Deep Learning-Based Mobile Application for Plant Disease Diagnosis: A Proof of Concept With a Case Study on Tomato Plant

Deep Learning-Based Mobile Application for Plant Disease Diagnosis: A Proof of Concept With a Case Study on Tomato Plant

Shradha Verma (GGSIP University, India), Anuradha Chug (GGSIP University, India), Amit Prakash Singh (GGSIP University, India), Shubham Sharma (GGSIP University, India) and Puranjay Rajvanshi (VIT University, India)
DOI: 10.4018/978-1-5225-8027-0.ch010
OnDemand PDF Download:
No Current Special Offers


With the increasing computational power, areas such as machine learning, image processing, deep learning, etc. have been extensively applied in agriculture. This chapter investigates the applications of the said areas and various prediction models in plant pathology for accurate classification, identification, and quantification of plant diseases. The authors aim to automate the plant disease identification process. To accomplish this objective, CNN has been utilized for image classification. Research shows that deep learning architectures outperform other machine learning tools significantly. To this effect, the authors have implemented and trained five CNN models, namely Inception ResNet v2, VGG16, VGG19, ResNet50, and Xception, on PlantVillage dataset for tomato leaf images. The authors analyzed 18,160 tomato leaf images spread across 10 class labels. After comparing their performance measures, ResNet50 proved to be the most accurate prediction tool. It was employed to create a mobile application to classify and identify tomato plant diseases successfully.
Chapter Preview


Advanced research in agriculture has led to the concept of smart farming and development of novel techniques, to improve the crop yield and provide assistance to the farmers, but pests and plant diseases are still a dominant threat to the overall plant health and food production.

With the current population growth rate, it is imperative that researchers investigate every potential solution that would facilitate the protection of crop yield. For improving the sustainability, we require to minimize the economic and production losses, along with the degrading impact of fertilizers and pesticides on the environment. Several researchers have consecrated their valuable efforts in developing new and innovative practices for disease diagnosis in plants (Razmjooy et al, 2012; Moallem et al, 2013, 2014). Previously, this domain has remained highly dependent on the conventional and long-established approaches.

Traditional methods encompass visual inspections by experienced individuals, be it the farmer or a botanist. Detecting plant diseases in a lab are one other option. It comprises bringing back samples from the field and conducting a microscopic evaluation and diagnostic experiments, such as ELISA, PCR etc. While this is an accurate procedure, it is time-consuming, not to mention costly, requiring laboratory equipment set-up and highly labor-intensive.

Given the restricted access to resources and limited expertise in plant pathology, there is a vital requirement for processes automation. Farmers are struggling throughout the world to protect their crop from the onslaught of several harmful microorganisms or pathogens such as a virus, bacteria, fungus, nematodes, protozoa (Shankar, Harsha, & Bhandary, 2014) as well as feeding off the insects.

A susceptible host and favorable environmental conditions are all these pathogens require, in order to infect the plants and eventually degrade their growth, sometimes resulting in high mortality. They are ubiquitous and stay inactive in the soil, air and/or water, even in the crop debris for multiple seasons, until they encounter suitable conditions to infest the plant.

In the event of infestation of pests in the crop, it undergoes biological and chemical stress that initially fails to provide any visual cues. It is only after the disease has manifested to a greater severity level, that visual symptoms appear and demand immediate action to control the outbreak.

The visual symptoms may cause some changes in different parts of the plant like shape, size, and color.

Also, to detect any such symptoms, constant monitoring of the crop is required, which can prove to be a tiresome task.

For Agronomists and researchers, plant disease diagnosis has become a greater concern over time and has led them to devise a multitude of techniques for early detection and prediction of diseases that acutely affect the yield and quality of the grain.

Recent research in the sensor based techniques has led to the identification and development of imaging technologies that are extensively being utilized for detection as well as identification of plant diseases, in addition to being non-destructive in nature.

Images capture any anomalies in the occurrence of a plant, which might indicate the attack of a pathogen. Optical properties of a plant are exploited to detect plant stress levels and disease severity. Lately, the emphasis has been towards early disease detection, which is possible due to the recent advances in the remote sensing and imaging technologies (Mahlein, 2016) such as multispectral imaging, hyperspectral imaging, thermal imaging etc. As in humans, early diagnosis in plants can help eliminate and cure the disease, with minimum cost, effort and use of pesticide, hence preserving the ecosystem from its damaging effects.

For more than last four decades, researchers have contributed significantly towards achieving this goal by utilizing the computational power of advanced technologies such as machine learning, deep learning, digital image processing, computer vision, genetic algorithms, big data, internet of things etc. These areas of computing have led to various suitable solutions in agriculture, from crop protection and monitoring to crop quality and its management.

With their combined application, the process of data collection and analysis can be largely automated, leading to the timely prediction of most probable diseases in crops. It would also eliminate the dependency on manual tasks and subject matter experts. Table 1 gives the nomenclature of the general terms used throughout the text.

Key Terms in this Chapter

Support Vector Machine (SVM): It is a supervised machine learning tool utilized for data analysis, regression, and classification.

Convolutional Neural Networks (CNN): It is a subtype of ANNs with a collection of deep FFNNs utilized mainly for image analysis and classification.

Image Processing: Image processing is the area of computer science that deals with the analysis, enhancement and manipulation of digital images for feature extraction, recognition, and classification purposes.

Recall: Similar to precision, it is also a statistical measure. It is the ratio of valid outputs to total number of relevant samples.

Precision: It is a statistical measure of random errors. It is the ratio of valid outputs (also known as true positives) to retrieved samples only.

F1 Score: Given precision and recall values, F1 score is computed as the harmonic mean of both.

Plant Pathology: It is the study of diseases in plants and their causal factors such as environmental conditions, pathogens, etc. that affect the overall plant growth.

Artificial Neural Networks (ANN): It is a computational technique inspired by the human brain. It consists of nodes (neurons) and connections (also known as synapses) between them, to exchange and transfer data. The network learns automatically according to the flow of the data.

Complete Chapter List

Search this Book: