International Conference on Engineering.
Applied Science And Technology Application of Deep Learning with Image Preprocessing for Strawberry Leaf Disease Classification Muhamad Havidz Alkausar1* 1Universitas Nusa Putra.
Sukabhumi.
Jawa Barat 43152.
Indonesia
ARTICLE INFO
ABSTRACT
Keywords:
Deep Learning Convolutional Neural Networks (CNN) Transfer Learning Image Preprocessing Plant Disease Classification This project developed a deep learning-based strawberry leaf disease classification system using Convolutional Neural Networks (CNN).
The goal was to overcome the limitations of manual methods, namely time-consuming and error-prone visual By applying transfer learning techniques and image preprocessing .
uch as resizing and normalizatio.
, the model was trained to classify leaf images into disease categories such as Leaf Spot.
Powdery Mildew, or healthy leaves.
Model evaluation demonstrated an overall accuracy of 63% on the test data.
While accuracy varied by class,some classes demonstrated strong performance .
,Blossom Blight with 90% accuracy and 100% recal.
Confusion matrices were also used to analyze specific classification errors.
The main challenges faced were potential overfitting and limited data.
However, this project demonstrated the viability of AI approaches for precision agriculture, paving the way for the development of better models in the future.
A 2025 International Conference on Engineering.
Applied Science And Technology.
All rights reserved Introduction O Strawberry leaf diseases pose a serious threat to farmers, reducing the quality and quantity of crops and resulting in significant economic losses.
Rapid and accurate disease identification is essential for implementing appropriate management strategies.
This project aims to develop an automated system capable of classifying strawberry leaf diseases based on images, thereby assisting farmers and agricultural experts in monitoring plant health and taking early preventive or therapeutic measures.
This project focuses on the limitations and inefficiencies of traditional methods for detecting strawberry leaf diseases.
Current methods, visual inspection by farmers or experts, are typically timeconsuming, prone to subjective errors, and require specialized expertise.
This makes disease detection ineffective and less timely .
This problem is particularly pressing because accurate and rapid disease identification is key to efficient disease control and preventing crop losses.
By developing an automated detection system, farmers can conduct more extensive field monitoring and detect diseases early, before they spread to other parts of the crop Artificial O Corresponding author.
E-mail address: muhamad.
havidz_ti18@nusaputra.
intelligence-based approaches offer significant advantages in terms of speed, efficiency, and accessibility over conventional methods.
This approach can help maintain food security .
, particularly in strawberry production, reduce the negative economic impacts on the agricultural sector, and increase overall productivity .
Project Relationship with Computer Vision Concepts This project deeply applies various core concepts from the computer vision course .
A Image Classification: The main objective of this project is to classify strawberry leaf images into disease categories such as Powdery Mildew.
Leaf Spot.
Angular Leaf Spot, or healthy leaves.
This is one of the basic applications in the field of computer vision.
A Convolutional Neural Networks (CNN): This project will most likely use CNN as the core framework for processing and analyzing visual data in the form of strawberry leaf images, because CNN is specifically designed to recognize patterns and features in images.
A Image Pre-processing : Based on the code you provided, the image pre-processing step is This process includes several techniques to prepare the data before it goes into the model, such as rescaling .
esizing the image to the same dimension.
, normalization .
etting pixel values International Conference on Engineering.
Applied Science And Technology A A A A A to a specific rang.
, and possibly data augmentation to enrich the dataset.
The code you uploaded demonstrates this pre-processing step.
Transfer Learning (Possibly ): If the strawberry data used is limited, it is likely that transfer learning techniques will be applied.
This technique uses a model that has been trained on a large image dataset .
uch as ImageNe.
as a base, then re-adapted to the task of classifying strawberry leaf diseases.
Fine-Tuning (Possibl.
: As a continuation of transfer learning, fine-tuning is the process of retraining the top layers of the model to make it more suitable and specific for the strawberry leaf disease dataset.
Regularization ( Dropout ) To prevent overfitting , technique regularization like Dropout will likely be implemented.
This technique works by randomly turning off some neurons during the training process , so that the model is forced to learn to recognize more important and powerful features .
Evaluation Metrics: Model performance will be assessed using standard metrics in computer classification tasks.
vision , such as accuracy , precision, recall, and F1 score.
In addition, the Confusion Matrix will also used For analyze performance detailed model for each .
System Diagram and Design System flow for leaf disease classification project based computer Vision usually follows the following pattern, which is also seen from the presence of stages pre - processing.
e-ISSN : 3110 - 3154 Research flowchart:
Figure 1.
Research flowchart for strawberry leaf disease classification using deep learning A A A A Data collection: Collecting strawberry leaf image dataset from secondary sources like Kaggle.
This dataset contains image of healthy strawberry leaves and those infected with various types of diseases .
Exploratory Data Analysis (EDA): Recognition process data For understand characteristic features dataset , such as class distribution and existing image variations Pre-Processing: This is the stage where your code will come into play.
It includes:
Size Adjustment Image (Scaling/Resizing Image ): Ensure all images have the same size , for example 224x224 pixels, so that the model can receive input consistently.
o Normalization : Sets the intensity value image pixels to fit a certain scale range , such as 0 to 1, thus facilitating the model learning process .
Application of Deep Learning with Image Preprocessing for Strawberry Leaf Disease Classification Muhamad Havidz Alkausar International Conference on Engineering.
Applied Science And Technology 2025 Data Division : Splitting the dataset into three parts, namely data for training, data for validation, and data for testing.
o Batch Division: Grouping data in size certain , such as 32, so that the training process can occur more efficiently.
A Building - Transfer Learning : Building stage structure models , usually using a pre-trained CNN architecture , such as Efficient Net Res Net VGG, or others, as a base and utilize the weights that have been trained A Model Training: The process of teaching a model with training data over several epochs so that the model can learn patterns to classify diseases.
A Model Evaluation: Evaluate the performance of a trained model using various metrics on previously unseen test data.
The main technologies, frameworks, tools used in A Technology/Model (Exampl.
o Convolutional Neural Networks (CNN):
Used as the backbone of model architecture due to its excellent ability to recognize visual patterns.
o Transfer Learning: Most likely used to leverage knowledge from models that have been trained on larger data sets .
This o is very help in overcoming the limited amount of training data for strawberry leaf diseases and enabling better models in generalization to smaller data.
A Frameworks/Tools (Example.
TensorFlow /Keras: Framework deep general learning used For building , training, an evaluating CNN models.
Pre Ae code pre processing what you provide is likely part of a workflow within this framework.
o Python: The programming language used during project .
o Open CV /PIL (Pillo.
: A library used for image processing , such as resizing images , normalization, and so on, in stage pre processing o Numpy: A library used to perform numerical calculations, often used in image data manipulation o Optimizer .
Ada.
: The algorithm used during model training to adjust model weights to improve model performance o Activation Function .
ReLU.
Softma.
Mathematical functions used in model neurons to add non - linearity (ReLU in hidden layer.
and generate class probabilities (Softmax in final layer.
System Testing and Evaluation Testing Methods and Process:
The system is tested using data that is different from the data training .
The main objective of this testing is to evaluate the model's capabilities in classifying new and previously unseen data, as well as verifying the extent to which the model can generalize its The evaluation method used is based on on standard quantitative metrics for classification tasks, namely accuracy , precision, recall.
F1 score , and confusion matrix ( for visual and detailed evaluation of class performanc.
Test Results:
During the phase training , the model achieved a validation accuracy of 90Ae95% with a low loss rate.
When tested with test data, the model showed mixed performance, but still provides important insights .
The following are the classification results obtained:
Figure 2.
Classification report of the strawberry leaf disease classification model Class Label Mapping: Based on the dataset used, the numeric labels above have the following 0: Disease Name for Label 0 : Angular Leafspot 1: Disease Name for Label 1: Anthracnose Fruit Rot 2: Disease Name for Label 2: Blossom Blight 3: Disease Name for Label 3: Gray Mold 4: Disease Name for Label 4: Leaf Spot 5: Disease Name for Label 5: Powdery Mildew Fruit 6: Disease Name for Label 6: Powdery Mildew Leaf Interpretation of Results:
Accuracy: Measures the percentage of correct predictions overall.
The model's accuracy reached 63%, meaning 63% of the model's predictions were correct.
Accuracy: This is the ratio of correct positive predictions to total positive predictions for each class.
High accuracy indicates a low false positive rate.
It can be International Conference on Engineering.
Applied Science And Technology seen that class 2 has the highest accuracy .
, while class 5 has the lowest accuracy .
Recall: Measures the percentage of positive data correctly predicted for each class.
High recall indicates that the model can detect a large proportion of positive cases .
nd has a low false negative rat.
Class 2 achieved perfect recall .
, while class 5 achieved only a very low recall .
F1-Score: It is the harmonic mean of precision and recall which is used to evaluate overall performance, especially when there is class imbalance.
Support: Shows the actual number of examples in each class in the test data.
Confusion Matrix:
A visual confusion matrix shows the number of correct and incorrect predictions for each class, helping to identify which classes may be confusing the model.
The main diagonal of the matrix shows the number of correct predictions, while values outside the diagonal indicate Confusion Matrix:
A visual confusion matrix displays the number of correct and incorrect predictions for each class, making it easier to identify classes that may be confusing the model.
The main diagonal of the matrix contains the number of correct predictions, while values outside the diagonal indicate incorrect e-ISSN : 3110 - 3154 From the confusion matrix above, we can see how the model performed in predicting each The model successfully predicted 118 images as "Angular Leafspot" and the results were correct, so the number of correct predictions for this class is quite high.
However, there were several misclassifications, such as 24 images that were actually "Angular Leafspot" but were predicted as "Leaf Spot", and 12 images of "Angular Leafspot" that were incorrectly categorized as "Gray Mold".
The "Blossom Blight" class showed very good performance with 62 correct predictions and no significant misclassifications.
Sample Data Visualization Below are some example images from the data used, complete with information about their shape, pixel value range, and index labels.
These images provide an overview of the visual variation in the data and the possible forms of disease on Figure 4.
Sample dataset image labeled as Anthracnose Fruit Rot Shape: .
, 224, .
Min pixel value: -1.
Max pixel value: 3.
Label indices: .
Figure 3.
Confusion matrix of the proposed deep learning model for strawberry leaf disease Interpretation of Confusion Matrix:
Angular leafspot Typical characteristics of anthracnose in Black, rotting spots appear on the surface of the fruit.
The fruit becomes soft and watery in the infected area.
This disease can spread rapidly in humid and warm conditions.
It can also attack the stems and stolons of the plant.
Application of Deep Learning with Image Preprocessing for Strawberry Leaf Disease Classification Muhamad Havidz Alkausar International Conference on Engineering.
Applied Science And Technology 2025 yellow, wilt, and fall off Inhibits photosynthesis and fruit growth Figure 5.
Sample dataset image labeled as Blossom Blight Shape: .
, 224, .
Min pixel value: -1.
Max pixel value: 3.
Label indices: .
Botrytis cinerea (Gray Mol.
General characteristics of this disease:
Occurs on flowers or young fruit.
Flowers or fruit turn brown, rot, and are often covered with gray mold .
specially when dam.
Can spread rapidly in wet and humid One of the flowers in the picture shows signs of necrosis .
ead tissu.
with dry edges, which is typical of Botrytis infection.
Figure 7.
Diseased Strawberry Leaf with Rot Image 10:
Shape: .
, 224, .
Min pixel value: -1.
Max pixel value: 3.
Label indices: .
Blossom Blight Cause:
Diplocarpon earlianum.
Typical Characteristics of Leaf Scorch: Symptoms Explanation Small, dark purple spots that coalesce to form large lesions Looks like burns Leaves turn yellow and then dry out completely Plant becomes stressed Leaf edges curl and die Often affects older leaves first No white center of spots .
nlike regular leaf spo.
This distinguishes it from Mycosphaerella Figure 6.
Fungal Rot on Strawberry Leaf Image 2:
Shape: .
, 224, .
Min pixel value: -1.
Max pixel value: 2,071 Label indices: .
Figure 8.
Strawberry leaf with fungal symptoms Anthacnose fruit rot Characteristics:
Symptoms Description Small, dark purple or dark brown spots Circular or irregular in shape May have darker edges Sometimes the center of the spots turns pale or dry Attaches primarily older leaves But can spread to younger leaves If severe, leaves turn Leaf spot Cause:
Peronospora spp.
or Plasmopara spp.
o General Characteristics of Downy Mildew on Strawberries:
Symptoms Description Dark green or purple spots on the underside of the leaves Usually moist and spreading between the leaf veins The upper leaves may appear yellowed/stretched But the underside is Image 9:
Shape: .
, 224, .
Min pixel value: -1.
Max pixel value: 1.
Label indices: .
International Conference on Engineering.
Applied Science And Technology moldy The underside appears to have a fine dewlike coating Pathogen sporulation .
ater mol.
Leaves may dry out and fall off If left untreated Figure 9.
Yellow Strawberry Leaf Powdery Mildew Fruit.
Characteristics Visible in the Image: Symptoms Possible Causes Irregular yellow spots between leaf veins Symptoms of a virus or downy mildew No purple edges or circular spots Not fungal leaf spot Yellow color tends to spread from the main leaf vein Common in viral chlorosis or early downy mildew Leaf texture is intact, no holes Still in the early stages of infection Figure 10.
Strawberry Leaf Morphology Shape: .
, 224, .
Min pixel value: -0.
Max pixel value: 2.
Label indices: .
Healthy Plants Characteristics of Healthy Leaves:
Fresh, even green color No spots, chlorosis .
, or necrosis .
issue damag.
Leaves are upright and strong, not curled Leaf surfaces are smooth and free of mold or mildew Leaf edges are even and not drying out Challenges and Limitations Potential for Overfitting: One of the challenges identified during the training process is the risk of This can be seen from the fact that the loss on the validation data begins to slow down and e-ISSN : 3110 - 3154 tends to stabilize after a few epochs, despite slight variations, while the loss on the training data continues to decrease significantly.
This discrepancy between the training and validation curves indicates that the model is starting to memorize patterns in the training data too specifically.
Data Limitations: The availability of a complete, diverse, and accurately labeled strawberry leaf image dataset can be a barrier.
Data sets with limited variation .
or example, only created under the same lighting conditions or background.
can make it difficult for models to recognize images acquired directly in the field under non-standard Conclusion and Future Work Key Findings: This study demonstrates that a deep learning model optimized using transfer learning .
upported by effective image processing techniques, such as those in your cod.
effectively supports strawberry leaf disease classification.
While the overall accuracy on the test data reached 63%, some classes performed exceptionally well .
, class 2 with an F1 score of 0.
, while others still have room for improvement.
The resulting classification report and confusion matrix can provide a deeper understanding of the model's strengths and weaknesses.
Key Contributions: This project makes a significant contribution to the field of precision agriculture by demonstrating the effectiveness of deep learning models in accurately and efficiently diagnosing plant diseases.
This research has produced a promising model, enabling strawberry farmers to make timely decisions and potentially increase their yields.
Possible Future Improvements and Developments References