Leaf-Classifier
! CODE UPDATED FOR MATLAB v2021b
Introduction
Disease detection in the field of agriculture is an important field of study in India. We find that 17.6% of the GDP is accounted for by Agriculture, Forestry and fishing. As such it becomes an important field to study since classifying diseased leafs using image analytics makes it significantly cheaper to treat.
To do this, a four step approach is applied, beginning with the acquisition of the image to its segmentation, feature extractions as well as classification which is the end result of the project.
Methodology
The approach to solve the problem is systematic, image segmentation is done with K-means clustering method and features are computed from disease affected cluster. The features that are extracted are fed to the ANN (Artificial Neural network) are essentially the Contrast, Correlation, Energy, Homogeneity, Mean, Standard Deviation and Variance. To classify the images, a support vector machine classifier was used. A color transformation was used to convert the leaves into HSV type format. The acquired images are processed into the program and are segmented using Kmeans, segmentation is the process of dividing the images into different parts, in essence clusters and is done using clustering algorithms, For Image processing and its clustering the Kmeans algorithm is used. Kmeans clustering identifies according to the number of clusters, the different clusters, each of these clusters have similar characteristics that is each cluster has parts of the image. This essentially means that one of the cluster would contain the diseased part of the leaf. We can further train the program to run with a command that asks what cluster to use to classify the disease and identify it.
Feature Extraction
Feature extraction is the process of extracting essential statistical values and numerics that are associated with the image. Once the cluster is entered, the image is converted into gray scale and gray level co-occurrence matrices are created. The program uses the said matrices to extract features. The derived features, Contrast, Correlation, Energy, Homogeneity, Mean, Standard Deviation and variance, are given as an input to the classifier.
Classification
Any classifier works on a Machine learning mode. In this case, a Neural Network was applied. The Extracted features are fed into the neural network that learns from the data and uses it to classify further inputs. It provides with the performance plot, Confusion matrix, error and histogram plot after the training of the network. Figure 3 contains a diagram of the neural network that is used, it is a basic push pull neural network.
Conclusion
Leaf disease detection is done using neural network classifier. The segmentation is done using k-means clustering. Various features like Contrast, Correlation, Energy, Homogeneity, Mean, Standard Deviation and Variance are extracted for cotton and tomato diseases. The diseased leaves considered for simulation are bacterial leaf spot, target spot septoria leaf spot and leaf mold disease. Features are computed from disease affected clusters 1 and 3. The features are fed to the classifier for recognising and classifying the diseases. Out of twenty cotton samples nine samples are classified correctly as bacterial leaf spot and one sample is misclassified as target spot. Eight samples are classified as target spot and two samples are misclassified as bacterial leaf spot. Out of twenty tomato samples 10 samples are classified as septoria leaf spot disease and 10 samples are classified as leaf mold disease. Accuracies for four diseases bacterial leaf spot, target spot septoria leaf spot and leaf mold are 90%, 80% and 100% respectively and its average classification accuracy is 100%.