Plant Leaf Recognition Using Machine Learning: A Review

Plant Leaf Recognition Using Machine Learning: A Review

3,4,5,6

Abstract - Plants classification through leaves is an innovating and fascinating area of research that can provide helpful information regarding plants. Plant identification using their leaves is important in agriculture for weed identification, plant growth assessment and classification of diseases in plants. In addition to this, leaves can prove tobean important factor in identification of plant species in comparison to other parts of plants including flowers, stems, and seeds. Although recent advancements in the field of machine learning have made leaf classification much easier Identifying plant species using their leaf images proves to bea challenge due to the vast variation among species and variations in their shape, size, and color. This review paper gives a detailed literature review of numerous tools and algorithms used in plant classification, providing their potential results and high accuracy. Some of the most commonly used leaf classification methods include support vector machines, convolutional neural networks, and decision trees. These algorithms have many applications, including estimating carbon uptake, predicting yields, and monitoring plant health and biodiversity. Plant classification through leaves can have applications in various areaofinterestsuch as agriculture, botanical research, medicine (Ayurveda) etc In Ayurveda, plants are used as medicines providing solutions to diabetes, digestive problems, diseases related to the heart, liver disorder, etc As machine learning and imagerecognition evolve, plant classification will have an even more significant impact in these fields.

Key Words: Machine Learning, Deep Learning, Plant Recognition, Pre-processing, Feature Extraction

1. INTRODUCTION

Plants play a crucial role in the ecosystem and have been used for various purposes throughout history. From agriculture to medicine, plants have been a source of sustenance and healing for humans. Identifying plants is important in agriculture for weed detection, plant growth estimation,anddiseasedetection.Manualidentificationof plantsthroughtheirleavesisatime-consumingandtedious job, which can be counteracted by the development of a plantidentificationsystem.Inrecentyears,technologyhas madeplantidentificationmoreaccessible,andvariousplant identificationsystemshavebeendeveloped.Leavesarethe most important part of a plant for classification as they

provide important information about the species. Leaf characteristicssuchastheshape,size,andcolour,aswellas the pattern of veins, hairs, or glands, can be used to differentiate between different plant species. The arrangement of leaves on the stem can also be used as a distinguishing feature. Moreover, leaves remain on the plantsformostoftheyear,makingthemanidealparttouse for plant identification. Machine learning algorithms have become a popular method for plant identification. These algorithms can recognize patterns and features in plant leavesandusethemtoidentifyunknownplantsaccurately andquickly.Variousmachinelearningalgorithmshavebeen usedinthedevelopmentofplantidentificationmodels,such assupportvectormachines,randomforests,anddeepneural networks.Additionally,image processing techniqueshave beenusedtoextractfeaturesfromplantleavesthatcanbe used for plant identification. This paper reviews various studies conducted to develop plant identification systems based on leaf characteristics. The paper discusses the differentmachinelearningalgorithmsusedandtheimage processingtechniquesappliedtoextractfeaturesfromplant leaves.Thepaperalsoexploresthedifferentapplicationsof plantidentificationsystemsinagricultureandhorticulture, suchasweeddetection,plantgrowthestimation,anddisease detection.Thedevelopmentofplantidentificationsystems hasopenednewopportunitiesintheidentificationofplants with medicinal properties. Plants have been used as medicinesforcenturies,andtheidentificationofplantswith medicinal properties can lead to the development of new drugs and treatments for various health disorders. The identification of plants with medicinal properties can be donethroughtheirleaves,andtheuseofplantidentification systems can hasten the process. In conclusion, the development of plant identification systems based on leaf characteristics has numerous applications in agriculture, horticulture, and medicine. The identification of plants throughtheirleaveshasbecomemoreaccessible,thanksto technology and the development of machine learning algorithms.Theuseofplantidentificationsystemscanhelp inthedetection of plantdiseases,weedcontrol,andplant growthestimation,makingitanessentialtoolinagriculture andhorticulture.Theidentificationofplantswithmedicinal propertiescanalsobedonethroughtheirleaves,leadingto thedevelopmentofnewdrugsandtreatmentsforvarious healthdisorders.Thepaperaimstoprovideanoverviewof

various studies conducted on plant identification systems andthedifferentapplicationsofplantidentificationsystems.

2. MOTIVATION

Plantleafrecognitionisveryinterestingandessentialfield withremarkablepotentialthatimpactsinvariousnumberof fields such as agriculture, medicine, forestry, and environmentalprotection.Byaccuratelyidentifyingdifferent plantspeciesbasedfromtheiruniqueleaffeatures,wecan gain insight into their growth patterns, their response to environmental factors, and their overall health. This knowledge can help us develop more efficient and sustainableagriculturalpractices,managenaturalresources better, and even discover new plant species. Additionally, with rapid advances in machine learning and computer vision, the development of automated leaf recognition systemscangreatlyincreasethespeedandaccuracyofplant identification, making it an essential tool for researchers, farmers,andecologists.Therefore,thereisagreatneedfor motivatedindividualstojointhefieldandcontributetheir skillsandknowledgetoadvanceourunderstandingofplant biologyandthenaturalworld

3. LITERATURE REVIEW

I. Leaf Analysis for Plant Recognition:

In this study, [1], a weighted K closest neighbor search algorithm is used to propose a leaf analysis systemforplantidentification.Thesystemconsistsof noise reduction preprocessing processes, feature extraction for computing scale invariant feature descriptors,andalgorithmicmatchingofplantspecies. TheLeafsnapdatasetisusedbytheauthorstotestthe system before it is put into use as a Windows phone app.

II. A Mobile Application for Plant Recognition through Deep Learning:

Thepaper[2]outlinesamethodfordeeplearningbased automated plant and flower recognition. This method makes use of video data to make up for any informationlossthatcanoccurwhencomparingstatic photographs of plants and flowers, in contrast to conventionalmethodsthatonlyemploystaticimages. Theapproach'sdeeplearningalgorithmsaswellasthe procedure for gathering, scrubbing, and purging data are described in the study. Also, a mobile iOS app is provided,andtheapproach'sresultsdemonstratethat 122/125 plants and 47/50 genera may be identified withadegreeofconfidenceupto95%.Theutilization of cloud-based resources to increase performance speedisalsocoveredinthestudy.

III. Tree Species Identification Based on Convolutional Neural Networks:

This paper [3] suggests an efficient convolutional neural network-based method for automatically classifying tree species (CNNs). The examination of numerous multi-dimensional characteristics of tree leaves,suchascolor,shape,andveinsignatures,isdone tocarryouttheidentification.Sinceitcanbedifficultto accurately identify a single leaf trait for a given tree species, CNNs are used to combine the multidimensionalinformation.Preprocessingproceduresare also used to increase the identification results' reliability. The Leafsnap database is used to test the proposedapproach,andtheresultsaregood.

IV. A Leaf Recognition Approach to Plant Classification Using Machine Learning:

Thepaper[4]introducesanautomatedmethodfor identifyingplantsthroughleafrecognition,whichisan importantpartofplantecologicalresearchworkflows. The proposed methodology is simple as well as efficient, which uses a combination of two texture features(BOFandLBP)asinputstoamulticlassSVM classifier.Themethodisevaluatedusinga leafimage databaseandshowsextremelyeffectiveresults.Their proposed method has great potential for practical applications in plant recognition due to its computational efficiency and ease of implementation usingcomputervisiontechniques.Overall,thispaper providesasignificantcontributiontothefieldofplant identification.

V. Plant identification using deep neural networks via optimization of transfer learning parameters:

In this paper [5], deep convolutional neural networkswereutilizedforthepurposeofidentifying plant species captured in photographs. The performanceofdifferentfactorsthataffecttheaccuracy of these networks was evaluated. Three popular and significantdeeplearningarchitectures,suchasAlexNet, GoogleNet, and VGGNet, were implemented for the purposeofthisstudy.Transferlearningwasemployed usingLifeCLEF2015planttaskdatasetsinordertofinetune the pre-trained models. Data augmentation techniques based on image transforms such as reflection, rotation, scaling, and translation were appliedtominimizetheriskofoverfitting.Inaddition, the network parameters were adjusted and different classifiers were combined to enhance overall performance. The best combined system achieved an accuracyof80%approximatelyusingthevalidationset andanapproximateinverserankscoreof0.752using officialtestset.Comparingtheseresultswiththoseof the LifeCLEF 2015 plant identification campaign, the topsystem'soverallvalidationaccuracywasimproved

by15%pointsanditsoverallinverserankscoreonthe testsetby0.1.Thetopthreecompetitionparticipants were also outperformed in all categories and their systemobtainedsecondplaceinPlantCLEF2016.

VI. Leaf Classification Project:

Usingashareddatasetofleafattributes,thispaper [6]analysesalternativemachinelearningmethodsfor classifying leaves. To analyze among effective and ineffectivecategorizationmodels,theauthorscompare andanalyzethemodeloutputs.

VII. Plant Recognition System based on Leaf Image:

Thepaper[7]suggestsanimage-basedautomatic identification method based on leaf structure. To identifyplantsfromphotographsofleaves,thesystem makes use of attributes like shape, texture, vein structure,andcolor.Forthepurposeofstoringimage dataandrelatedinformation,theauthorsadditionally createacloud-baseddatabase.

VIII. Leaf shape extraction for plant classification:

Inordertoclassifyplants,thispaper[8]focuseson leaf form extraction from photos. In order to extract biometric properties of leaves for categorization, the authorssuggestemployingavarietyofoperatorsand image processing approaches. This paper states the necessityforautomatedmethodsandshowshowtimeconsumingmanualclassificationis.

IX. A study on plant recognitionusingconventionalimage processing and deep learning approaches:

Thepaper[9]proposestwoapproaches,traditional and deep learning, to address this issue. In the traditionalapproach,shape,texture,andcolorfeatures are extracted, and various classifiers are used for classification. The deep learning approach involves testingdifferentdeeplearningarchitecturesforplant species recognition. Four datasets, including three standarddatasetsandonereal-timedataset,areused forevaluation.TheresultsdemonstratethattheVGG16 CNN models outperformed traditional methods in terms of accuracy. The combination of color channel statistics,LBP,Hu,andHaralickfeatureswithaRandom Forestclassifierachievedaplantidentificationaccuracy of82.38%fortheLeaf12datasetusingthetraditional method. VGG 16 CNN architecture with logistic regressionachievedagreateraccuracyof97.14%for theLeaf12dataset,whileVGG19CNNarchitecturewith logisticregressionachievedanaccuracyof96.53%for Folio, 96.25% for Flavia and 99.41% for Swedish datasets,respectively.

X. Plant Identification Methodologies using Machine Learning Algorithms:

Themethodsusedarewhatdeterminehowaplant isidentified,itisaprocessthathasevolvedoverages. Identificationofplantsisimportantbecauseitenables the retrieval of necessary data related to various species,whichisnecessaryforcertainapplications.This paper [10] offers numerous methods and strategies fromvariouswritersforidentifyingplants.

XI. Identification of Plants using Deeplearning:AReview:

Traditionalmethodsofplantidentificationbasedon physical characteristics can be time-consuming and challenging. To address this issue, researchers have exploredtheuseofadvancedtechnologiessuchasdeep learning and image recognition to develop more efficient plant identification methods. In a review of academicliteraturepublishedbetween2015and2020, it has been observed that convolutional neural networks (CNNs), a type of deep learning algorithm, have shown promising results in the area of plant identification. This has led to the development of various techniques and methods for leaf recognition usingCNNs.Thispaper[11]aimstocontributetothe existing body of literature on plant identification by discussingtheconceptsofdeeplearninganddifferent leaf recognition methods. By analyzing the latest researchinthisfield,thepaperprovidesanacademic database of knowledge that can be used to improve plant identification and further advance the field of ecology.

4. CONCLUSION

Manuallyidentifyingplantscanbeareallytiresomeprocess. So, to make this task easy automated methods such as machine learning and deep learning models can be implemented.Inthispaper,differentmachinelearningand deep learning a0lgorithms for the purpose of plant recognition through their leaves have been reviewed. Authorsofthesepapershavesuggestedvarioustechniques in order to achieve highest accuracy possible. These techniques include algorithms such as Random Forest, KNearestNeighbors,SVMclassifier,LogisticRegression,etc., andpopulardeeplearningarchitecturessuchasGoogleNet, AlexNet, VGGNet and these techniques give different accuraciesifusedonasingledataset.

5. REFERENCES

[1] Aparajita Sahay, Min Chen, “Leaf Analysis for Plant Recognition”,20167thIEEEInternational Conference onSoftwareEngineeringandServiceScience(ICSESS)

[2] Min Gao, Lang Lin, Richard O. Sinnott, “A Mobile Application for Plant Recognition through Deep

Learning”,2017IEEE13thInternationalConferenceon eScience

[3] HongZhou,ChenjunYan,HuahongHuang,“TreeSpecies IdentificationBasedonConvolutionalNeuralNetworks”, 2016 8th International Conference on Intelligent Human-MachineSystemsandCybernetics

[4] Redha Ali, Russell Hardie, Almabrok Essa, “A Leaf Recognition Approach to Plant Classification Using MachineLearning”,2018IEEE

[5] Ghazi, M.M., “Neurocomputing” (2017), http://dx.doi.org/10.1016/j.neucom.2017.01.018

[6] Jiacheng Hu, Yitao Liu, Jia Liu, “Leaf Classification Project”,2020ECE228

[7] SwatiP.Raut,Dr.A.S.Bhalchandra,“PlantRecognition SystembasedonLeafImage”,ProceedingsoftheSecond InternationalConferenceonIntelligentComputingand ControlSystems(ICICCS2018)

[8] M.M.Amlekar,A.T.Gaikwad,R.R.Manza,P.L.Yannawar, “Leaf Shape Extraction for Plant Classification”, 2015 International Conference on Pervasive Computing (ICPC)

[9] S. Anubha Pearline, V. Sathiesh Kumar, S. Harini, “A study on plant recognition using conventional image processing and deep learning approaches”, Journal of Intelligent&FuzzySystems36(2019)

[10] Skanda H N, Smitha S Karanth, Suvijith S, Swathi K S, Pragati P, “Plant Identification Methodologies using MachineLearningAlgorithms”,InternationalJournalof EngineeringResearch&Technology(IJERT)Vol.8Issue 03,March-2019

[11] RakibulSk,AnkitaWadhawan,“IdentificationofPlants Using Deep Learning: A Review”, International SymposiumonIntelligentControl(ISIC)2021