A Development Of Neuro Expert System For Diagnosingpaddy Disease

A Development of Neuro Expert System for Diagnosing Paddy Disease Azlin Ahmad1, Shamimi A.Halim and 1 Norzaidah Md Noh1 1

Faculty of Information Technology and Quantitative Science, Shah Alam {azlin, shamimi,norzaidah}@tmsk.uitm.edu.my

ABSTRACT Paddy production has led to the development of economic growth to Malaysia, especially to the northern states of Malaysia. Based on the statistical data from Muda Agricultural Development Authority (MADA), the average of gross harvested production for the last 5 years is 4,687,636 metric tones. Due to the whether, the vectors and insects can easily affect the paddy plant. Paddy diseases are major biological constraints in rice production where more than 60 diseases are well described. The Neuro Expert System for Diagnosing Paddy Disease is focused on the 2 types of diseases caused by fungus, which are blast disease and sheath blast disease, criteria and common symptoms of the diseases in MADA territory, and used the Case Based Reasoning (CBR) and Neural Network (NN) techniques. The research uses the images of paddy leaves. Samples of infected leaves were captured and pre-processed using the image processing toolbox in MATLABv7.0 and Adobe Photoshop. The neural network part of this prototype uses the Lavenberg–Marquardt Backpropagation Neural Network to identify type of diseases infected. While nearest matching algorithm has been chosen as the engine of the Case Based Expert System. Keywords: paddy disease, case based reasoning, backpropagation, image processing

1. INTRODUCTION Paddy plantation is a major agricultural activity in Malaysia. According to MADA, the average gross of harvested production was 4,687,636 metric tones for the last five years. However, the processed product has loss to 3,745,239 metric tones. This situation is due to several factors such as season. Basically in Malaysia, there are two production seasons, which are; i) main season which is hot and humid, and ii) off season which is hot and dry. These two different seasons have caused the vectors and insects affect the paddy plant easily. As a result, the paddy plant will easily infected by several diseases. Traditionally, the farmers use manual approach in identifying the diseases assisted by MADA officer. This paper proposed a new approach that combines two different techniques which are NN and CBR. The paper is organized as follows. Section 2 and 3 would discuss paddy diseases in MADA area and related works in this field. In section 4, an appropriate approach and methodology for this research will be explained. Section 5 would discuss about the finding and results of the research while in Section 6, there would be some conclusion about the result. 2. PADDY DISEASE IN MADA AREA Paddy diseases caused by fungus are numerous (Ou, 1972). Sheath Blight (Hawar Seludang), Bacterial Leaf Blight (Hawar Daun), Bacterial Leaf Streak (Jalur Daun), Bakanea and Brown Spot (Bintik Perang) and Blast (Karah) are commonly diseases that can be found in the MADA area (Abdullah, 1991). Blast and sheath blight are the more serious fungus paddy diseases. Blast is an ancient disease of paddy and the symptoms on the paddy plant include spots or lesion on leaves, nodes and branches. Bacterial leaf blight and leaf streak are diseases that most often attack at the leaf tip and rods, respectively. While bakanae

Paper number: 3027987

disease is cause by fungus Gibberella fujikuroi (Anderson et al, 1999), the brown spot disease is caused by an air-borne fungus. 3. RELATED WORK The integration of Artificial Intelligence (AI) technique with agricultural field can be seen as a means of diversifying the usage of latest technology in respective area. Based on research carried out by Abdullah et al (2007) from Faculty of Information Science and Technology, University Kebangsaan Malaysia, a Fuzzy Logic approach has been used to handle the uncertainty and vagueness in order to identify the paddy disease. In this research, ten linguistic variables have been identified such as color and type of lesion, color of boundary and leaves, percentage of damages lesion and paddy age level. 4. METHODOLOGY This research involves several phases: preliminary studies, knowledge acquisition, knowledge representation, image processing, feature extraction, algorithm adaptation, prototype design and development, testing and fine tuning. 4.1 Knowledge Acquisition and Representation This phase started with data collection activity from several resources, including Lembaga Kemajuan Pertanian Muda (MADA) Ampang Jajar, Kedah, the domain expert from MADA, and huge number of literatures. The data collected from MADA includes the detail of paddy development in Malaysia, data on area of paddy plants grown, harvest and products, types of diseases that affected in Muda territoryand some images of infected paddy plants. Then, the data and knowledge were represented into the form of cases. Cases construction was done because it can represent the knowledge and manage to express the relation, recommendation, directive, strategies and heuristic of the cases. The basic structure of case-based reasoning (CBR) is used to apply the expert system. It contains four cycles: retrieve, reuse, revise and retain. Some examples of the cases constructed are shown in the figure 1 below:

Case 1: Hawar Seludang

Case 3: Karah

-

-

-

bintik-bintik pada seludang / upih daun bintik-bintik membesar dan memanjang tompok-tompok berwarna kekelabuan dan perang di sekelilingnya.

-

-

Case 2: Bakanae -

-

anak pokok padi yang lebih tinggi dan daunnya berwarna pucat terdapat akar pada ruas anak padi akan mati satu persatu

bintik-bintik berbentuk lonjong dan tajam di hujungnya tengah daun berbintik warna kelabu/ keputihan dan di tepi berwarna kemerahan pokok akan musnah jika serangan peringkat semaian

Case 4: Bintik Perang -

bintik-bintik berbentuk bijian berwarna perang di tepi dan kelabu/ keputihan di tengah, pada daun dan buah padi

Figure 1: Example of cases of paddy diseases Paper number: 3027987

4.2 Image Processing and Feature Extraction This phase started when the images of the infected paddy plants were collected from several sources. Then, the images were preprocessed to make sure the images are ready for feature extraction. This diagnosing system is applying the pattern recognition concept. Therefore, it is important to verify and identify the exact variables which then will be the input for the system. The RGB value of the image was identified as the most suitable feature. This is because the major differences between the diseases are the color of the infected paddy leaves. 4.3 Backpropagation Neural Network (BPNN) In the algorithm adaptation phase, BPNN was used in training the network. The training input patterns were presented to the network input layer. Then the network propagates the input pattern until the output pattern is generated by the output layer. Sigmoid function is used as the activation function of the neural network. 4.4 Prototype Design & Development The process flow of the prototype is represented in the figure 2 below.

Home/ Mainpage ADMIN

Create Retrieve

USER

Answer Questions Paddy Diagnose

Update Database Delete

Result -Type of Paddy Disease & how to control it

Information of Paddy Type of Paddy Disease Type of Insect’s Pest

Figure 2: Process flow Figure 3 below presents the research model of the prototype. The doted box represents the Expert System Engine.

Paper number: 3027987

BPNN Engine Uploaded Image

Disease

Symptoms Knowledge Base

Disease & Recommendation

Treatment

Symptoms Symptoms

Questions

User

Symptoms

User Interface

Inference Engine

Answer

Working Memory Solution / recommend treatment Expert System Engine

Figure 3: Prototype Research Model The main developments of this prototype are the image recognition and the CBR engine. MATLAB is used to develop the image processing part. In the CBR inference engine, K-nearest neighbor technique is used to identify the most similar cases to the symptoms provided by the user. The certainty factor (CF) is then calculated. The equation used to combine the certainty factors for a hypothesis is as following: Total CF value = ∑ CFcombine (CF1, CF2) * 100 where CF1 is certainty factor value for one type of disease and CF2 is another certainty factor value for the same disease. The total of this combination is as follows: CFcombine (CF1, CF2) = CF1 + CF2 ( 1 – CF1) 5. RESULTS AND DISCUSSIONS After several analyses done onto the BPNN engine, the best architecture of neural network is using 11 hidden neurons as shown in table 1. The level of accuracies to identify Karah disease using the image is 84%. Table 1: Level of accuracies according to number of hidden neurons. No of hidden neurons 7 11 13

No of Epoch 400 400 400

Karah (%) 83 84 83

The hybrid system of BPNN and CBR has produced a good result in identifying the paddy disease based on several symptoms, and the image of the infected paddy leaves.

Paper number: 3027987

6. CONCLUSIONS This research has shown that, by implementing a neuro expert system will definitely be helpful to diagnose paddy disease. The hybridization of the two approaches in diagnosing paddy disease is quite a success. Having the pattern recognition part, allows the user to identify a disease based on the image of the infected paddy leaves. Case-based expert system allows the user to understand clearly each diseases, and identify the disease before it might rise. However, this prototype has its own strength and weaknesses. Having a small database to store all the possible cases might not covers every cases exist in the whole country. Besides that, the cases need to be maintained and examined frequently by the domain experts in order to produce the best result. In order to improve the prototype, there are several activities might be done to upgrade the system. This includes to upgrade the system to a web-based application, use centralized database system to make The development of the prototype in sure the diseases might cover a larger area of paddy plantation.฀ assisting the aforementioned application domain is expected to improve the understanding of this framework structure. The system will help and assist the farmers by giving them the knowledge required about all paddy disease that might occur in their area. At the same time, this system will help them to control any disease before it might destroy the whole plantation. ACKNOWLEDGEMENT This research is a self funding research. We are grateful and proud to produce this research with the lehp of several parties, especially the domain expert, Mr Muhamad Hisham b Mohd Noor, an agricultural officer of MADA. And our special thanks to Lembaga Kemajuan Pertanian Muda (MADA) for all support that they give to make sure this system a success. REFERENCES Abdullah, R. (1991). Malaysia Bumi Bertuah. Penerbit Prisma Sdn Bhd. Abdullah, S., Bakar, A. A., Mustafa, N., Yusuf, M. and Hamdan, A. R. (2007). Fuzzy Knowledge Modelling for Image-based Paddy Disease Diagnosis Expert System, Proceedings of International Conference on Electrical Engineering and Informatics, Bandung. Anderson, L., Webster, R.K. (1999). Bakanae – Disease of Rice in California. Department of Plant Pathology, University of California. Fausett, L. (1994). Fundamentals of Neural Network. Library of Congress in Publication Data. Negnevitsky, M. (2002). Artificial Intelligence: A Guide to Intelligent System. Pearson Education. Ou, S.H. (1972). Rice Diseases.1st ed. International Research Rice Institute, Los Banos, Philippines. Ou, S.H. (1985). Rice Diseases.2nd ed. International Research Rice Institute, Los Banos, Philippines. Verlag, S. (1980). Proceedings of Symposium on Paddy Soil. Beijing Science Press.

Paper number: 3027987