Agricultural Research Management Information System

Home
Research Summary
Search
About Us
- ARMIS
- Brochure
Contact Us
Report
User Request
Data Input
Help
- Operation Manual
  - PDF
  - Video
- Program Area & Commodity
We have reached 37600 number of research entries at this moment.

- Logout

Research Detail

Home
Research
Detail

Mango Leaf Diseases Recognition Using Neural Network and Support Vector Machine

Md. Rasel Mia
Department of Computer Science and Engineering Gono Bishwabidyalay Dhaka, Bangladesh

Sujit Roy
Department of Computer Science and Engineering JKKNIU and Gono Bishwabidyalay Dhaka, Bangladesh

Subrata Kumar Das*
Department of Computer Science and Engineering Jatiya Kabi Kazi Nazrul Islam University Mymensingh, Bangladesh

Md. Atikur Rahman
Department of Computer Science and Engineering Gono Bishwabidyalay Dhaka, Bangladesh

Abstract/ Executive Summary:

This paper presents a Neural Network Ensemble (NNE) for Mango Leaf Diseases Recognition (MLDR). Mango trees are affected by various diseases and identifying diseases is a complex task till now because those diseases are detected manually. This study intends to detect the symptoms of plant diseases easily with machine learning than a manual monitoring system. Here, Trained data are produced by classification technique collecting images of leaves that were various disease affected. A machine learning system is designed to identify the symptom of mangoes leaf diseases automatically uploading and matching new images of an affected leaf with trained data. The proposed system could successfully detect and classify the examined disease with an average accuracy of 80%. This proposed solution would clinch the Mango plants. The system will help to detect disease without the presence of agriculturists. It would also save time to identify the disease with a machine instead of the traditional system, which helps to treat the affection of mango leaf disease properly, increase the production of mango and meet the demand of the global market.

Keywords: Artificial Neural Network (ANN); Disease; K-Nearest Neighbor (KNN); Mango plant; Pattern recognition; Support Vector Machine (SVM)

Location: Department of Computer Science and Engineering Gono Bishwabidyalay Dhaka, Bangladesh

Start Date:

End Date:

Program Area: Pest Management

Commodity: Diseases, Mango

Objectives:

The evaluation and development of leaf diseases has become more common nowadays, as environment and climate condition are more unstable than ever. In this changing environment, appropriate and timely disease identification including early prevention has been more important.

Methodology:

In order to develop MLDR among various NNE techniques, SVM has been employed. The SVM technique along with efficient feature extraction enables MLDR to attain faster training and recognition performance. A. Proposed Model When a new image inserted into the system that new data would be compared with trained data.

Mathematical Analysis Gray-Level Co-occurrence Matrix (GLCM) is a-level spatial dependence matrix. It is frequently used in the literature without a hyphen, co-occurrence. Gray Level Co-occurrence Matrix creates the GLCM =Graycomatrix(I). Here grayscale intensity value I occurs horizontally adjacent to a pixel with the value j. Each element (I, J) in GLCMS specifies the number of times that the pixel with value I occurred horizontally adjacent to a pixel ith value j. We will evaluate some features like Contrast, Correlation, Energy, and Homogeneity by using GLCMS. Contrast: Returns a measure of the intensity contrast between a pixel and its neighbor over the whole image. Range = [0 (size (GLCM, 1)-1) ^2]. Contrast is zero for a constant image.

Input Data (Image) Analysis The inputted data (image) is needed to classify to recognize the disease. For the recognition, we have used SVM classification. Image pre-processing: Firstly, we have taken the image from the current directory. For getting the better performance we can edit the image using any kind of photo editor (just like Adobe Photoshop). Actually, the photo editor's work is to increase or decrease of contrast of brightness or resizing of the image, or rotating the image. Classify the Color in A* B*: We have to convert the image from RGB to the lab. For this, we have to use the srgb2lab function. This function performs the work. Here it creates three images. Those images are being resized according to their previous size. It makes three color images. One is less brightness. And final is according to B*(indicating where the color falls along the blue-yellow).

Feature Extraction: Feature plays a very important role in the area of image processing. Before getting features, various image pre-processing techniques like contrast, correlation, energy etc. homogeneity are applied on the sampled image. For SVM classification firstly we had to load .mat type file, which contained data sets. Then we had to run svmtrain (train, group), train consisted of our training data. Group consisted of normal, dag disease, golmachi, moricha disease, shutimold. Then finally we had to run a method for the classification named svmclassify (svmstruct, sample).

Source of Information: International Journal of Computer Science Engineering (IJCSE) ISSN : 2319-7323 Vol. 8 No.01 Jan-Feb 2019

Article Link (Online Journal):

Funding Source:

Budget:

Total Budget (Tk.) :

Achievement/Findings:

Bangladesh is an agriculture-based country. Its economy is solely dependent on agriculture. Food inflation is a national challenge. Scientists, agriculturists work day and night to promote the yield of food grains. It is very difficult to infer the varieties of a leaf disease by simple visual observation. It is very time-consuming and can be accomplished by the trained botanists. Research outcome can help to recognition of Mango leaf diseases quickly and easily by machine instead of manual system. After observing and classifying all these data, for getting more accuracy we need more data. For matching the disease we need various types of data just like left rotate, right rotate, up to rotate, down rotate, 100%, 120%, 140%, 160%,180%, 200% zoom out. In this primary stage, we have detected four types of disease and one normal among a total of five groups. Each group consists of four different types of images. For getting more accuracy, we need to add more images in each group. The accuracy depends on training data. In our work, we take 20 training data having 13 features each. To improve performance, we need to add more training data. Only for classification, we have used an SVM classifier. We have to use the classifier more sufficiently for increasing efficiency. Future work will include research along with two directions: first comparing textures base features, second color features for improving recognition accuracies.

Knowledge Management: Journal