ifb 0

     i0

     (1)

     LSTM is as follows in eq. (3), eq. 94), eq. (5), eq. (6), and eq. (7):

     0

     c(b) , otherwise

     (2)

     gt pll ut1, ht wl

     (3)

     Where,

     r = radius

     d pl

     u

     ,h w

     t m t1 t

     m (4)

     xb , yb =coordinate of centre pixel

     q pl

     s ,u

     , h w

     t n t

     t1 t

     n (5)

     b = threshold difference

     eam =threshold

     s d s

     • g tanul

       u

       ,h

       (6)

       t t t1 t

       b t1 t

   • Color features [18]: The LDF filters and arranges images from a database using global criteria. After that, they look for regional traits in both the target image and all other images. They must also evaluate the image's color

   ut qt tanust

   Where,

   ht = input LSTM

   (7)

   distribution data. They use a histogram that contains site color occurrence of picture colors and local color occurrence within particular regions for each bin block to generate an area-wise directional statistical histogram. Color feature is depicted as ColorF .

   ut1 = hidden layer output

   gt ,dt ,qt = input gate input

   ll , ld , lq = weight matrix

   wl , wd , wq = offset vector

   The final feature set

   Final F is depicted as

   Final F {Mean, Mid , STD, LGP, ColorF}.

4. OPTIMIZED LSTM CLASSIFIER

   To find the plant leaf disease, the chosen feature set

   lb = the cell unit's weight in relation to the input

   ut = hidden layer output

   J = sigmoid function

   The SI-BMO (Self Improved Blue Monkey Optimization) algorithm, which is thoroughly described in section VI, is used to optimize the weight W in the

   Final F

   will be put to the test. In this study, we identify and

   LSTM classifier.

   categorize plant leaf diseases using an optimized LSTM [19]. Furthermore, a recently introduced SI-BMO algorithm performs the training by tuning the optimal weights, which

   A. Objective function and solution encoding

   In this proposed work, weight is optimized by defining the objective

   improves the results of the designed LSTM classifier.

   function Obj F

   given by eq. (8).

   The test pattern must be acquired and retained by the

   Obj F min(error )

   (8)

   W1 W2

   .. WN

   Solution encoding of proposed SI-BMO algorithm is depicted in fig. 2.

5. SELF IMPROVED BLUE MONKEY OPTIMIZATION (SI-BMO) ALGORITHM

   Fig. 2. Solution encoding

   particular gated units in order for LSTM to sustain relatively long series data dependence and achieve high precision forecasting. The forget gate, input gate, and output gate are the three gating components that make up the majority of the LSTM neural network. The specialized gate units learn and retain the sequence analysis in order to preserve lengthy period series information reliance and make significant predictions. The forget gate regulates how efficiently a cell can retain knowledge from the past. The neuron's output is represented by

   A novel metaheuristic program optimization called BM is based on how blue monkey swarms function in nature. The number of men in a group is determined by the BMO algorithm [20]. The blue monkeys differ from other species of monkeys. They typically reside in social systems where women predominate, meaning that females remain in their birth groupings. The males depart from their groupings as soon as they reach the mature stage. There are typically many females and young in each blue monkey group, but still only one male. Inbreeding is made challenging bythis issue. The BM algorithmic rule imitates Blue Monkey behavior. Every group of monkeys had to go across the search area in order to simulate

   these interactions. Little to no interaction exists between the ale Cercopithecus mitis and the young. Young males should

   venture outside as soon as they can because of the cercopithecus mitis's territorial character in order to succeed. Typically, blue

   W

   child

   i

   X child i

   = child weight

   = child position

   monkey groups consist of one male and a large number of females and young.

   • Improved Position Update: In conventional position update, better blue monkey position in each group

     As per the SI-BMO algorithm, rand value of children update can be calculated using logistic map function. A logistic map [22] produces chaotic patterns in (0, 1). The following equation (15) provides the formal definition of this map:

     determines the updating position for every blue monkey in that group. The following equations 9 and 10 characterize this behavior:

     xk 1 Qxk (1 xk )

     Where,

     Q

     (15)

     Ratingi1 (0.7* Ratingi ) (Wlead Wi )*rando*(Mbest Mi )

     (9)

     Mi1 Mi Ratingi1 * rando

     (10)

     = adjustable parameter

     The pseudo-code of proposed SI-BMO algorithm is shown below:

     Algorithm 1: Self Improved Blue Monkey Optimization

     Initialize the population of BM and child

     Initialize the weight W and the power rate Rating .

     Put the blue monkeys into teams (T ) at random, with the entire child in one team.

     Determine the fitness levels of each group's BM and youngsters.

     Choose the worst and greatest fitness values for each group, and then save them in Current Best. Children choose the fittest option.

     t 1

     while t maxiiterations

     Swapping the least fit members of each group for the fit members of the child's group.

     Improved position update is done in eq. (11) and eq. (12)

     Improved children update is done in eq. (15)

     Fitness update of every BM and child.

     Current best update

     Current Best Equals New Best if New Best is superior to Current Best.

     t t 1 child child

     end while

     return optimal BM

     Where,

     Wlead =leader weight Wi =monkey weight Mbest =leader position Mi =position of monkey

     rando=random number

     Rating = power rate of monkey

     As per the SI-BMO algorithm, the improved position update is depicted in eq. (11).

     Mi1 Mi Ratingi1 * rando Levy()

     (11)

     Levy( ) [21] is calculated in eq. (12)

     Levy( ) 2 t * r1*

     Where,

     t

     max

     1

     | r2 |

     (12)

     t = current iteration

     tmax =maximum iteration

     = constant

     r1, r2 =random number

   • Children update: The model equation (13) and (14) were utilized for children update of BM.

   i 1

   Ratingchild

   (13)

   i

   (0.7 * Ratingchild

   ) (W W

   child ch

   leader i

   )* rando*(Mbest

   • Mi )

   M

   child i1

   (14)

   Where,

   Ratingchild

   M

   child i

   i1

   • Ratingchild

   * rando

6. RESULTS AND DISCUSSION

   W

   child lead

   = child power rate

   =weight of leader child

   1. Simulation set up

      This work was done in python. The performance of LSTM + SI-BMO for detecting plant leave disease was

      computed using data from [23]. The adopted LSTM + SI-BMO was distinguished with LSTM + SSO, LSTM + CMBO, LSTM

      + PRO, LSTM + BMO, and CNN, SVM, DBN and GRU.

      Dataset description: Human society needs to increase food production by an estimated 70% by 2050 to feed an expected population size that is predicted to be over 9 billion people.

      Currently, infectious diseases reduce the potential yield by an average of 40% with many farmers in the developing world experiencing yield losses as high as 100%. The widespread distribution of smart phones among crop growers around the world with an expected 5 billion smart phones by 2020 offers the potential of turning the smart phone into a valuable tool for diverse communities growing food. One potential application is

      the development of mobile disease diagnostics through machine learning and crowd sourcing. Here we announce the release of over 50,000 expertly curated images on healthy and infected leaves of crops plants through the existing online platform Plant Village. We describe both the data and the platform. These data are the beginning of an on-going, crowd sourcing effort to enable computer vision approaches to help solve the problem of yield losses in crop plants due to infectious diseases. The sample image showing the affected and healthy leaves after HE is shown in Fig. 3.

      Original leaves

      Pre- processed leaves

      1. (b) (c) Fig. 3. Sample image revealing (a) image 1 (b) image 2 (c) image 3

   2. Convergence Analysis

      The cost analysis of SI-BMO scheme for diverse iterations is exposed in Fig. 4. Here, analysis was done on SI-BMO over LSTM + SSO, LSTM + CMBO, LSTM + PRO, LSTM + BMO.

      On examining Fig. 4, SI-BMO has got lesser cost from iteration 17 to 25. On the other hand, a high cost of 1.061 is achieved by CMBO from iteration 0 to 8. Next to that, BMO has achieved a high cost of 1.060. Thus, when compared over others, SI-BMO has accomplished lesser cost.

      Fig. 4. Convergence analysis of SI-BMO model over others

   3. Performance Analysis

      The SI-BMO is validated regarding accuracy, sensitivity, specificity, and precision, FPR, FNR and so on. The analysis using SI-BMO for varied LPs is shown in Fig. 5 7. On investigating Fig. 5, a high accuracy of 0.95 is achieved by SI- BMO at 90th LP, while at 60th, 70th and 80th LPs; the accuracy

      achieved by SI-BMO is comparatively less, at 60th, 70th and 80th LPs, the accuracy achieved by SI-BMO is 0.9, 0.91 and

      1. respectively. Likewise, the FNR achieved by SI-BMO at 90th LP is much less than LSTM + SSO, LSTM + CMBO, LSTM + PRO, LSTM + BMO. Table II shows the examination of SI-BMO on varied classifiers like CNN, SVM, DBN and GRU. In Table II, for every variant of LP, SI-BMO model has depicted better values over 90%.

         (a)	(b)
         (c)	(d)

         1. (b)

      (c)

      Fig. 6. Analysis on (a) F-measure (b) MCC and (c) NPV

      (a)	(b)

      Fig. 7. Analysis on (a) FPR (b) FNR

      TABLE II. ANALYSIS OF SI-BMO ON VARIED CLASSIFIERS

      Metrics	CNN	SVM	DBN	GRU	SI-BMO
      NPV	0.60256 6	0.53604 2	0.50256 4	0.54617 4	0.80898 9
      Specificit y	0.81619 9	0.77189 7	0.74741 9	0.77900 2	0.92911 7
      F-measure	0.87662	0.8441	0.82562	0.84939 5	0.95448 8
      Accuracy	0.85844 8	0.82211 1	0.80163	0.828	0.94426 2
      MCC	0.51884 1	0.42244 3	0.37133 6	0.43754 9	0.79534 8

      Precision	0.89557 9	0.86729 9	0.85108 3	0.87192 6	0.96493 8
      FPR	0.31805 2	0.37966	0.41172 7	0.37009 8	0.12689 9
      Sensitivit y	0.68194 8	0.62034	0.58827 3	0.62990 2	0.87310 1
      FNR	0.14155 2	0.17788 9	0.19836 4	0.172	0.05573 8

   4. Ablation Study

      Table III examines the presented LSTM + SI-BMO model over other extensive features. This section explains the impact of developed model on optimization. Since, we have deployed enhancements in conventional BMO optimization, better outcomes are achieved. In Table III, it could be noted that the

      accuracy for LSTM with no optimization (0.739268) is less than LSTM with optimization (0.944262). Thus, enhancements done in conventional BMO had outcome in better outputs.

      TABLE III. ANALYSIS OF PROPOSED MODEL OVER PROPOSED WITHOUT OPTIMIZATION

      Metrics	Proposed with no optimization	LSTM + SI- BMO
      MCC	0.322225	0.795348
      Specificit y	0.681059	0.929117
      FNR	0.260732	0.055738
      Precision	0.811235	0.964938
      F-measure	0.773581	0.954488
      Sensitivit y	0.517932	0.873101
      NPV	0.414807	0.808989
      Accuracy	0.739268	0.944262
      FPR	0.482068	0.126899

7. CONCLUSION

Proposed plant leaf disease detection has 3 steps: preprocessing, feature extraction and detection. In preprocessing phase, histogram equalization will takes place. A technique for adjusting contrast in image processing is called histogram equalization. In feature extraction, 3 features were extracted like statistical features, LGP and color features. In detection phase, we use an optimized LSTM to identify plant leaf diseases and classify them. SI-BMO algorithm is proposed to optimize the weight in the LSTM classifier. Finally, SI-BMO algorithm performance was evaluated and the result was analyzed successfully.

[2] Jeong, J., Kwon, S., Hong, MP. et al. Adversarial attack-based security S. Ashwinkuma, S. Rajagopal, V. Manimaran, B. Jegajothi, "Automated plant leaf disease detection and classification using optimal MobileNet based convolutional neural networks", 2021.

[3] M. Yogeshwari and G. Thailambal, "Automatic feature extraction and detection of plant leaf disease using GLCM features and convolutional neural networks", 2021.

[4] Navneet Kaur and V. Devendran, "Novel plant leaf disease detection based on optimize segmentation and law mask feature extraction with SVM classifier", 2020.

[5] U. Maheswaran, Ravindra Babu Kallam, B. Arathi, Koppula Prawan, Anitha G, "Efficient plant leaf disease identification Material Fabrication using lightweight device", 2021.

[6] V. Singh, A.K. Misra, Detection of Plant Leaf Diseases Using Image Segmentation and Soft Computing Techniques, Information Processing in Agriculture, 2018.

[7] M. Sardogan, A. Tuncer and Y. Ozen, "Plant Leaf Disease Detection and Classification Based on CNN with LVQ Algorithm," 2018 3rd International Conference on Computer Science and Engineering (UBMK), 2018, pp. 382-385, doi: 10.1109/UBMK.2018.8566635.

[8] S. S. Chouhan, A. Kaul, U. P. Singh and S. Jain, "Bacterial Foraging Optimization Based Radial Basis Function Neural Network (BRBFNN) for Identification and Classification of Plant Leaf Diseases: An Automatic Approach Towards Plant Pathology," in IEEE Access, vol. 6, pp. 8852- 8863, 2018, doi: 10.1109/ACCESS.2018.2800685.

[9] X. Liu, W. Min, S. Mei, L. Wang and S. Jiang, "Plant Disease Recognition: A Large-Scale Benchmark Dataset and a Visual Region and Loss Reweighting Approach," in IEEE Transactions on Image Processing, vol. 30, pp. 2003-2015, 2021, doi: 10.1109/TIP.2021.3049334.

[10] L. Li, S. Zhang and B. Wang, "Plant Disease Detection and Classification by Deep LearningA Review," in IEEE Access, vol. 9, pp. 56683-56698, 2021, doi: 10.1109/ACCESS.2021.3069646.

[11] Deepa, N.R., Nagarajan, N. RETRACTED ARTICLE: Kuan noise filter with Hough transformation based reweighted linear program boost classification for plant leaf disease detection. J Ambient Intell Human Comput 12, 59795992 (2021). https://doi.org/10.1007/s12652-020- 02149-x

[12] Zhang, S., You, Z. & Wu, X. Plant disease leaf image segmentation based on superpixel clustering and EM algorithm. Neural Comput & Applic 31, 12251232 (2019). https://doi.org/10.1007/s00521-017-3067-8

[13] [13] Prasad, S., Peddoju, S.K. & Ghosh, D. Multi-resolution mobile vision system for plant leaf disease diagnosis. SIViP 10, 379388 (2016). https://doi.org/10.1007/s11760-015-0751-y

[14] Chouhan, S.S., Singh, U.P. & Jain, S. Automated Plant Leaf Disease Detection and Classification Using Fuzzy Based Function Network. Wireless Pers Commun 121, 17571779 (2021). https://doi.org/10.1007/s11277-021-08734-3

[15] Barbedo, J.G.A. A novel algorithm for semi-automatic segmentation of plant leaf disease symptoms using digital image processing. Trop. plant pathol. 41, 210224 (2016). https://doi.org/10.1007/s40858-016-0090-8

[16] R. Dorathy, RM. Joany, R. Joseph Rathish, S. Santhana Prabha, S. Rajendran, "Image Enhancement by Histogram Equalization", 2016.

[17] U. Habiba, M. R. Howlader, R. H. Faisal and M. M. Rahman, "hLGP: A Modified Local Gradient Pattern for Image Classification," 2019 International Conference on Electrical, Computer and Communication Engineering (ECCE), 2019, pp. 1-6, doi: 10.1109/ECACE.2019.8679470.

[18] Jyoti Narwade, Binod Kumar (2016). Local and Global Color Histogram Feature for Color Content-Based Image Retrieval System. In: Satapathy, S., Bhatt, Y., Joshi, A., Mishra, D. (eds) Proceedings of the International Congress on Information and Communication Technology. Advances in Intelligent Systems and Computing, vol 438. Springer, Singapore. https://doi.org/10.1007/978-981-10-0767-5_32

[19] Keqiao Chen, "APSO-LSTM: An Improved LSTM Neural Network Model Based on APSO Algorithm", 2020.

[20] Maha Mahmood and Belal-AI-Khateeb, "The Blue Monkey: A New Nature Inspired Metaheuristic Optimization Algorithm", 2019.

REFERENCES

[21] Çig d em

Inan Ac and Hakan Gu l can, "A Modified Dragonfly

[1] R. Sujatha, Jyotir Moy Chatterjee, NZ Jhanjhi, Sarfraz Nawaz Brohi,"Performance of deep learning vs machine learning in plant leaf disease detection ", Microprocessors and Microsystems.

Optimization Algorithm for Single- and Multiobjective Problems Using Brownian Motion", 2019.

[22] Hui Lu, Xiaoteng Wang, Zongming Fei and Meikang Qiu, "The Effects of Using Chaotic Map on Improving the Performance of Multiobjective Evolutionary Algorithms", 2013.