A system for classifying apple leaf diseases using deep learning and feature fusion

Abstract

A system for classifying apple leaf diseases using deep learning and feature fusion, consisting of: A data input module, which includes a data storage module, is configured to store a data set created using various data sources, with the data set containing apple leaf images being derived from the data sets "Apple Leaf 9", "Kashmiri Apple Plant Disease" and "Plant Village Apple Leaf"; a data preprocessing module configured to perform preprocessing of the newly prepared dataset of apple leaves, wherein the data preprocessing module is configured to perform data decoding, expansion, resizing, segmentation, scaling and color conversion of the input image data; a feature extraction module that is operationally connected to the data processing module and is configured to receive preprocessed data and transfer the preprocessed data to one or more convolutional neural network models for feature extraction; a feature fusion module configured to combine the extracted features from the Convolutional Neural Networks to develop a fused feature vector; a hyperparameter optimization module configured to optimize the hyperparameters of the feature extraction models by implementing a particle swarm optimization algorithm; a classification module configured to classify the fused trait vector into 13 apple leaf disease classes using a random forest classifier; and a user interface connected to the classification module, configured to display the classification results.

Description

AREA OF INVENTION

[0001] The present disclosure relates to a system for classifying apple leaf diseases using deep learning and feature fusion. More specifically, the present invention relates to a deep learning (DL)-based system called DensoMobileForestNet for classifying apple leaf diseases. The system uses feature fusion by combining the outputs of convolutional neural networks and performs classification using random forest (RF) to improve the accuracy of disease classification. BACKGROUND OF THE INVENTION

[0002] Current agricultural disease detection systems face significant technical limitations that prevent accurate and efficient classification of plant diseases. Traditional manual inspection methods suffer from subjectivity, time constraints, and scalability issues, particularly on large farms. Existing deep learning-based classification systems encounter persistent challenges, including inconsistent data labeling, class imbalances, and environmental variability, all of which negatively impact model performance and reliability.

[0003] Modern computer vision systems for plant disease detection are typically based on single convolutional neural network architectures. This limits their ability to capture comprehensive feature representations necessary for accurate disease classification. These systems often struggle with hyperparameter optimization and rely on computationally intensive raster or random search methods that do not achieve optimal performance configurations. Furthermore, existing systems lack robust feature fusion mechanisms that can leverage the complementary strengths of multiple deep learning models to improve classification accuracy.

[0004] The lack of intelligent hyperparameter optimization in current systems leads to suboptimal model performance, requiring extensive manual adjustments and expertise. Conventional optimization approaches are inefficient when exploring the complex hyperparameter space, resulting in longer development cycles and reduced system effectiveness. Furthermore, most existing systems operate with limited datasets from single sources, which impairs their generalizability across diverse agricultural environments and disease manifestations.

[0005] Current classification systems also lack comprehensive preprocessing modules capable of handling the fluctuations in image quality, orientation, and environmental conditions that frequently occur in agriculture. The inability to effectively process datasets from multiple sources and adapt to varying imaging conditions significantly limits the practical applicability of existing systems for plant disease detection in real-world agricultural scenarios.

[0006] Therefore, there is an urgent need for an advanced system that integrates multiple deep learning architectures through intelligent feature fusion, employs metaheuristic optimization algorithms for automatic hyperparameter optimization, and includes comprehensive preprocessing capabilities to overcome the limitations of current technologies for classifying plant diseases. Summary of the invention

[0007] This disclosure relates to a system for classifying apple leaf diseases using deep learning and feature fusion. The deep learning-based system for automated apple leaf disease classification integrates multiple convolutional neural network models using feature fusion techniques. The system processes apple leaf images from various datasets, extracts features using MobileNetV2 and DenseNet201 models, fuses these features into a combined vector, and uses a random forest classifier to identify diseases into 13 classes with increased accuracy through particle swarm optimization.

[0008] This disclosure relates to providing a system for classifying apple leaf diseases using deep learning and feature fusion. The system comprises: a data input module with a data storage module configured to store a dataset created from various data sources, wherein the apple leaf image dataset is derived from the "Apple Leaf 9," "Kashmiri Apple Plant Disease," and "Plant Village Apple Leaf" datasets; a data preprocessing module configured to preprocess the newly created apple leaf dataset, wherein the data preprocessing module is configured to decode, expand, scale, segment, and color-convert the input image data;a feature extraction module operationally connected to the data processing module and configured to receive preprocessed data and transfer the preprocessed data to one or more convolutional neural network models for feature extraction; a feature fusion module configured to combine the extracted features from the convolutional neural networks to create a fused feature vector; a hyperparameter optimization module configured to optimize the hyperparameters of the feature extraction models by implementing a particle swarm optimization algorithm; a classification module configured to classify the fused feature vector into 13 apple leaf disease classes using a random forest classifier; and a user interface connected to the classification module configured to display the classification results.

[0009] One objective of the present disclosure is to provide a system for classifying apple leaf diseases using deep learning and feature fusion.

[0010] Another objective of the present disclosure is to provide an automated disease classification system that can accurately identify and classify apple leaf diseases using advanced deep learning techniques, thereby reducing the need for manual inspections and providing rapid diagnostic capabilities for agricultural applications.

[0011] Another objective of this disclosure is to improve classification accuracy through feature fusion by combining the strengths of several convolutional neural network models (MobileNetV2 and DenseNet201) to create a more robust and comprehensive feature representation for disease detection.

[0012] Another objective of the present disclosure is to optimize system performance through intelligent hyperparameter tuning using particle swarm optimization algorithms to automatically select the best configuration parameters and thus ensure maximum classification accuracy while minimizing computational effort.

[0013] To further clarify the advantages and features of the present disclosure, the invention is explained in more detail with reference to specific embodiments illustrated in the accompanying drawings. These drawings merely show typical embodiments of the invention and are therefore not to be understood as limiting its scope. The invention is described and explained more precisely and in greater detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE FIGURES

[0014] These and other features, aspects, and advantages of the present disclosure will be better understood if the following detailed description is read with reference to the accompanying drawings, in which identical symbols consistently represent identical parts. The following applies: Fig. Figure 1 shows a block diagram of a system for classifying apple leaf diseases using deep learning and feature fusion according to an embodiment of the present disclosure. Fig. Figure 2 shows a diagram illustrating the operation of the proposed system according to an embodiment of the present disclosure.

[0015] Experts will also recognize that the elements in the drawings are presented for the sake of simplicity and are not necessarily to scale. For example, the flowcharts illustrate the process by highlighting the main steps to enhance understanding of the aspects of this disclosure. Furthermore, with regard to the design of the device, one or more components of the device may be represented in the drawings by conventional symbols, and the drawings may show only the specific details relevant to understanding the embodiments of this disclosure, so as not to clutter the drawings with details that are readily apparent to those skilled in the art after reading this description. DETAILED DESCRIPTION:

[0016] For a better understanding of the inventive principles, reference is made below to the embodiment shown in the drawings, which is described in specific language. However, this does not limit the scope of the invention. Changes and further modifications of the illustrated system, as well as further applications of the inventive principles, are possible, as would normally occur to a person skilled in the art in this field.

[0017] It is clear to the person skilled in the art that the preceding general description and the following detailed description are exemplary and explanatory of the invention and are not intended as a limitation of it.

[0018] References in this specification to “an aspect”, “another aspect”, or similar expressions mean that a particular feature, structure, or property described in connection with the embodiment is included in at least one embodiment of the present disclosure. Therefore, occurrences of the expressions “in one embodiment”, “in another embodiment”, and similar expressions in this specification may all refer to the same embodiment, but need not.

[0019] The terms "includes," "include," or other variations thereof are intended to cover non-exclusive inclusion, such that a process or method that includes a list of steps may not only contain those steps but may also include other steps not expressly listed or inherent in such process or method. Likewise, the statement "includes..." in the case of one or more devices, subsystems, elements, structures, or components does not, without further limitations, preclude the existence of other devices, subsystems, elements, structures, components, or additional devices, subsystems, elements, structures, or components.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by a person skilled in the art in the field of the invention. The system, methods, and examples provided here serve only for illustration and are not to be construed as a limitation.

[0021] Embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0022] The functional units described in this specification are referred to as devices. A device may be implemented in programmable hardware such as processors, digital signal processors, central processing units, field-programmable gate arrays, programmable array logic systems, programmable logic devices, cloud processing systems, or the like. Devices may also be implemented in software for execution by various processor types. An identified device may contain executable code and consist, for example, of one or more physical or logical blocks of computer instructions, which may be organized, for example, as an object, procedure, function, or other construct.However, the executable file of an identified device does not need to be physically stored in the same location, but can consist of different commands stored in different locations which, logically linked together, form the device and fulfill its purpose.

[0023] The executable code of a device or module can consist of one or more instructions and may even be distributed across multiple code segments, different applications, and multiple storage devices. Similarly, operational data within the device can be identified and represented, and organized in any form and data structure. This operational data can be captured as a single data record or distributed across different locations, including various storage devices, and may exist, at least partially, as electronic signals within a system or network.

[0024] References in this description to “a selected embodiment”, “an embodiment”, or “an embodiment” mean that a particular feature, structure, or property described in connection with the embodiment is included in at least one embodiment of the disclosed subject matter. Therefore, the expressions “a selected embodiment”, “in an embodiment”, or “in an embodiment” appearing at different points in this description do not necessarily refer to the same embodiment.

[0025] Furthermore, the described features, structures, or properties can be combined in any way in one or more embodiments. The following description contains numerous specific details to enable a comprehensive understanding of the embodiments of the disclosed subject matter. However, those skilled in the art will recognize that the disclosed subject matter can also be implemented without these specific details or with other methods, components, materials, etc. In other cases, known structures, materials, or processes are not presented or described in detail so as not to obscure aspects of the disclosed subject matter.

[0026] According to the exemplary embodiments, the disclosed computer programs or modules can be executed in a variety of ways, for example, as an application in the memory of a device or as a hosted application running on a server and communicating with the device application or browser via various standard protocols such as TCP / IP, HTTP, XML, SOAP, REST, JSON, and other suitable protocols. The disclosed computer programs can be written in exemplary programming languages ​​that are executed from the device's memory or from a hosted server, such as BASIC, COBOL, C, C++, Java, Pascal, or scripting languages ​​such as JavaScript, Python, Ruby, PHP, Perl, or other suitable programming languages.

[0027] Some of the disclosed embodiments involve or otherwise involve data transmission over a network, for example, the transmission of various inputs or files over the network. The network may include, for example, the Internet, wide area networks (WANs), local area networks (LANs), analog or digital wired and wireless telephone networks (e.g., PSTN, Integrated Services Digital Network (ISDN), mobile networks, and Digital Subscriber Line (xDSL)), radio, television, cable, satellite, and / or other transmission or tunneling mechanisms for data transmission. The network may comprise multiple networks or subnetworks, each containing, for example, a wired or wireless data path. The network may include a circuit-switched voice network, a packet-switched data network, or another network for transmitting electronic communications.For example, the network can include networks based on the Internet Protocol (IP) or the Asynchronous Transmission Mode (ATM) and supporting voice communication via VoIP, Voice over ATM, or other comparable protocols. In one implementation, the network includes a cellular network configured for exchanging text or SMS messages.

[0028] Examples of networks include a Personal Area Network (PAN), a Storage Area Network (SAN), a Home Area Network (HAN), a Campus Area Network (CAN), a Local Area Network (LAN), a Wide Area Network (WAN), a Metropolitan Area Network (MAN), a Virtual Private Network (VPN), an Enterprise Private Network (EPN), the Internet, a Global Area Network (GAN), etc.

[0029] Fig. Figure 1 shows a block diagram of a system for classifying apple leaf diseases using deep learning and feature fusion according to an embodiment of the present disclosure.

[0030] Referring to Fig. 1 The system (100) comprises: a data input module (102), which includes a data storage module (102a) and is configured to store a dataset created using various data sources, wherein the dataset of apple leaf images is derived from the datasets “Apple Leaf 9”, “Kashmiri Apple Plant Disease”, and “Plant Village Apple Leaf”; a data preprocessing module (104), which is configured to preprocess the newly created dataset of apple leaves, wherein the data preprocessing module (104) is configured to decode, expand, resize, segment, scale, and color-convert the input image data; a feature extraction module (106), which is operationally connected to the data processing module (104) and is configured to receive preprocessed data and to transfer the preprocessed data to one or more convolutional neural network models for feature extraction;a feature fusion module (108) configured to combine the extracted features from the convolutional neural networks to develop a fused feature vector; a hyperparameter optimization module (110) configured to optimize the hyperparameters of the feature extraction models by implementing a particle swarm optimization algorithm; a classification module (112) configured to classify the fused feature vector into 13 apple leaf disease classes using a random forest classifier; and a user interface (114) associated with the classification module configured to display the classification results.

[0031] In one embodiment, the data preprocessing module (104) is configured to: decode apple leaf image data into JPEG format; perform image enlargement by rotating the images at angles of 30, 60, 90, 120 and 180 degrees; resize all images to 224 × 224; normalize pixel values ​​to a range between 0 and 1 by dividing by 255.0; and perform color conversion for image processing.

[0032] In one embodiment, the feature extraction module (106) is configured to extract features from preprocessed apple leaf images by using a MobileNetV2 model of a convolutional neural network, extract features from preprocessed apple leaf images by using a DenseNet201 model of a convolutional neural network, and reduce extracted features from both models to a 256-dimensional feature set.

[0033] In one embodiment, the feature extraction module (106) further comprises: a convolution layer configured to extract spatial features from apple leaf images; a rectified linear unit (ReLU) activation function configured to introduce nonlinearity; a max pooling layer configured to perform downsampling and control overfitting; a global average pooling layer configured to average features from the last convolution layer; and a fully connected layer configured to flatten features for classification input.

[0034] In one embodiment, the feature fusion module (108) is configured to concatenate the extracted features from fully connected layers of the MobileNetV2 model and the DenseNet201 model into a combined feature vector and to apply dimensionality reduction to optimize the fused feature vector.

[0035] In one embodiment, the hyperparameter optimization module (110) is also configured to: optimize learning rate, decay rate, dropout and number of neurons using the particle swarm optimization algorithm, evaluate hyperparameter configurations based on classification accuracy and select optimal hyperparameter values ​​for the MobileNetV2 and DenseNet201 models.

[0036] In one embodiment, the classification module (112) comprises: a random forest classifier configured with multiple decision trees; wherein the random forest classifier is configured to: receive the fused feature vector as input; classify apple leaf diseases into 13 classes, including healthy leaves and 12 disease states; and output classification results with confidence values.

[0037] In one embodiment, the data input module (102), the data preprocessing module (104), the feature extraction module (106), the feature fusion module (108), the hyperparameter optimization module (110), the classification module (112) and the user interface (114) can be implemented in programmable hardware devices such as processors, digital signal processors, central processing units, field-programmable gate arrays, programmable array logic, programmable logic devices, cloud processing systems or the like.

[0038] The present invention relates to an advanced system for classifying apple leaf diseases, which utilizes deep learning technologies and feature fusion methods to enable accurate and efficient disease identification. The system addresses the urgent need for automated plant disease detection in agriculture, where conventional manual inspection methods are time-consuming, subjective, and prone to error.

[0039] The system architecture comprises several interconnected modules that work in a coordinated manner to process apple leaf images and classify diseases. The data input module contains a comprehensive dataset from three different databases: the Apple Leaf 9 dataset, the Kashmiri Apple Plant Disease dataset, and the Plant Village Apple Leaf dataset. This diverse data collection ensures robust training and testing capabilities under varying environmental conditions and disease manifestations.

[0040] The data preprocessing module performs comprehensive image preparation operations, including format standardization to JPEG, multi-angle image enlargement by rotating the image by specific degrees, dimension normalization to 224 × 224 pixels, pixel value normalization to a range of 0 to 1, and color space conversion for optimal feature extraction. These preprocessing steps ensure consistent input quality and improve the system's ability to generalize across varying image conditions.

[0041] The feature extraction module represents the core intelligence of the system and utilizes two state-of-the-art convolutional neural network architectures: MobileNetV2 and DenseNet201. These models were specifically chosen for their complementary strengths: MobileNetV2 offers efficient feature extraction through depth-separable convolutions, while DenseNet201 enables improved feature reuse through dense connections. The module includes standard CNN components, including convolutional layers for spatial feature extraction, ReLU activation functions for nonlinearity, Max pooling layers for downsampling and overfitting control, global average pooling for feature aggregation, and fully connected layers for the final feature representation.

[0042] The feature fusion module implements an innovative approach to combine the extracted features from both CNN models. By concatenating features from the fully connected layers of MobileNetV2 and DenseNet201, the system creates a comprehensive feature vector that captures various aspects of disease characteristics. Dimension reduction optimizes this fused feature vector to ensure computational efficiency while preserving essential distinguishing information.

[0043] The hyperparameter optimization module uses particle swarm optimization algorithms to automatically optimize critical system parameters such as learning rates, decay rates, dropout probabilities, and neuron counts. This intelligent optimization approach eliminates the need for manual parameter adjustments and ensures the system operates at maximum performance by systematically exploring the hyperparameter space and selecting configurations that maximize classification accuracy.

[0044] The classification module uses a random forest classifier with multiple decision trees to process the fused feature vectors. This ensemble approach provides robust classification capabilities for 13 different classes, including healthy leaves and 12 different disease states. The classifier not only outputs the disease classification but also confidence scores, providing the user with reliability indicators for each diagnosis.

[0045] The system features a user interface that presents classification results in an accessible format, enabling agricultural professionals, farmers, and researchers to quickly interpret disease identification results and make informed decisions regarding plant management and treatment strategies.

[0046] Fig. Figure 2 shows a diagram illustrating the operation of the proposed system according to an embodiment of the present disclosure.

[0047] In Fig. 2. The system processes apple leaf images from various datasets, extracts features using the MobileNetV2 and DenseNet201 models, combines these features into a combined vector, and uses a random forest classifier to identify diseases in 13 classes with improved accuracy through particle swarm optimization.

[0048] The system comprises a data input module with a data storage module configured to store a newly developed apple leaf image dataset. This dataset is created using three different publicly available sources: the Apple Leaf 9 dataset, the Kashmiri Apple Plant Disease dataset, and the Plant Village Apple Leaf dataset, accessible via Kaggle and the PlantVillage database. The dataset used in the system is a multi-class apple leaf disease dataset consisting of images of 12 different apple leaf disease states as well as healthy leaf samples, thus forming a total of 13 different classes. More specifically, four disease classes were obtained from the Plant Village dataset, seven classes from the Kaggle-based Apple Leaf 9 dataset, and the remaining two classes from the Kashmiri Apple Plant Disease dataset, which is also available on Kaggle.The collected data set is then divided into training, testing and validation subgroups, with 80% reserved for training and the remaining 20% ​​for testing and validation purposes.

[0049] The system also includes a data preprocessing module that applies a series of preprocessing operations to the compiled dataset to prepare for feature extraction. The preprocessing workflow involves decoding the image data into JPEG format, followed by image extension to increase dataset variability. The extension techniques used include rotating the images by several angles, including 30, 60, 90, 120, and 180 degrees. Subsequently, all image samples, which were originally in different dimensions such as [256×256], [256×518], [256×462], [512×512], [383×256], and [341×256], are uniformly scaled to a fixed size of [224×224] pixels. Furthermore, the pixel values ​​of the images are normalized to a scale between 0 and 1 by dividing the intensity of each pixel by 255.0.A color conversion step is also included to convert image color schemes into a format optimized for further processing, improve consistency across all samples, and support effective feature extraction.

[0050] The system includes a feature extraction module that receives preprocessed apple leaf images from the data preprocessing module and is configured to extract meaningful features relevant for classification. After preprocessing the apple leaf disease dataset, the processed image data is passed to the convolutional neural network (CNN) models integrated into the feature extraction module. These CNN models are configured to extract relevant features across a series of computational layers. The extracted features are then used by the classification module to categorize the data into 13 classes representing various apple leaf diseases, including healthy leaves.

[0051] In one embodiment, the convolution process plays a crucial role within the feature extraction module. It is performed using a convolution layer that recognizes spatial patterns from the input images. This layer computes output feature maps by applying convolution kernels to localized areas of the image data. Each kernel glides over the input feature maps and computes a weighted sum for each area, followed by the addition of a bias term. This operation generates a new set of feature maps, with each output corresponding to a separate kernel and capturing different spatial features of the apple leaf patterns. After the convolution operation, the system applies a rectified linear unit (ReLU) activation function to introduce nonlinearity into the model. This activation function enables the CNN to learn complex representations by suppressing negative values ​​and preserving positive ones.This improves its ability to distinguish between subtle visual features in diseased and healthy apple leaves. The ReLU function used in the system operates by mapping all negative inputs to zero while preserving positive values. This contributes to better training convergence and model performance. To control overfitting and reduce the spatial dimensions of the feature maps, the feature extraction module includes a max-pooling layer. This layer performs downsampling by selecting the maximum value from a defined range of each feature map. This preserves the most prominent features and reduces the number of parameters. Max-pooling is particularly well-suited to this apple leaf dataset due to the fine variations in disease symptoms on the leaf surface, ensuring the robustness of feature extraction even with such small variations.After the max-pooling phase, a global average pooling layer is applied. This layer averages each feature map from the last convolutional layer, effectively converting the spatial feature maps into a single vector per feature map. This process significantly reduces the number of trainable parameters and minimizes the risk of overfitting. The resulting feature vector is a compact and representative summary of the high-level spatial features extracted from the input apple leaf image. Finally, the output of the global average pooling layer is smoothed and passed to a fully connected (FC) layer, which serves as the final stage of the feature extraction module. The FC layer integrates the flattened feature vector and prepares it for classification by establishing dense connections between all neurons in the previous and current layers.In the classification module, this fully connected representation is then used by the random forest classifier to assign the image to one of the 13 defined classes, thus completing the classification process.

[0052] In one embodiment, the system is developed by evaluating multiple machine learning (ML) and deep learning (DL) models to determine the most effective architecture for classifying apple leaf diseases. The system includes a feature extraction module using deep convolutional neural networks (CNNs), which employs feature fusion techniques to combine outputs from different layers. Feature fusion is achieved by utilizing feature vectors from earlier layers of CNN models such as MobileNetV2 and DenseNet201. These extracted features undergo dimensionality reduction to generate an optimized 256-dimensional feature vector. The fused and reduced features are then passed to the classification module, which uses a random forest classifier to categorize them into 13 apple leaf disease categories.

[0053] In one implementation, the system's framework comprises three phases. In the first phase, features from two separate CNN models, MobileNetV2 and DenseNet201, are extracted and fused. These models initially create high-dimensional feature vectors, which are then reduced to a 256-dimensional representation to improve classification efficiency. The classification module then receives these fused vectors for disease identification. The second phase uses an improvised version of the VGG16 model. Features from this model are combined within the feature fusion module with those from the highest-performing CNN model. After alignment and dimensionality reduction, the optimized feature set is classified using the random forest algorithm. In the third phase, the system integrates features from the two highest-performing CNN models, MobileNetV2 and DenseNet201.These models independently extract features, which are then combined at the fully connected level into a unified feature vector that is subsequently classified using the Random Forest classifier. This final hybrid model is called DensoMobileForestNet and represents the comprehensive classification framework used in the system for the accurate detection of various types of apple leaf diseases.

[0054] During implementation, a comparative analysis of individual machine learning (ML) and deep learning (DL) models is performed using preprocessed data. Machine learning (ML) algorithms such as logistic regression, K-nearest neighbors (KNN), decision trees, support vector machines (SVM), and random forests are implemented, all optimized with the Adam optimizer. The data preprocessing module provides processed images from which the system extracts features using classic techniques such as Hu moments, histograms, Haralick texture features, gray-level co-occurrence matrices (GLCM), and local binary patterns (LBP). These techniques offer diverse visualizations such as shape, texture, and intensity distribution, and help evaluate the performance of the ML model. Performance evaluation metrics are used to identify the most accurate ML model.In the second phase of implementation, the system utilizes several deep learning (DL)-based CNN models, including ResNet50, ResNet101, DenseNet201, XceptionNet, VGG16, VGG19, and MobileNetV2. These models perform automated feature extraction from the apple leaf dataset, and the extracted deep features undergo dimensionality reduction, resulting in a 256-dimensional feature set. This reduced set is then classified in a single step using the Adam optimizer. A performance comparison of these CNN models identifies the most suitable model for the task.

[0055] In one embodiment, the system performs a feature fusion with models VGG16 and VGG19 in a configuration of the feature fusion module. Features are extracted from layers FC6 and FC7 of VGG16 and VGG19, respectively, and combined in the fusion module. The fused features are then classified using the random forest classifier. The performance evaluation determines the superior model based on the accuracy of the classification results.

[0056] In another embodiment, the feature fusion module is configured to implement enhanced feature fusion by integrating the features of the DenseNet201 model with features extracted from an improvised version of VGG16. This fusion involves aligning the feature dimensions and combining the representations into a consolidated feature vector. The fused vector is then processed by the classification module, using the random forest classifier to identify the disease class. This fusion framework improves robustness and accuracy by leveraging the strengths of both models in capturing different features.

[0057] The system proposed in the present invention integrates MobileNetV2 and DenseNet201 into the feature extraction module. Each model independently processes the preprocessed image data to generate feature vectors, which are then combined on the fully connected plane to form a composite feature representation. The classification module subsequently uses the Random Forest classifier to assign these fused features to one of the 13 apple leaf disease categories. This final system, named DensoMobileForestNet, demonstrates superior disease classification performance, making it ideally suited for applications in precision agriculture and automated plant health monitoring.

[0058] The system also includes a hyperparameter optimization module that utilizes the Particle Swarm Optimization (PSO) algorithm to optimize key parameters of the DensoMobileForestNet model. This module is configured to optimize the learning rate, decay rate, dropout rate, and the number of neurons used in the CNN models. During training, the PSO algorithm evaluates various hyperparameter configurations and identifies the combination with the highest classification accuracy. Once the optimal parameters are determined, the DensoMobileForestNet model uses these configurations for feature extraction, and the random forest classifier performs the final classification. This integration of PSO-based optimization increases the overall accuracy and efficiency of the system, thus improving its performance in real-world agricultural disease diagnostics.

[0059] The system's performance is evaluated and a statistical analysis is performed. Performance evaluation of different system implementations is conducted in an experimental setup developed in the Kaggle Notebook environment, using a system with 73.1 GB of disk space, 13 GB of RAM, and 15.4 GB of GPU memory, powered by a P100 GPU. Training was performed with a batch size of 128 over 30 epochs, using the Adam optimizer with an initial learning rate of 0.001. After applying metaheuristic optimization algorithms, optimal hyperparameters were selected and applied to improve the model's performance. To assess classification accuracy, the study used a comprehensive set of performance metrics, including accuracy (AC), precision (PR), recall (TPR), F1 score, Matthews correlation coefficient (MCC), and Cohen's kappa (K).These were evaluated using micro- and macro-averaging methods to measure performance at the individual class level and across all classes. The metrics included true positives, false positives, true negatives, and false negatives to provide meaningful insights into model effectiveness. In the initial experimental simulations of the system, conventional machine learning models were evaluated, with the Random Forest (RF) classifier consistently outperforming other models such as logistic regression, K-nearest neighbors, decision trees, and SVM. In the deep learning category, DenseNet201 emerged as the top performer, followed by MobileNetV2. These models demonstrated superior classification capabilities across the entire integrated dataset. Subsequent phases evaluated various combinations of feature fusion and classification.The fusion of the FC6 and FC7 layers of VGG16 and VGG19, followed by classification using the RF model, showed that VGG16 (FC6 + FC7) achieved high accuracy with RF. The system then moved on to combining features of DenseNet201 and an improvised VGG16 model. Integrating these features led to significantly better results in RF classification, thus confirming the benefits of combining complementary CNN architectures. The final model, named DensoMobileForestNet, integrated features of MobileNetV2 and DenseNet201. This hybrid approach proved highly effective, achieving 94.65% accuracy in classifying the 13 disease classes. The model's performance was further enhanced through hyperparameter optimization using Particle Swarm Optimization (PSO).Compared to other metaheuristic techniques such as Ant Colony Optimization (ACO) and Differential Evolution (DE), the PSO-based DensoMobileForestNet achieved better results in terms of accuracy and convergence. Statistical analysis confirmed the model's reliability and evaluated metrics such as maximum and minimum accuracy, average performance, standard deviation, and average runtime. Low standard deviation values ​​indicated the model's robustness and consistency across multiple trials. A Wilcoxon signed-rank test was performed to compare the PSO-based model with other variants. This test showed that it significantly outperformed the others in most comparisons and only rarely achieved comparable performance. Convergence analysis demonstrated that the PSO-based DensoMobileForestNet consistently minimized the misclassification rate faster and more effectively than other models.Overall, the proposed framework represented a significant improvement over existing methods and surpassed previous approaches, which were typically based on single models and smaller class sets. The integration of preprocessing, feature fusion, and PSO-based optimization contributed to a more accurate and reliable classification system for apple leaf diseases.

[0060] The results showed that the present invention provides a robust, PSO-optimized DensoMobileForestNet system capable of classifying 13 apple leaf disease classes with high accuracy. Although the approach yielded better results, the longer training time due to the size and complexity of the dataset was a notable limitation.

[0061] The drawings and the preceding description show examples of embodiments. Those skilled in the art will recognize that one or more of the described elements can be combined to form a single functional element. Alternatively, certain elements can be divided into several functional elements. Elements of one embodiment can be added to another embodiment. For example, the sequence of the processes described here can be changed and is not limited to the manner described here. Furthermore, the actions of a flowchart need not be implemented in the sequence shown; nor does it necessarily have to be performed by all actions. Actions that are not dependent on other actions can also be performed in parallel with the other actions. The scope of the embodiments is in no way limited by these specific examples.Numerous variations are possible, whether explicitly stated in the specification or not, such as differences in structure, dimensions, and material use. The range of embodiments is at least as broad as specified in the following claims.

[0062] Advantages, further benefits, and problem solutions have been described above with reference to specific embodiments. However, the advantages, benefits, problem solutions, and all components that can lead to an advantage, benefit, or solution occurring or becoming more apparent are not to be construed as critical, necessary, or essential features or components of individual or all claims. REFERENCES 100 A system for classifying apple leaf diseases using deep learning and feature fusion. 102 Data Entry Module 102a Data storage module 104 Data preprocessing module 106 Module for Feature Extraction 108 Feature Fusion Module 110 Module for Hyperparameter Optimization 112 Classification module 114 User interface 202 Input level 204 Preprocessing 206 Mobilenetv2 208 features 210 Densenet201 212 Fully bonded layer 214 Random Forest Classifier 216 Output level

Claims

[1] A system for classifying apple leaf diseases using deep learning and feature fusion, consisting of: A data input module, which includes a data storage module, is configured to store a data set created using various data sources, with the data set containing apple leaf images being derived from the data sets "Apple Leaf 9", "Kashmiri Apple Plant Disease" and "Plant Village Apple Leaf"; a data preprocessing module configured to perform preprocessing of the newly prepared dataset of apple leaves, wherein the data preprocessing module is configured to perform data decoding, expansion, resizing, segmentation, scaling and color conversion of the input image data; a feature extraction module that is operationally connected to the data processing module and is configured to receive preprocessed data and transfer the preprocessed data to one or more convolutional neural network models for feature extraction; a feature fusion module configured to combine the extracted features from the Convolutional Neural Networks to develop a fused feature vector; a hyperparameter optimization module configured to optimize the hyperparameters of the feature extraction models by implementing a particle swarm optimization algorithm; a classification module configured to classify the fused trait vector into 13 apple leaf disease classes using a random forest classifier; and a user interface connected to the classification module, configured to display the classification results. [2] System according to claim 1, wherein the data preprocessing module is configured to decode apple leaf image data into JPEG format, to perform image enlargement by rotating the images at angles of 30, 60, 90, 120 and 180 degrees, to resize all images to the dimension 224 × 224, to normalize pixel values ​​to a range between 0 and 1 by dividing by 255,0 and to perform color conversion for image processing. [3] System according to claim 1, wherein the feature extraction module is configured to extract features from preprocessed apple leaf images by using a MobileNetV2 model of a convolutional neural network; extract features from preprocessed apple leaf images by using a DenseNet201 model of a convolutional neural network; and reduce extracted features from both models to a 256-dimensional feature set. [4] System according to claim 1, wherein the feature extraction module further comprises: a convolution layer configured to extract spatial features from apple leaf images; a rectified linear unit (ReLU) activation function configured to introduce nonlinearity; a max pooling layer configured to perform downsampling and control overfitting; a global average pooling layer configured to average features from the last convolution layer; and a fully connected layer configured to flatten features for classification input. [5] System according to claim 1, wherein the feature fusion module is configured to concatenate the extracted features from fully connected layers of the MobileNetV2 model and the DenseNet201 model into a combined feature vector and to apply dimensionality reduction to optimize the fused feature vector. [6] System according to claim 1, wherein the hyperparameter optimization module is further configured to optimize learning rate, decay rate, dropout and number of neurons using the particle swarm optimization algorithm, to evaluate hyperparameter configurations based on classification accuracy and to select optimal hyperparameter values ​​for the MobileNetV2 and DenseNet201 models. [7] System according to claim 1, wherein the classification module comprises: a random forest classifier configured with multiple decision trees; wherein the random forest classifier is configured to: receive the fused feature vector as input; classify apple leaf diseases into 13 classes, including healthy leaves and 12 disease states; and output classification results with confidence values.

Images