CN7023 Artificial Intelligence & Machine Vision Assignment Sample 2023
Abstract:
The rice leaf disease prediction system is more important in the agriculture field. Deep learning methods are one of the machine learning methods which are used to make the pattern matching processes effectively. The convolutional neural network is the important method in deep learning. This method is more efficient and this method is the main method for pattern matching process. The rice image dataset was collected from the online repository. This data set was the recently updated. The main objective of this automated system is predicting the disease affected leafs in the rice crop using deep learning method. The convolution networks are powerful which can be directly applied on the images. The prediction accuracy is calculated. This automated prediction system produced 90% overall accuracy.
Keywords: Machine Learning, Deep Learning, Convolutional Neural Network, Principal Component Analysis.
Introduction
Rice is the main food item for all over the world. The agriculture field is affected by various diseases to the crops. So the production of food items will be reduced. The food material production is reduced by this king of disease. Identification and prediction of disease affection is more important. The early prediction can save the maximum loss for the farmers and the production can be increased by taking the necessary action to recover the crops from the disease. This disease can affect the food material production from 60% to 100%. This will give great loss to the farmers and also it will create the demand for the food materials. Various research works are done by the researchers to provide the good system for this disease prediction.
Deep Learning algorithms are powerful and efficient to handle the digital images. The Convolutional Neural network is used for this prediction system. This system helps to provide the information about the disease prediction and control. This will also provide the information related to the disease control and the economic loss. From this system, the diseases of rice leaf will be predicted quickly. The time take for this disease identification and prediction is very less when compared with the manual prediction process (Wan-jie Liang. 2019). The MATLAB tool provides implementation support for the automated rice crop disease prediction system. The dataset for this implementation is taken form the online sites. The dataset will be divided three such as Training dataset, Validation dataset and testing dataset. The taken dataset consist 5069 images with four different class labels.
The convolutional neural network considers the neuron based process which receives number of inputs. Then the activation function is processed to handle the input valued and produced the desires output class label for the image. The image processing methods and feature extraction processes are essential to identify the disease in the rice leaf. The main advantage of using this convolutional neural network is the automatic detection features without requiring any supervision process from human. The computational process of this CNN is more effective than the machine learning methods in this image processing sector.
Literature Survey
The rice leaf disease identification system is developed with deep convolutional neural network (Yang Lu. 2017). This system considered 500 natural images which are with diseases and healthy crops. The convolutional neural network was trained to identify the 10 different diseases in the rice leaf. This system used 10 fold cross validation method for finding the accuracy of the system. The image pre-processing techniques and the feature extraction techniques were used to make prediction of the disease. To get the better prediction system, the experimental process involved Particle Swarm optimization method and support vector machine algorithm for classification. The standard back propagation algorithm was also used in this prediction system. the training accuracy and the testing accuracies were analyzed. This automated system produced 95.48 % accuracy in the field of disease prediction in rice leafs.
Deep learning methods were adapted to predict the disease in the Rice crops (V. Vanitha. 2019). This rice disease prediction system used convolutional neural network for the identification and classification. Based on the given training, the CNN was effectively predicting the disease in the leaf. The 500 data samples were collected for disease identification. There were three different rice disease analyzed and the classification was based on those disease prediction. The rice diseases are Leaf brown spot, Sheath Rot and Bacterial Blight. This rice disease prediction system provided the framework for making the disease prediction process. The data preparation methods were used to arrange the data samples for implementation. The training data base was selected to give training for the CNN to make disease prediction. The convolutional layer, pooling layer, the fully connected layer and The ReLu activation function were used to make the effective disease identification. Feature extraction is the main process which focuses the important features to predict the disease.350 9mage samples were used for the CNN training. The VGG16, ResNet50 and InceptionV3 architectures were used in this convolutional neural network. From these CNN architecture, the ResNet50 reached the highest accuracy than the other architectures used. This method provided 99.53% of accuracy in the classification process.
The smart paddy crop disease identification system was developed with the help of Convolutional neural network (R.Rajmohan. 2018). The sensor based approach was implemented in this disease identification. The main process of this system was collecting the data samples for creating the database. The disease identification and prediction was done with the help of convolutional neural network and SVM classifier. The database was built with the disease syndromes and the possibility of treatment details. This system also provides the disease management process for providing the treatment details for the affected diseases. Image capture and selection methods were used to make the database. The Image Zoom and Crop methods were used to make the region process for identifying the disease affected area of the leaf. The 250 image samples were used in the training process of convolutional neural network. For testing 200 image samples were used. 50 images were identified as the noisy images. The image processing methods were implemented to reduce the noisy images from the dataset. The CNN and SVM methods were used to predict the disease and provided the better performance. The accuracy of the system was calculated using confusion matrix formula. The proposed methodology provided the higher accuracy with 87.50 %.
Paddy leaf disease identification and recognition system was developed with Optimized deep neural network (S. Ramesh. 2019). This recognition system used jaya optimization algorithm for the classification process. the image pre-processing methods were used to convert the color images into the HSV images. The image segmentation process was also used to split the images as disease affected parts and the normal parts. The feature extraction process was used to fins the disease affected parts from the normal parts. The color, shape and the textures are the main features in this disease identification and recognition. The optimized deep learning methods were used to find the disease affected parts in the images. The jaya optimization algorithm was also used to make the disease identification and recognition in an effective way. The received results were compared with ANN model to check the accuracy in prediction. The proposed method achieved 98.9% prediction accuracy.
Analysis of Automatic rice Disease Classification was implemented using image processing techniques (G. Jayanthi 2019). The digital image processing methods were implemented to predict the rice disease. The image segmentation process was implemented in the RGB color images. The GLCM and SURF features were used for the extraction of the image features. The Edge detection methods were used for segmentation. The FCM method was also used to support the image segmentation process. The benchmark datasets were collected from the repositories to implement the disease detection system. The Image pre-processing techniques were used to remove the noisy images for the training and testing process. The infected parts of the crop were identified with adaptation of image segmentation process. Five different Rice diseases were identified using this detection system. The machine learning classifiers were used to make the classification of images in this rice disease detection system. The accuracy, sensitivity, and specificity were calculated for this prediction process.
Localization and classification of rice disease prediction system was developed using the Deep Convolutional Neural Network (Liu 2016). This system also used the Saliency Map for the identification of differentiation in the images. Bounding square method was used to extract the features from the large dataset. The self-learning process for features selection was implemented. The important parameters are analyzed in this prediction system. The deep convolutional neural network was used to identify and predict the rice disease. The dataset contained digital images which are collected form the online sites. The pre-processing methods were implemented on the collected images for removing and resizing the images. The training database was used to identify the disease with the convolutional neural network. The testing dataset evaluates the prediction accuracy of this disease identification system. This proposed model reached 95% of accuracy in the rice disease prediction.
The rice plant disease classification system was developed using deep conventional neural network and transfer learning method (Vimal K. Shrivastava. 2019). The rice crop disease prediction system was more important in the agriculture field. The real time dataset was used in this prediction system. This dataset also considered four different diseases. The pre-processing methods were used to reduce the noisy images and resize the images for prediction process. Three kind of dataset division was done to analyze the accuracy of the prediction system. The classification accuracy was high when the training phase considered 80% of data samples and the testing phase considered 20% of data sample. This disease prediction system provided 91.37 % of prediction accuracy.
The Identification and recognition system for rice disease and pests using deep learning method was developed (Chowdhury Rafeed Rahman. 2020). This system was used to make the early prediction of the rice diseases to avoid the economic loss for the producer. The deep learning methods were preferred because these learning methods were efficient to handle the images for pattern matching. The convolutional neural network was used to identify and predict the disease affected parts using the digital images. The VGG16 and inceptionV3 models were adapted to make this prediction process. The real time dataset was used for the disease prediction. The feature selection methods were utilized to extract the important features of the images which helped to find the diseases in the rice leafs. The prediction accuracy was calculated and compared with these two architectures. The comparison results showed that the VGG16 model achieved the higher accuracy than the other model.
Materials and methods
The proposed system will give the enhance result of rice disease prediction. The Rice images are collected form Kaggle site with more than 5000 image samples. This dataset contains three different disease classes. This dataset contains Brown spot, Hispa, Leaf Blast and healthy rice leaf images. All these data samples are divided into 70:10:20. This indicates that the 70 percentage of dataset is used for the training of conventional neural network. The 10 percentage of dataset is used for the validation function and 20 percent dataset is used for the testing function. The Image pre-processing methods and feature extraction process will be done to get the enhanced dataset with dimensionality reduction and the noisy image reduction. Convolutional neural network is the deep learning techniques which are used for the image [pattern identification. There are different types of architecture applications supported by this convolutional neural network. Those are LeNet-5, AlexNet, VGG16, Inception-v1, Inception-v3, ResNet-50, Xception, and ResNeXt-50. These architectures are used to provide the better disease prediction results with less computation time.
| Architecture | Description |
| LeNet-5 | This architecture consist 5 layers which consist 2 convolutional layer and 3 fully-connected layers. The average pooling layer is also used to calculate the average weigh of the network. |
| AlexNet
|
AlexNet consist 8 layers which have 5 conventional layers and 3 fully connected layers. This architecture supports to ReLu activation function. |
| VGG-16 | This layer contains 13 convolutional neural network and 3 fully connected layers. This architecture consists of 138M parameters and takes up 500MB of storage space. |
| Inception-V1 | This architecture provides 22 layers with 5M parameters. This supports the parallel towers for convolutional network with various filters. |
| Inception V3
|
This supports the 22 layer process with 24M parameters. This avoids the representation of bottlenecks and using the factorization method for effective computation. |
| ResNet-50 | This provides the basic building blocks such as conv and identity. This can make more deeper network up to 152 layers |
| Xception | This supports the depthwise separable convolutional layers for the detection process. |
| ResNeXt-50 | This provides 50M parameters and scaling up the number of parallel towers. |
These network architectures are more efficient to make the pattern matching process with the images.
The VGG16 architecture is selected for this proposed disease prediction system. The convolutional layer, pooling layer, ReLu activation function and fully connected neural network are used to predict the disease in the rice leafs. The evaluation methods are used to check the performance of this architecture.
Image Pre-Processing
Image pre-processing is the initial process for this rice disease prediction system. These pre-processing methods are used to make the resize process for all the images in this database. All the data samples are resized with same size factor to make the computation process in an efficient manner. The filtering methods are also used for the removal of unwanted parts such as background removal. Making the background as black will provide more memory efficient process. The image processing methods provides filtering which consist smoothing, sharpening and enhancement of edges. The filtering mask is the small matrix process which is used to make sharpening, blurring and embossing the images. This process is done with the convolutional process between the kernel and an image.
Feature Extraction
Feature extraction is used to find the key factors of the image which is more useful to find out the disease affected area. Most familiar features are color, shape and textures. This is used to make the dimensionality reduction to provide the storage efficiency for the computation process. The compact feature vector is used to scale the images and make all the images into same size with dimensionality reduction. The feature extraction is done with various methods such as projection and profiling, zoning, standard deviation, density and mean etc.
Deep learning methods:
Deep learning is one of the machine learning methods which is giving more effective results in image pattern matching. This learning method is based on the neural network process which consist number of neurons for making the process. This method also consist the input layers, hidden layers and the output layer process for the certain activity.
Convolutional neural network
The convolutional neural network is used to provide various layer processes to achieve the better results. This network consist convolutional layer, pooling layer, fully connected layer and the activation layer with activation function with Principal Component Analysis (PCA). This method is used for reducing the dimensionality of the dataset.
Convolutional layer
The convolutional layer is used to make the filtering process and map the input values using feature map. This method is one of the basic building blocks of the convolutional neural network which helps to automatically learn the huge number of filters to a training dataset. This summarized the detected features in the input image. This layer consist multiple filters for learning the features through to make the feature map. Color images consider multiple channels to recognize color features.
Pooling layer
The Pooling layer is the next important layer which is followed by the convolutional layer. This layer is used to reduce the dimensionality and parameters for the computation process. These less parameters can improve the performance efficiency of the network. This layer handles the feature map independently. This consist two main pooling such as max pooling and average pooling. The average pooling is used to calculate the average value of the entire feature map. The max pooling is used to choose the maximum elements form the region which is covered by the feature map.
ReLu activation function
The rectified linear activation is used to predict the output if the input value is positive, otherwise it will return zero. This activation function is processes like a linear function to avoid the non-linear difficulties. The main advantage of this ReLu activation method is the process simplicity. This also gives the sparsity representational to calculate the negative inputs.
Fully Connected layer
This layer is known as the feed forward layer. The input values to this fully connected layer are the output of convolutional or pooling layers. This layer is the classic architecture which supports all the input neurons to be connected with the next layer for the computation process. This fully connected layer consist huge number of connections and network parameters. The fully connected input layer consist the output values of the pooling or conventional layer as input value. The fully connected output layer produces the class label with probability.
Experimental Results:
The dataset is taken form the Kaggle site which contains more than 5000 images of rice leaf with 4 class labels. MATLAB is the familiar tool which provides the implementation platform for this rice disease prediction system using convolutional neural network.
Training model:
The Convolutional neural network must be involved to get training to identify the disease affected regions of the images. The 70 percentage of data samples are used for the training of neural network for effective prediction.
| Category | Images Count | Training 70% | |
| BrownSpot | 1249 | 874 | |
| Healthy | 1132 | 792 | |
| Hispa | 1438 | 1007 | |
| LeafBlast | 1250 | 875 |
Table No:1 Training details using CNN
This table indicates the data samples counts which are involved in the training process.
Validation Model:
The validation model is used to check the performance of the training process. this evaluation helps to identify the efficiency of the used method in the rice disease prediction.
| Category | Images Count | Validation 10% | |
| BrownSpot | 1249 | 125 | |
| Healthy | 1132 | 113 | |
| Hispa | 1438 | 144 | |
| LeafBlast | 1250 | 125 |
Table No: 2 Data samples for validation model
Testing model:
Testing is the evaluation process of the trained model for the rice disease prediction. The convolutional neural network is used for the disease prediction. Form the dataset, 20 percentages of the data samples are used for the testing function.
| Category | Images Count | Testing 20% | |
| BrownSpot | 1249 | 250 | |
| Healthy | 1132 | 226 | |
| Hispa | 1438 | 288 | |
| LeafBlast | 1250 | 250 |
Table No: 3 Data samples for Testing model
Confusion Matrix:
This confusion matric is the essential function which is used to describe the efficiency of the used model. The performance of that model is calculated using the confusion matrix table. This table consists four main components such as True positive, True negative, False Positive and False negative.
This confusion matrix is sued to calculate the accuracy, misclassification rate, precision and recall.
Formula to calculate the accuracy is
Formula to calculate the Recall is
Formula for calculating the Precision is
Using these formulas the efficiency of the proposed system is calculated.
Conclusion:
The convolutional neural network is used for the rice disease prediction system. The VGG16 architecture is used to make the convolutional neural network. The data samples are taken from Kaggle. This implementation is used PCA analysis for dimensionality reduction. The prediction accuracy for 4 different classes is evaluated using confusion matrix. This Automated Rice Leaf Disease Prediction System provides 93.46% accuracy in brown spot prediction, 92.96% accuracy in healthy leaf prediction, 93.64% accuracy in Hispa Prediction and 93.12% in LeafBlast prediction.
References:
Bibliography
Chowdhury Rafeed Rahman., Preetom Saha Arko., Mohammed Eunus Ali., Mohammad Ashik Iqbal Khan., Sajid Hasan Apon., Farzana Nowrin., Abu Wasif.,. “Identification and Recognition of Rice Diseases and Pests Using Convolutional Neural Networks.” arXiv.org, 2020.
- Jayanthi, K.S. Archana, A. Saritha., . “Analysis of Automatic Rice Disease Classification Using Image Processing Techniques.” International Journal of Engineering and Advanced Technology, 2019: Volume-8, Issue-3S.
Liu, Z. Y., Gao, J. F., Yang, G. G., Zhang, H. & He, Y. “Localization and Classification of Paddy Field Pests using a Saliency Map and Deep Convolutional Neural Network.” Scientific Reports, 2016.
R.Rajmohan., M.Pajany., R.Rajesh.,D.Raghu Raman., U. Prabu.,. “SMART PADDY CROP DISEASE IDENTIFICATION AND MANAGEMENT USING DEEP CONVOLUTION NEURAL NETWORK AND SVM CLASSIFIER .” International Journal of Pure and Applied Mathematics, 2018: Volume 118, No. 15 , PP 255-264.
- Ramesh., D. Vydeki.,. “Recognition and classification of paddy leaf diseases using Optimized Deep Neural network with Jaya algorithm.” Information Processing in Agriculture, 2019.
- Vanitha., . “Rice Disease Detection Using Deep Learning.” International Journal of Recent Technology and Engineering, 2019: Volume-7, Issue-5S3.
Vimal K. Shrivastava., Monoj K. Pradhan., Sonajharia Minz., Mahesh P. Thakur.,. “RICE PLANT DISEASE CLASSIFICATION USING TRANSFER LEARNING OF DEEP CONVOLUTION NEURAL NETWORK.” Remote Sensing and Spatial Information Sciences, 2019.
Wan-jie Liang., Hong Zhang.,Gu-feng Zhang., Hong-xin Cao.,. “Rice Blast Disease Recognition Using a Deep Convolutional Neural Network.” Scientific Reports, 2019.
Yang Lu., Shujuan Yi.,Nianyin Zeng., Yurong Liu., Yong Zhang.,. “Identification of Rice Diseases using Deep Convolutional Neural Networks.” Neurocomputing, 2017: Vol- 267, PP 378-384.
Know more about UniqueSubmission’s other writing services:
