Artificial Intelligence Mobile App For Identification Of Cassava Diseases In Farm

The project expects to radically transform disease monitoring by using artificial intelligence (AI), advanced sensor technology and crowd sourcing capable of connecting the global agricultural community to help smallholder farmers. In this work, a Convolutional neural network (CNN) was used. Convolutional neural network (CNN) models have the potential to improve plant disease phenotyping where the standard approach is visual diagnostics requiring specialized training. In scenarios where a CNN is deployed on mobile devices, models are presented with new challenges due to lighting and orientation. It is essential for model assessment to be conducted in real world conditions if such models are to be reliably integrated with computer vision products for plant disease phenotyping. We train a CNN object detection model to identify foliar symptoms of diseases in cassava (Manihot esculenta Crantz). We then deploy the model in a mobile app and test its performance on mobile images and video of 720 diseased leaflets in an agricultural field in Nigeria. Within each disease category we test two levels of severity of symptoms-mild and pronounced, to assess the model performance for early detection of symptoms. In both severities we see a decrease in performance for real world images and video as measured with the F-1 score. The F-1 score dropped by 32% for pronounced symptoms in real world images (the closest data to the training data) due to a decrease in model recall. If the potential of mobile CNN models are to be realized our data suggest it is crucial to consider tuning recall in order to achieve the desired performance in real world settings. In addition, the varied performance related to different input data (image or video) is an important consideration for design in real world applications.

TABLE OF CONTENTS

COVER PAGE

TITLE PAGE

APPROVAL PAGE

DEDICATION

ACKNOWELDGEMENT

ABSTRACT

CHAPTER ONE

1.0     INTRODUCTION

• • BACKGROUND OF THE STUDY

• • PROBLEM STATEMENT

• • AIM AND OBJECTIVE OF THE STUDY

• • SIGNIFICANCE OF THE STUDYT

• • PROJECT ORGANISATION

CHAPTER TWO

LITERATURE REVIEW

• • INTRODUCTION

• • REVIEW OF THE STUDY

• • OVERVIEW OF CASSAVA

• • REVIEW OF DIFFERENT TYPES OF CASSAVA DISEASES

CHAPTER THREE

• • MATERIALS AND METHOD

CHAPTER FOUR

• • RESULT

• • DATA PREPROCESSING

• • CNN MODEL

CHAPTER FIVE

• • DISCUSSION AND CONCLUSION

• • RECOMMENDATION

• • REFERENCES

CHAPTER ONE

1.0                                          INTRODUCTION

1.1                            BACKGROUND OF THE STUDY

Cassava (Manihot esculenta Crantz) is the most widely grown root crop in the world and a major source of calories for roughly two out of every five Africans. In 2014, over 145 million tonnes of cassava were harvested on 17 million hectares of land on the African continent. It is considered a food security crop for smallholder farms, especially in low-income, food-deficit areas as it provides sufficient yields in low soil fertility conditions and where there are irregular rainfall patterns.

Smallholder farmers, representing 85% of the world’s farms, face numerous risks to their agricultural production such as climate change, market shocks, and pest and disease outbreaks (Nagayet, 2005). Cassava, an exotic species introduced to Africa from South America in the 16th century, initially had few pest and disease constraints on the continent. In the 1970s two arthropod pests, the cassava mealybug [Phenacoccus manihoti(Matt.-Ferr.)] and the cassava green mite [Mononychellus tanajoa (Bond.)] were accidentally introduced from the neotropics (Legg, 1999), becoming the most economically threatening pests. Cassava virus diseases, in particular cassava mosaic disease (CMD) and cassava brown streak disease (CBSD), have a longer history on the continent. Mosaic disease was the first to be recorded in Tanzania towards the end of the 19th century (Warburg, 1894). In East Africa, the outbreak of a severe form of the virus in the 1990s, termed East African cassava mosaic virus (EACMV-UG or UgV), coupled with the sensitivity of local cultivars, resulted in a threat to food security in the region as farmers’ only solution was to abandon cultivation (Thresh et al., 1994). Thresh et al. (1997) estimated annual yield losses to CMD at 15–24%, or 21.8–34.8 million tons, at 1994 production levels. CBSD was reported in the 1930s (Storey, 1936). With limited success in controlling CMD and CBSD, the two diseases have become the largest constraints to cassava production and food security in sub-Saharan Africa resulting in losses of over US$1 billion every year (Legg et al., 2006).

In order to manage the detection and spread of cassava diseases, early identification in the field is a crucial first step. Traditional disease identification approaches rely on the support of agricultural extension organizations, but these approaches are limited in countries with low logistical and human infrastructure capacity, and are expensive to scale up. In such areas, internet penetration, smartphone and unmanned aerial vehicle (UAV) technologies offer new tools for in-field plant disease detection based on automated image recognition that can aid in early detection at a large scale.

Conventional plant disease diagnosis by human experts is inherently subjective and limited to regions that can support the required human infrastructure (Bock et al., 2010). Computer vision algorithms show promise to transform this field with the landmark result of a deep convolutional neural network (CNN) winning the Imagenet competition to classify over 1 million images from 1,000 categories, almost halving the error rates of its competition (LeCun et al., 2015). This success brought about a revolution in computer vision with CNN models dominating the approach for a variety of classification and detection tasks. As CNN models become the standard computer vision model to be deployed in real-time vision applications, assessing and reporting whether the results of their performance translates from research datasets to real time scenarios is crucial. Results of different CNN architectures are usually reported on standard large scale computer vision datasets of a million and more static images (He et al., 2016, 2017; Szegedy et al., 2016; Howard et al., 2017). Domain specific datasets like medical imagery or plant diseases, where transfer learning is often applied to CNN models, comprise smaller datasets as expert labeled images are more challenging to acquire (Masood and Ali Al-Jumaily, 2013; Ramcharan et al., 2017). In a recent assessment for a skin lesion classification task, researchers reported the performance of the deep learning model matched at least 21 dermatologists tested across three critical diagnostic tasks (Esteva et al., 2017). This study was done on a labeled dataset of 129,450 clinical images and the researchers concluded that the technology could be deployable on a mobile device but further evaluation in real-world settings is needed. Similar promising results have been shown in studies of plant disease classification (Mohanty et al., 2016; Johannes et al., 2017; Ramcharan et al., 2017) and health care (Miotto et al., 2017). Deploying on mobile devices would also be beneficial in democratizing access to algorithms while maintaining user privacy by running inference offline.

Despite the ubiquity of smartphones there are few examples of CNN models deployed on these phones categorizing visual scenes in the real world where performance is affected by input data type and compounded by wide extremes in lighting as is normal in outdoor settings. Clear examples of computer vision in real world settings such as autonomous vehicles (cars and drones) leverage multiple sensors in both the visible and non-visible spectrum (Floreano and Wood, 2015; Janai et al., 2017). If mobile CNN models are to achieve their promise it is important to recognize the constraint of a single sensor (i.e., camera) and test the performance of a CNN on mobile devices in conditions they are intended to be used in.

Here, we investigate plant disease diagnostics on a mobile device. We deploy and test the performance of a CNN object detection model for real-time plant disease diagnosis in an agricultural field. Within each disease category we test two levels of severity of symptoms – mild and pronounced, to assess the model performance for early detection of symptoms. We report precision, recall, F-1 score, and accuracy for mobile images and video to assess how the CNN performs in a real world app with different types of input data.

1.2                                   PROBLEM STATEMENT

In the last decade, cassava diseases are been identified by an agriculturist only and this takes time to identified and sometimes error can also occur. This mobile based means of identifying cassava disease made it possible for anyone to identify cassava disease in the farm.

1.3                         AIM / OBJECTIVES OF THE STUDY

The study discusses how mobile app can be used to identify cassava diseases in the farm. It aims to increase the effectiveness of farm-level advice. At the end of this work, student involve shall be able to understand:

1. 1. How mobile app can be used to identify cassava diseases.

1. 1. Different cassava diseases

1.4                             SIGNIFICANCE OF THE STUDY

Technology plays a big role in warning the communities of possible outbreaks and also as a monitoring tool for pests and diseases. So the new app will be used against the cassava brown streak disease.

This new app will attempt to solve the problem of pests and diseases that can face farmers with small holder farmers hardly hit. It’s estimated that every year pests and diseases costs billions of dollars to potential agricultural economy as they damage agricultural outputs such as crops, livestock and fish harvests.

With artificial intelligence mobile app for cassava diseases identification, diseases of cassava can be identified easily by anybody.

1.5                                                         PROJECT ORGANISATION

The work is organized as follows: chapter one discuses the introductory part of the work, chapter two presents the literature review of the study, chapter three describes the methods applied, chapter four discusses the results of the work, chapter five summarizes the research outcomes and the recommendations.