Intelligent diagnosis of diseases in plants using a hybrid Multi-Criteria decision making technique

https://doi.org/10.1016/j.compag.2016.12.003Get rights and content

Highlights

  • An Expert System that can intelligently diagnose diseases in plants is proposed.

  • A model for framing diseases and their pathological characteristics is presented.

  • The disease diagnosis method uses Multi-Criteria Decision Making techniques.

  • The AgriDiagnose system consists of a Pathology tool and a mobile app.

  • Experimental results show an accuracy of over 95%.

Abstract

This paper describes an Expert System that can intelligently diagnose diseases in plants. The system is dialog-based and uses a Multi-Criteria Decision Making technique that is a hybrid of Analytic Hierarchy Process and Sensitive Simple Additive Weighting. The paper describes an approach for disease modeling that uses a set of characteristics which are weighted for each disease using two types of weights: Relative Weights and Scales. The diagnostic process involves calculating the utility value for each disease based on the utility values of its characteristics. Experimental results show an accuracy of over 95%. The system implemented is called AgriDiagnose and it consists of a web-based pathology tool to model the diseases and a mobile app for farmers to interact with the system for disease diagnosis in the field.

Introduction

Diseases have the potential to destroy large numbers of crops and can result in significant losses and food shortages if not detected and controlled in time. For example, the Papaya Ringspot virus affected the country of St. Kitts and destroyed about 90% of that country’s production (Chin et al., 2007). Many developing countries organize plant clinics for farmers at which farmers can be educated about various pests and diseases and where plant Pathologists can diagnose diseases from samples that farmers bring to the clinic. This is often in addition to visits to the farms by Agriculture Extension Officers. Much work has also been done in trying to automate diagnosis (Barbedo, 2016, Gonzalez-Andujar et al., 2006, Mansingh et al., 2007). These Artificial Intelligence systems generally either apply image processing techniques to images of diseased plants or use a data entry dialog system to attempt a diagnosis.

In this paper, we present a dialog based system for diagnosis of plant diseases. The system uses a multi-criteria decision making technique that is a hybrid of Analytic Hierarchy Process (AHP) (Saaty, 1977) and Sensitive- Simple Additive Weighting (S-SAW) (Goodridge, 2016) to dynamically put forward questions to the farmers in an optimal way and to reason through their responses returning a diagnosis. A major contribution of the paper is the approach presented for modeling diseases using a consistent set of characteristics (criteria). The AHP is used for determining weights of these characteristics for all diseases in the system. The diagnosis process uses S-SAW for sensitivity analysis. The S-SAW is an extension of the popular SAW method (Hwang and Yoon, 1981) which allows the decision maker to define an objective function which governs the optimization goals of each criterion. This is used in calculating the utility value of each characteristic.

This technique was implemented in a system called AgriDiagnose, a system that consists of a back-end, web-based pathology tool and a front-end mobile app for farmers. The results obtained from experimentation gives a 95.9% accuracy for diagnosing the correct disease and a 100% sensitivity result that the system returns a positive result when the plant is indeed diseased.

The rest of the paper is organized in this manner. In Section 2, we review some of the approaches taken in the literature to intelligent diagnosis of plant diseases. In Section 3, we describe the disease modeling that is configured in the Pathology tool and in Section 4, we describe the diagnostic process. We trace one case study throughout these two sections so that the reader can follow the process with data. In Section 5, we reveal the results obtained from our simulation exercises and introduce four metrics for measuring these results. We conclude in Section 6.

Section snippets

Background and related work

Farmers would benefit from a diagnostic process that could intelligently act as a human pathologist. This diagnostic process would take the same inputs that farmers typically provide to a pathologist, process them and return real-time diagnoses. Many of the expert systems that have been developed receive input data from images and use image processing methods or through data entry by users through a user interface. Barbedo (2016) provides a comprehensive survey of expert systems applied to

Disease modeling

In the approach presented in this paper, all information about each disease is brought together using the concept of a disease model where a model fundamentally contains all the characteristics of one disease. Each model consists of many characteristics where a characteristic in the simplest terms describes something about a model, for example, spot color. Characteristics may be grouped into categories/types as detailed in Table 2.

Therefore, for each disease there exists a Disease Model di that

Disease diagnosis

Farmers interact with the system by answering a series of questions which describe the appearance of the affected plant. Characteristic Relative Weights and the Sensitive Simple Additive Weighting (S-SAW) are used to determine which questions are asked and how disease diagnosis is determined.

Experimental results

The primary purpose of the experimental simulations is to determine how well the system was able to act as a human Pathologist and correctly diagnose diseased plants. The metrics used to evaluate the system’s diagnostic process are similar to those typically used to evaluate clinical tests (Stojanovic et al., 2014). Four measures were used: sensitivity, specificity, precision and accuracy. We first explain the terms used in these metrics in Table 10 then give the definitions in Equations (9),

Conclusion

The diagnostic process described in this paper combines components of different multi-criteria decision making (MCDM) techniques to form a dialog-based decision support system for plant disease diagnosis. This automated process can result in fast, real-time diagnoses being provided to farmers in the field allowing for early detection. The techniques have been implemented in a computer system called AgriDiagnose that comprises a web-based Pathology tool for configuring the disease models and a

Acknowledgements

This project is funded by the University of the West Indies - Government of Trinidad and Tobago Research and Development Impact fund.

References (18)

There are more references available in the full text version of this article.

Cited by (0)

View full text