Intelligent Image Analysis for Plant Phenotyping

Ashok Samal, Sruti Das Choudhury, Ashok Samal, Sruti Das Choudhury

• Editorial: CRC Press
• Año: 2020
• ISBN:
  9781351709989
• Edición: 1
• Categoría: Informatica
• Categoría: Grafica per computer

Domesticated crops are the result of artificial selection for particular phenotypes or, in some cases, natural selection for an adaptive trait. Plant traits can be identified through image-based plant phenotyping, a process that was, until recently, strenous and time-consuming. Intelligent Image Analysis for Plant Phenotyping reviews information on time-saving techniques, using computer vision and imaging technologies. These methodologies provide an automated, non-invasive, and scalable mechanism by which to define and collect plant phenotypes. Beautifully illustrated, with numerous color images, the book focuses on phenotypes measured from individual plants under controlled experimental conditions, which are widely available in high-throughput systems.

Features:

• 
  
  

• Presents methodologies for image processing, including data-driven and machine learning techniques for plant phenotyping.

• 
  
  

• Features information on advanced techniques for extracting phenotypes through images and image sequences captured in a variety of modalities.

• 
  
  

• Includes real-world scientific problems, including predicting yield by modeling interactions between plant data and environmental information.

• 
  
  

• Discusses the challenge of translating images into biologically informative quantitative phenotypes.

A practical resource for students, researchers, and practitioners, this book is invaluable for those working in the emerging fields at the intersection of computer vision and plant sciences.

Part I

Basics

1 Image-Based Plant Phenotyping

Opportunities and Challenges

Ashok Samal, Sruti Das Choudhury, and Tala Awada

CONTENTS

1.1 Introduction

1.2 Importance of Phenotyping Research

1.3 Plant Phenotyping Analysis Framework

1.4 Plant Phenotyping Networks

1.5 Opportunities and Challenges Associated with High-Throughput Image-Based Phenotyping

1.6 Image-Based Plant Phenotyping Analysis

1.7 Data Management for Plant Phenotyping

1.8 Computational Challenges in Image-Based Plant Phenotyping

1.8.1 Computational Resources

1.8.2 Algorithm Robustness

1.8.3 Inference from Incomplete Information

1.8.4 Large Phenotype Search Space

1.8.5 Analysis of Image Sequences

1.8.6 Lack of Benchmark Datasets

1.9 Looking into the Future

1.9.1 Imaging Platforms

1.9.2 Integrated Phenotypes

1.9.3 Learning-Based Approaches

1.9.4 Shape Modeling and Simulation for Phenotyping

1.9.5 Event-Based Phenotypes

1.10 Summary

References

1.1 Introduction

A plant phenotype is defined as the quantifiable and observable morphological, physiological, and biochemical properties of a plant, resulting from the interaction of its genotype with the environment. The concept of “genotype to phenotype” was first introduced by Wilhelm Johannsen in 1909, while working on germination studies in barley and common beans (Roll-Hansen 2009). Since a plant’s pattern of growth progression is influenced by climatic variations, phenotypes may record the ontogenetic properties of a plant, which refers to the developmental history of the plant during its lifetime, from germination to senescence. Phenotypes can be measured either by considering the whole plant (holistic phenotypes), or its individual organs (component phenotypes). Examples of holistic phenotypes include plant height, total leaf area, and biomass. Examples of component phenotypes include the area of individual leaves, the average cross-sectional area of the stem, and the volume of a fruit (Das Choudhury et al. 2019).

Phenotypes are regulated by the plant’s genetics and by its interaction with the environment, so that different genotypes will probably produce different phenotypes, and plants of the same genotype might have different phenotypes under different environmental conditions. It is similarly possible for plants of different genotypes to have similar phenotypic traits. Thus, the relationship between phenotypes to genotypes is many-to-many in nature. Some traits are influenced more by genetic factors, while the plant’s growing environment is more determinative for others. Ultimately, phenotypes are significant, as they influence resource acquisition by the plant and its yield.

1.2 Importance of Phenotyping Research

We are in the midst of a significant change in global climate patterns, where environmental stresses, e.g., droughts, floods, extreme temperatures, and disease outbreaks, are expected to increase in frequency and intensity in many regions and are predicted to reduce crop yields in the impacted areas. This is expected to be exasperated by the need to ensure food security for a growing world population in the presence of dwindling natural resources (FAO 2017). These complex challenges make it necessary to accelerate research for boosting plant yield and adaptation to local environments. Increasing and, in some cases, maintaining yields will require implementation of novel techniques in gene discovery and plant breeding, as well as advanced technologies such as precision agriculture (Chawade et al. 2019). Precise and fast assessment of phenotypes for crop varieties or populations is needed to accelerate breeding programs (Lobos et al. 2017), and to aid in the development of advanced management practices for precision agriculture. Plant phenotyping tools have the potential to accomplish these goals (Camargo and Lobos 2016).

Phenotyping can also play an essential role in the understanding of critical plant physiological processes. Traits, especially complex ones, reflect the biological processes that govern growth, phenology, and productivity of plants (Lynch and Walsh 1998). Understanding the underlying processes can lead to the “elucidation of the mechanisms impacting important ecophysiological traits” (Pauli et al. 2016).

1.3 Plant Phenotyping Analysis Framework

Figure 1.1 shows an overall schematic of a typical high-throughput plant phenotyping framework. The images of the plants are captured by multiple cameras of different modalities in an automated high-throughput plant phenotyping facility at regular intervals for a specific period of study. The images thus captured are first stored in a centralized database, and then transferred to the centralized server for analysis and integration with other datasets and/or for distribution to the different end-users for scientific discovery. The computing and data can be hosted locally or delivered through a cloud computing platform over the internet. The plant phenotyping analysis is a multidisciplinary research field at the intersection of computer vision, statistics, engineering, machine learning, plant science, and genetics. To fully exploit the benefits of image-based plant phenotyping, raw images, metadata, and computed data, as well as genomic data, must be stored in a searchable database so that they can be efficiently queried and analyzed for data mining and knowledge discovery. Significant insights can be obtained using these techniques, both within experiments and across multiple experiments in plant phenotyping, particularly in a large phenotyping facility that supports diverse experiments generating heterogeneous, high-velocity and high-volume data.

FIGURE 1.1 A typical framework for high-throughput plant phenotyping analysis.

1.4 Plant Phenotyping Networks

Due to the growing significance of phenotyping analysis and the need to develop long-term plans, a number of international, national, and regional plant phenotyping networks (summarized in Table 1.1) have been organized over the past decade (Carroll et al. 2019), with the following common missions:

TABLE 1.1
Plant Phenotyping Networks around the World

Network	Name	Web Location
APPF	Australian Plant Phenotyping Facility	http://www.plantphenomics.org.au/
APPN	Austrian Plant Phenotyping Network	http://www.appn.at/
DPPN	German Plant Phenotyping Network	https://dppn.plant-phenotyping-network.de/
EPPN	European Plant Phenotyping Network	https://www.plant-phenotyping-network.eu/
FPPN	French Plant Phenotyping Network	https://www.phenome-fppn.fr/
IPPN	International Plant Phenotyping Network	https://www.plant-phenotyping.org/
LatPPN	Latin American Plant Phenotyping Network	https://www.frontiersin.org/articles/10.3389/fpls.2016.01729/full
NaPPI	Finland National Plant Phenotyping Infrastructure	https://www.helsinki.fi/en/infrastructures/national-plant-phenotyping
NPPN	Nordic Plant Phenotyping Network	https://www.forageselect.com/nppn
Phen-Italy	Italian Plant Phenotyping Network	http://www.phen-italy.it/
UKPPN	UK Plant Phenotyping Network	http://www.ukppn.org.uk/

1. 1. Accelerating plant phenotyping research by representing a multidisciplinary community, comprising plant biologists, ecologists, engineers, agronomists, and computational scientists, to foster collaboration, innovation, and the initiation of multi-investigator and multi-institution projects.
2. 2. Promoting a framework for data standards to facilitate data sharing, accessibility, interoperability, and reusability worldwide.
3. 3. Incentivizing mutually beneficial research between public and private sectors.
4. 4. Facilitating the interdisciplinary training needed for effective translational plant phenotyping research.

As Table 1.1 shows, most of the efforts are concentrated around Western Europe. However, there are some networks located in other parts of the world, including Asia, Australia, and North America. These networks have considerably advanced the research in plant phenotyping, by managing extensive involvement of individuals in transdisciplinary and transinstitutional research collaborations globally.

1.5 Opportunities and Challenges Associated with High-Throughput Image-Based Phenotyping

Low-throughput plant phenotyping has traditionally been performed manually, with the help of various sensors or instruments, e.g., leaf area meters, chlorophyll meters, infrared gas analyzers, etc. Whereas some measurements, e.g., plant height, can be performed non-destructively and repeatedly over time, many other traits necessitate destructive measurements, e.g., weight, plant architecture, nutrient content, and water relations. Additionally, low-throughput phenotyping is labor-intensive and time-consuming, particularly for large experiments with hundreds or thousands of plants that require measurement of multiple traits (Reynolds et al. 2019). Furthermore, manual measurements, even by trained individuals, naturally entail ...