Description:[001] The present invention relates to the field of agricultural technology, particularly to methods and systems for plant disease detection and classification. More specifically, it focuses on the identification of plant curl diseases using advanced image processing techniques combined with machine learning algorithms. The invention is designed to provide accurate and early detection of diseases in plant leaves, enabling effective disease management and improved crop yield.
[002] The invention integrates the use of active contour models for precise image segmentation, Fourier descriptors for detailed feature extraction, and 1D Convolutional Neural Networks (CNNs) for robust disease classification. This combination of technologies addresses the limitations of traditional methods, which are often labor-intensive, time-consuming, and less accurate. By leveraging digital image processing and artificial intelligence, the invention offers a scalable solution for real-time monitoring and diagnosis of plant diseases in various agricultural settings.
[003] The field of application for this invention extends to horticulture, agronomy, and plant pathology, where early and precise identification of plant diseases is crucial for maintaining crop health and productivity. It can be particularly beneficial for detecting diseases in crops such as tomatoes, peaches, and guavas, which are prone to curl diseases caused by bacterial, fungal, and viral infections. This innovative approach has the potential to revolutionize plant disease management practices, contributing to sustainable agriculture and food security.
BACKGROUND OF THE INVENTION
[004] Plant diseases represent a significant challenge to global agriculture, adversely affecting crop yields and quality. Among these, leaf curl diseases caused by bacterial, fungal, and viral infections are particularly problematic. These diseases disrupt the normal growth and development of plants, leading to severe losses in agricultural productivity. Traditional methods of diagnosing and managing plant diseases often rely on visual inspection and expert analysis, which are labor-intensive, time-consuming, and prone to human error. Consequently, there is an urgent need for automated, accurate, and efficient methods to identify and manage plant diseases.
[005] Advancements in image processing and machine learning technologies have opened new avenues for developing automated systems capable of diagnosing plant diseases from digital images. Image processing techniques enable the extraction of critical features from images, such as shape, color, and texture, which are essential for identifying disease symptoms. Machine learning algorithms, particularly convolutional neural networks (CNNs), have demonstrated significant potential in classifying and recognizing patterns in large datasets, including images of diseased plant leaves.
[006] Despite these advancements, existing methods for plant disease detection often face limitations in accuracy and computational efficiency. Many techniques struggle with accurately segmenting diseased regions from complex backgrounds, leading to suboptimal feature extraction and classification. Moreover, traditional 2D CNNs require substantial computational resources, which can be a barrier to real-time applications in agricultural settings.
[007] This invention addresses these challenges by introducing a novel method for identifying plant curl diseases using an active contour model for image segmentation, Fourier descriptors for feature extraction, and a 1D CNN for classification. The active contour model effectively isolates diseased regions from the background, while Fourier descriptors provide a compact and robust representation of shape features. The 1D CNN, with its lower computational complexity compared to 2D CNNs, enables efficient and accurate classification of diseases. This integrated approach not only enhances the accuracy of disease detection but also reduces computational requirements, making it suitable for real-time applications in agriculture.
[008] The proposed method has been tested on datasets of peach and guava leaf images affected by curl diseases, demonstrating high accuracy in disease detection and classification. By enabling early and accurate identification of plant diseases, this invention offers a valuable tool for farmers and agricultural professionals to manage and mitigate the impact of plant diseases, ultimately contributing to improved crop yields and quality.
SUMMARY OF THE INVENTION
[009] The present invention relates to an innovative method for identifying plant curl diseases using advanced image processing and machine learning techniques. The invention addresses the limitations of traditional methods by providing a more accurate, efficient, and automated solution for early detection of plant diseases, specifically those causing leaf curl.
[010] The method comprises the following steps:
Image Acquisition: High-resolution images of plant leaves are captured using digital imaging devices.
Image Segmentation: The acquired leaf images are processed using an active contour model, also known as "snakes," to accurately segment and isolate the diseased regions from the healthy parts of the leaves. This model works by minimizing an energy function that consists of internal energies controlling the smoothness of the contour and external energies derived from the image's intensity gradients.
Feature Extraction: From the segmented images, features such as shape, color, and texture are extracted using Fourier descriptors. This involves converting the boundary coordinates of the segmented regions into a complex number sequence and applying the Discrete Fourier Transform (DFT) to obtain Fourier coefficients. These coefficients effectively represent the shape features of the diseased regions.
Disease Classification: The extracted features are input into a 1D Convolutional Neural Network (CNN) for classification. The CNN is designed with convolutional layers, pooling layers, and fully connected layers to learn and identify patterns associated with various plant curl diseases. The network is trained on labeled datasets of healthy and diseased leaf images to improve its accuracy and reliability.
Output: The method provides a classification of the disease present in the leaf image, enabling early diagnosis and effective disease management.
[011] The proposed method offers significant advantages, including higher accuracy in disease detection, reduced computational requirements, and suitability for real-time applications. It provides a robust solution for improving crop yield and quality by facilitating early and accurate identification of plant diseases.
BRIEF DESCRIPTION OF THE DRAWINGS
[012] The accompanying drawings illustrate the various steps and components of the method for identifying plant curl diseases using active contour models, Fourier descriptors, and a 1D Convolutional Neural Network (CNN). The figures are as follows:
Figure 1:
(a) & (b) show tomato leaves affected by Late Blight.
(c) displays a color-segmented image of the affected leaves.
(d) presents binary-segmented images used for further analysis.
Figure 2:
Demonstrates the object outline matching the leaf using the active contour model.
Figure 3:
Illustrates a sample 1D CNN configuration with three CNN layers and two MLP (Multi-layer Perceptron) layers.
Figure 4:
Shows three consecutive hidden CNN layers of a 1D CNN, highlighting the sub-sampling factor and filter sizes.
Figure 5:
Input image of a peach leaf affected by curl disease.
Figure 6:
Input image of a guava leaf affected by curl disease.
Figure 7:
Depicts the training process for peach leaf disease classification, including graphs of accuracy and loss over training epochs.
Figure 8:
(a) Input image of a leaf.
(b) Corresponding mask image.
(c) Contour region identified by the active contour model.
(d) Segmented output image.
Figure 9:
Graphical representation of features extracted using the Fourier descriptor.
Figure 10:
Classification output for peach leaf disease using the trained 1D CNN.
Figure 11:
(a) Input image of a guava leaf.
(b) Corresponding mask image.
(c) Contour region identified by the active contour model.
(d) Segmented output image.
Figure 12:
Graphical representation of features extracted using the Fourier descriptor for guava leaves.
Figure 13:
Classification result for guava leaf disease using the trained 1D CNN.
Figure 14:
Depicts the training process for guava leaf disease classification, including graphs of accuracy and loss over training epochs.
Figure 15:
(a) Input image of a guava leaf.
(b) Corresponding mask image.
(c) Contour region identified by the active contour model.
(d) Segmented output image.
Figure 16:
Graphical representation of Fourier descriptor features for guava leaf.
Figure 17:
Classification output for guava leaf disease using the trained 1D CNN.
Figure 18:
(a) Input image of a guava leaf.
(b) Corresponding mask image.
(c) Contour region identified by the active contour model.
(d) Segmented output image.
Figure 19:
Graphical representation of Fourier descriptor features for guava leaf.
Figure 20:
Final classification output for guava leaf disease using the trained 1D CNN.
[013] These figures collectively illustrate the process and results of the proposed method for identifying plant curl diseases.
DETAILED DESCRIPTION OF THE INVENTION
[014] The present invention relates to a method for identifying plant curl diseases using advanced image processing and machine learning techniques. This method aims to provide accurate and early detection of plant diseases by leveraging an active contour model for image segmentation, Fourier descriptors for feature extraction, and a 1D Convolutional Neural Network (CNN) for classification. The primary focus is on diseases affecting the leaves of plants, which are critical indicators of overall plant health.
[015] Image Acquisition
The process begins with acquiring high-resolution images of plant leaves. These images can be captured using standard digital cameras or specialized imaging devices. The quality of the images is crucial, as higher resolution images provide more detailed information, facilitating more accurate segmentation and feature extraction.
[016] Active Contour Model for Image Segmentation
The active contour model, also known as "snakes," is employed to segment the plant leaf images and isolate the regions affected by diseases. This model involves the minimization of an energy function that represents the internal and external energies of the contour. The internal energy ensures the smoothness of the contour, while the external energy pulls the contour towards the edges of the diseased regions. The energy function is defined as:

where is the parametric representation of the contour with sss being the curve length parameter. The internal energy E_int is given by:

where a\alphaa and ß\betaß are weighting parameters controlling the elasticity and rigidity of the contour, respectively. The external energy E_ext is derived from the image intensity gradients and is defined as:

Where G_s is a Gaussian function used for smoothing, ? is the gradient operator, * denotes convolution, (I(x,y)) is the image intensity function.
The active contour model effectively isolates the diseased regions from the background, providing a precise segmentation that is essential for subsequent feature extraction.
[017] Fourier Descriptors for Feature Extraction
Once the diseased regions are segmented, Fourier descriptors are used to extract meaningful features from these regions. Fourier descriptors involve converting the boundary coordinates of the segmented regions into a complex number sequence and applying the Discrete Fourier Transform (DFT) to obtain
the Fourier coefficients. These coefficients serve as compact representations of the shape features, which are crucial for classification.
The resulting Fourier coefficients represent the shape features of the diseased regions. These features are invariant to transformations such as translation, rotation, and scaling, making them robust for classification.
[018] 1D Convolutional Neural Network (CNN) for Classification
The extracted Fourier descriptors are then fed into a 1D Convolutional Neural Network (CNN) for classification. The CNN architecture consists of convolutional layers, pooling layers, and fully connected layers. The network is trained on labeled datasets of healthy and diseased leaf images, allowing it to learn and identify patterns associated with various curl diseases.
[019] CNN Architecture
1. Input Layer: Receives the Fourier descriptors as input.
2. Convolutional Layers: Apply convolution operations to extract local patterns and features from the input descriptors. Each convolutional layer is followed by an activation function, typically ReLU (Rectified Linear Unit).
3. Pooling Layers: Perform down-sampling to reduce the dimensionality of the feature maps and retain the most important features. Max-pooling is commonly used.
4. Fully Connected Layers: Serve as the classifier, where the features extracted by the convolutional and pooling layers are used to predict the class labels of the input images.
The final output layer consists of neurons equal to the number of disease classes, and a softmax activation function is used to produce probability scores for each class.
[020] Training the CNN
The CNN is trained using a dataset of labeled images, where each image is annotated with the corresponding disease label. The training process involves:
1. Forward Propagation: Calculating the output of the CNN by passing the input through the network layers.
2. Loss Calculation: Computing the loss between the predicted labels and the true labels using a suitable loss function, such as cross-entropy loss.
3. Backpropagation: Adjusting the network weights to minimize the loss by computing the gradients of the loss function with respect to the weights.
4. Optimization: Using an optimization algorithm, such as stochastic gradient descent (SGD), to update the weights based on the computed gradients.
[021] Experimental Results
The proposed method was tested on datasets of peach and guava leaf images affected by curl diseases. The following steps were performed:
1. Segmentation: The active contour model was used to segment the leaf images, isolating the diseased regions accurately.
2. Feature Extraction: Fourier descriptors were computed from the segmented regions to extract shape features.
3. Classification: The extracted features were classified using a 1D CNN, which was trained on a labeled dataset of healthy and diseased leaf images.
[022] Results
The method demonstrated high accuracy in classifying various plant curl diseases. The segmented images, feature extraction, and classification results are illustrated in the accompanying figures.
• Figure 5: Input Peach Leaf
• Figure 6: Input Guava Leaf
• Figure 7: Training Process for Peach Leaf
• Figure 8: Active Contour Segmentation Results
• Figure 9: Fourier Descriptor Feature Extraction Results
• Figure 10: Classification Output for Peach Leaf
• Figure 11: Active Contour Segmentation for Guava Leaf
• Figure 12: Fourier Descriptor Feature Extraction for Guava Leaf
• Figure 13: Classification Output for Guava Leaf
[023] The proposed method provides an effective and efficient solution for identifying plant curl diseases. By combining active contour models for segmentation, Fourier descriptors for feature extraction, and a 1D CNN for classification, the method achieves high accuracy and can be implemented in real-time applications. This invention offers significant advantages for early detection and management of plant diseases, contributing to improved crop yield and quality.
[024] Implementation
The method was tested on datasets of peach and guava leaf images affected by curl diseases. The process demonstrated high accuracy in disease detection, with reduced computational requirements, making it suitable for real-time applications in agricultural settings.
[025] The present invention provides an advanced and efficient method for identifying plant curl diseases through the integration of image processing techniques and machine learning. The proposed approach, which employs an active contour model for precise image segmentation, effectively isolates the diseased regions from the plant leaf images. This segmentation is crucial for ensuring the accuracy of subsequent feature extraction and classification steps.
[026] By utilizing Fourier descriptors, the invention captures essential shape, color, and texture features from the segmented leaf images. These descriptors serve as compact and informative representations of the leaf's characteristics, enabling a more robust analysis. The extracted features are then processed by a 1D Convolutional Neural Network (CNN), which has been specifically designed and trained to recognize patterns associated with various curl diseases. The 1D CNN's architecture, which includes convolutional, pooling, and fully connected layers, allows for effective feature learning and classification.
[027] This method has demonstrated significant improvements in disease detection accuracy, as evidenced by experimental results on datasets of peach and guava leaf images. The ability to identify diseases early is particularly beneficial for managing and mitigating the impact of plant diseases on crop yields. The reduced computational requirements of the proposed method make it suitable for real-time applications, providing a practical solution for use in agricultural settings.
[028] The integration of active contour models, Fourier descriptors, and 1D CNNs offers a powerful tool for the early detection and classification of plant curl diseases. This invention not only enhances the accuracy of disease identification but also contributes to the broader goal of improving agricultural productivity and sustainability. Through early diagnosis and effective disease management, farmers can better protect their crops, reduce losses, and ensure the production of high-quality yields. This innovative approach holds great promise for the future of plant disease detection and management in the agricultural industry. , Claims:1. A method for identifying plant curl diseases comprising acquiring images of plant leaves, segmenting the leaf images using an active contour model, extracting features from segmented images using Fourier descriptors, and classifying diseases using a 1D Convolutional Neural Network based on extracted features.
2. The method for identifying plant curl diseases wherein the active contour model is utilized to accurately contour the boundaries of diseased regions by minimizing an energy function.
3. The method for identifying plant curl diseases wherein the energy function in the active contour model includes internal and external energies derived from the image’s intensity gradients.
4. The method for identifying plant curl diseases wherein Fourier descriptors are used to convert the boundary coordinates of the segmented regions into a complex number sequence for feature extraction.
5. The method for identifying plant curl diseases wherein the Discrete Fourier Transform is applied to the complex number sequence to obtain Fourier coefficients representing shape features.
6. The method for identifying plant curl diseases wherein the 1D Convolutional Neural Network is trained on labeled datasets of healthy and diseased leaf images to learn patterns associated with various curl diseases.
7. The method for identifying plant curl diseases wherein the 1D Convolutional Neural Network architecture includes convolutional layers, pooling layers, and fully connected layers for disease classification.
8. The method for identifying plant curl diseases wherein the classification process involves feeding segmented and feature-extracted images into the 1D Convolutional Neural Network and utilizing trained network weights to classify the input images.
9. The method for identifying plant curl diseases wherein the segmentation and feature extraction processes are optimized for real-time application in agricultural settings.
10. The method for identifying plant curl diseases wherein the classification results provide early diagnosis and effective management of plant diseases, enhancing crop yield and quality.