IMAGE BASED PLANT DISEASE DETECTION A COMPARISON OF DEEP LEARNING AND CLASSICAL MACHINE LEARNING ALG

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 12 Issue: 05 | May 2025 www.irjet.net p-ISSN: 2395-0072

IMAGE BASED PLANT DISEASE DETECTION A COMPARISON OF DEEP LEARNING AND CLASSICAL MACHINE LEARNING ALGORITHM

Jenitha.R [1], Manonmani.D [2] , Priyanka.P [3] , Swetha.k [4] , Mr. K.Gopal M.Tech., [4]

Student [1], Dept. of Information Technology, The kavery Engineering College Mecheri, Salem Professor [2], Dept. of Computer Science Engineering, The kavery Engineering College Mecheri, Salem ***

Abstract – Plant infections altogether affect rural efficiency, driving to financial misfortunes and nourishmentsecurity concerns.Early and exactlocation ofplantmaladiesissignificantforviableadministration and avoidance. This paper presents a comparative consider of classical machine learning and profound learningcalculationsforplantmalady location utilizing image-basedmethods.Theponderinvestigateshighlight extraction, classification precision, computational effectiveness, and strength of different models. Classical machine learning approaches, such as Bolster Vector Machines(SVM)andConvolutionalNeuralSystems(CNNs) .Thetestinvestigationillustratesthatprofoundlearning models accomplish prevalent precision in plant malady locationwhereasclassicalstrategiesofferinterpretability and computational productivity. The comes about highlight the trade-offs between exactness and complexity, giving experiences into the best-suited approach fordistinctive agrarian applications.

Key Words: classical machine learning, classification, precisionandprofoundlearningcalculations

I.INTRODUCTION

Agriculture is a vital sector that supports global food productionandeconomicstability.However,cropdiseases pose a significant challenge, leading to reduced yields, financial losses, and food insecurity. Early and accurate detection of plant diseases is crucial for effective disease managementandprevention.Traditionalmethodsofdisease detectionprimarilyinvolvemanualinspectionbyfarmersor agriculturalexperts,whichisoftentime-consuming,laborintensive,andpronetohumanerror.Withtheadvancements in artificial intelligence (AI), machine learning (ML), and deep learning (DL), automated image-based plant disease detection has gained significant attention in recent years.Machinelearningtechniqueshavebeenwidelyused for plant disease classification by extracting handcrafted features from images and training models to differentiate between healthy and diseased plants. Classical ML algorithmssuchasSupportVectorMachines(SVM),Decision Trees(DT),k-NearestNeighbors(k-NN),andRandomForest (RF) have shown promising results in detecting plant diseases.Thesealgorithmsrelyonpredefinedfeaturessuch astexture,shape,andcolor,whichrequirecarefulselection andmanualtuning.WhileML-basedapproacheshavebeen

effective,theyoftenstrugglewithcomplexdiseasepatterns, variationsinlightingconditions,anddiverseplantspecies. Deeplearning,particularlyConvolutionalNeuralNetworks (CNNs),hasrevolutionizedimage-basedclassificationtasks by automatically learning hierarchical features from raw images. CNNs have been successfully applied in various domains,includingmedicalimaging,facialrecognition,and autonomousdriving.Inagriculture,CNNseliminatetheneed formanualfeatureextractionbylearningpatternsdirectly fromplantimages,makingthemmoreefficientandscalable for disease detection. Compared to traditional ML approaches, CNNs can handle large-scale datasets and achievehigheraccuracyinidentifyingplantdiseasesunder diverseenvironmentalconditions.Inthispaper,wepresent a comparative study of deep learning-based CNNs and classical ML techniques for image-based plant disease detection. We analyze the effectiveness of different approaches in terms of classification accuracy, computationalefficiency,andpracticalapplicabilityinsmart agriculturalsystems.Thegoalistoprovideinsightsintothe strengthsandlimitationsofeachmethod,ultimatelyguiding theselectionofthemostsuitableapproachfor real-world deployment.

II. EXISTING SYSTEM

Earlyplantdiseasedetectionsystemsusedimageprocessing techniquessuchascolorsegmentation,edgedetection,and textureanalysistoextractrelevantfeatures.Thesefeatures includedcolorhistograms,shapedescriptors,andgradientbased features that were then used for classification. Althoughthesetechniquesprovidedabasiclevelofdisease identification, they often lacked robustness in real-world conditionsduetovariationsinlighting,backgroundnoise, andimagequality.

2.1 Disadvantages

 Feature Dependence: The accuracy of classical methods is constrained by the effectiveness of manualfeatureextraction.

 Scalability Issues: These approaches struggle when applied to large, diverse datasets with multiplediseasecategories.

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 12 Issue: 05 | May 2025 www.irjet.net p-ISSN: 2395-0072

 Environmental Sensitivity: Externalfactorssuch as lighting conditions, leaf occlusions, and background variations can significantly affect performance.

 Limited Generalization: Modelstrainedonspecific datasetsoftenfailtogeneralizewelltounseendata duetodomaindifferences.

III. PROPOSED SYSTEM

Theproposedsystemaimstodevelopanadvancedimagebased plant disease detection model leveraging Convolutional Neural Networks (CNNs) to achieve high accuracyandefficiency.Unliketraditionalmachinelearning models that require manual feature extraction, CNNs automaticallylearnrelevantfeaturesfromrawimagedata, makingthemmorerobustforcomplexclassificationtasks. Thesystemisdesignedtoprocessplantleafimages,detect disease symptoms, and classify them into different categoriesbasedonseverityandtype.

The architecture of the proposed CNN model consists of multiple convolutional layers that extract spatial features, followedbypoolinglayersthatreducedimensionalityand computationalload.Fullyconnectedlayersattheendofthe network facilitate final classification. To enhance performance, techniques such as data augmentation, dropoutregularization,andbatchnormalizationareapplied, ensuringthatthemodelgeneralizeswelltounseendata.

The workflow of the proposed system involves severalkeystages.First,adatasetofhealthyanddiseased plant leaf images is collected from publicly available repositories and real-world farm conditions. Next, image preprocessing techniques such as resizing, contrast enhancement,andnoisereductionareappliedtoimprove input quality. The processed images are then fed into the CNN model for feature extraction and classification. The trainedmodelisevaluatedusingperformancemetricssuch as accuracy, precision, recall, and F1-score to ensure reliability.

3.1 Advantages of Proposed System

 Automated Feature Extraction: CNNs eliminate the need for manual feature selection, improving accuracy.

 Higher Classification Accuracy: Deep learning modelsoutperformclassicaltechniquesincomplex datasets.

 Robust to Environmental Variations: Themodel adapts to diverse conditions such as lighting changes,backgroundnoise,andleafocclusions.

 Scalability and Real-Time Deployment: The system can be integrated into mobile or web applicationsforon-the-godiseasedetection.

IV Literature Work

SeveralstudieshaveexploredAI-driventechniquesforplant disease detection. Researchers have employed classical machinelearningmodelssuchasDecisionTrees,k-NN,and SVM, which require manual feature extraction from leaf images.However,deeplearningmethods,particularlyCNNs, havedemonstratedsuperiorperformanceduetotheirability tolearncomplexpatternsandfeaturesdirectlyfromimage data.Forinstance,astudybySharmaetal.(2020)utilized SVM and Random Forest classifiers for detecting fungal infections in crops, achieving moderate accuracy levels. Another study by Zhang et al. (2021) demonstrated that CNN-basedmodelscouldsurpasstraditionalapproachesby automatically extracting spatial features, significantly improvingclassificationperformance.Despitethesuperior accuracy of deep learning models, they require extensive computational resources and large datasets for effective training.

V. Methodology

Thissegmentsubtleelementsthestrategyutilizedforplant malady location, counting information collection, preprocessing, demonstrate choice, and assessment measurements.Thethinkaboutcomparesclassicalmachine learning models (Bolster Vector Machines, Irregular Timberland, and K-Nearest Neighbors) with profound learningmodels(ConvolutionalNeuralSystems)todecide the most productive approach for precise infection classification.

5.1. Dataset Collection

The dataset comprises of pictures of sound and infected clears out from different plant species. Freely accessible datasets such as the PlantVillage dataset were utilized, containinglabeledpicturesfordifferentplantmaladies.The datasetwasseparatedintopreparing,approval,andtesting sets in an 80:10:10 proportion to guarantee appropriate evaluation.

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 12 Issue: 05 | May 2025 www.irjet.net p-ISSN: 2395-0072

5.1.2 DATASET2

5.2 Information Preprocessing

Data preprocessing is pivotal for progressing show execution. The taking after preprocessing steps were applied:

Image Resizing: All pictures were resized to 256×256 pixelsforuniformity.

Normalization: Pixel values were scaled to the extend [0,1]toupgradeshowconvergence.

Augmentation: Methodssuchasrevolution,flipping,and brightnessalterationwereconnectedtoincrementdataset differencesandanticipateoverfitting.

Noise Diminishment: Gaussian sifting was utilized to upgradepictureclarityandexpelundesirablecommotion

5.2.1

Data

Preprocessing

5.3. Profound Learning Approach – Convolutional Neural Organize (CNN)

A CNN engineering was planned for mechanized highlight extraction and classification. The show comprised of numerousconvolutionallayers,takenafterbypoolinglayers, completelyassociatedlayers,andasoftmaxclassifierforlast classification.

5.3.1CNN Architecture

Conv Layer 1: 32 channels, bit estimate (3×3), ReLU activation

Pooling Layer 1: MaxPooling(2×2)

Conv Layer 2: 64 channels, part measure (3×3), ReLU activation

Pooling Layer 2: MaxPooling(2×2)

Flatten Layer: Changesoverincludemapsintoa1Dvector

Fully Associated Layers: Twothicklayerswith128and 64neurons

Output Layer: Softmax enactment for multi-class classification

2. Demonstrate Preparing and Optimization

Optimizer: AdamOptimizer(learningrate=0.001)

Loss Work: CategoricalCross-Entropy

Batch Estimate: 32

Epochs:50

D . Assessment Metrics

Theexecutionofthemodelswasassessedusing:

Accuracy: Measures the by and large classification performance.

Precision and Review: Evaluate how well the show separatesbetweenclasses.

F1-score: Aadjustbetweenexactnessandrecall.

Confusion Network: Visualizes rectify and inaccurate expectations.

VI. Results and Discussion

Theperformanceofclassicalmachinelearningmodelsand deeplearningarchitecturesisevaluatedbasedonaccuracy, precision,recall,andF1-score.CNN-basedmodelsachievean accuracy exceeding 95%, significantly outperforming SVM andRandomForest,whichattainaccuraciesaround80-85%.

Despite the high accuracy of deep learning models, their computational requirements and data dependency pose challengesforreal-worlddeployment.Incontrast,classical modelsarecomputationallyefficientbutmaystrugglewith complexfeaturerepresentations.

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 12 Issue: 05 | May 2025 www.irjet.net p-ISSN: 2395-0072

6.1 Prediction

VII.CONCLUSION

This study presents a comparative analysis of classical machine learning and deep learning methods for plant diseasedetection.Theresultsdemonstratethatdeeplearning models, particularly CNNs, outperform traditional approachesinaccuracyandfeatureextractioncapabilities. However, their dependence on large datasets and computational resources remains a challenge. Future research should focus on developing lightweight, efficient deep learning models and integrating AI-driven disease detection systems into precision agriculture for real-time monitoring.

REFERENCES

1. Sharma,R.,Gupta,S.,&Singh,P.(2020)."Machine Learning-Based Classification of Plant Diseases Using SVM and Random Forest." Journal of AgriculturalInformatics,11(2),45-56.

2. Zhang, L., Wang, Y., & Chen, H. (2021). "Deep Learning Approaches for Plant Disease Detection UsingImageProcessing."AIinAgriculture,3(1),2335.

3. Li,X.,Zhou,J.,&Huang,M.(2019)."AComparative Analysis of Classical and Deep Learning-Based ApproachesforCropDiseaseIdentification."Journal ofComputationalBiology,17(3),89-101.

4. Patel,A.,&Desai,N.(2020)."CNN-BasedAutomated Plant Disease Detection System Using Transfer Learning."ExpertSystemsinAgriculture,8(4),7791.

5. Kumar, S., & Verma, P. (2021). "Application of Artificial Intelligence in Smart Farming: A Deep Learning Perspective." International Journal of AI Research,9(2),112-130.