Biological Shape Analysis - Proceedings Of The 2nd International Symposium

Pete E Lestrel

• Publisher: WSPC
• Year: 2013
• ISBN:
  9789814518420
• Topic: Biological Sciences
• Subtopic: Genetics & Genomics
• Index: Biological Sciences

This proceedings volume describes the current state of research dealing with biological shape analysis. The quantitative analysis of the shape of biological organisms represents a challenge that has now seen breakthroughs with new methodologies such as elliptical Fourier analysis, quantitative trait loci analysis (QTLs), thin plate splines, etc. The volume also illustrates the diversity of disciplines that are actively involved in the characterization and analysis of the biological shape. Some of the papers deal with the need to relate the underlying genome responsible for the actual observed characteristics of form. Moreover, many of the papers focus on the relationship of the shape to the processes that determine the biological form, an issue of major continuing concern in biology.

This volume brings together for the second time practitioners from a variety of disciplines who have been concerned with the necessity of applying new methods to the analysis of biological shape. Previous methodologies based on the conventional metrical approach (distances, angles and ratios), have not been able to adequately capture — in quantitative terms — the subtleties and complexities of biological form due to its irregularity. This volume represents an initial attempt to quantitatively characterize the biological form in both two- and three-dimensions, as it is actually perceived.

There is no volume available that deals with the subject matter of these Proceedings. The papers represent, as in the first proceedings, a unique look at: (1) new methodologies developed and used quantitatively describe the biological form; (2) the need to relate the observed biological shape to the underlying processes that determine the shape; and (3) the tremendous diversity of disciplines actively involved in the characterization and analysis of biological shapes. These range from physical anthropology, anatomy, genetics, botany, entomology, forensics, to applied mathematics, etc.

Contents:

• Agricultural Crop Selection:
  • Can Machine Vision Substitute for Plant Breeders' Eye? A Case of Whole Crop Shape Selection in Soybean Breeding (Seishi Ninomiya)
• Entomological Studies:
  • Genetic Architecture of the Developmental Buffering Machinery for Wing Shape in Fruit Flies (K H Takahashi)
  • Effect of Male Genital Spines on Female Remating Propensity in the West Indian Sweet Potato Weevil, Euscepes postfasciatus (N Kumano, T Kuriwada, K Shiromoto and H Tatsuta)
  • Morphometric Studies on the Variation of Male Lucanid Beetle Mandibles (H Tatsuta, H Iwata and K Goka)
• Human Morphological Studies:
  • Skull and Cranium: Craniofacial Morphology in Human Genetics (T Yamaguchi, R Kimura, A Kawaguchi, Y Tomoyasu and K Maki)
  • Vertebral Morphology: An Application of Fourier Transform of Two-dimensional Images: A Case Study of Human Vertebral Tuberculosis of Hokkaido Ainu (O Kondo)
  • Mandibular Studies: Representation of the Mandible as a Curve in 3-space: A Preliminary Study Using Fourier Descriptors (R Khullar, P E Lestrel, W Moon and C A Wolfe)
  • Mandibular Studies: Mandibular Shape Analysis of Plio-Pleistocene Hominins: Fourier Descriptors in Norma lateralis (P E Lestrel, C A Wolfe and A Bodt)
  • Whole Body Studies: Assessment of Body Image Perception: A Preliminary Study Using Elliptic Fourier Descriptors (P E Lestrel, N Miyake, M Ishihara and C A Wolfe)
• Primate Studies:
  • Craniofacial Covariation in Extant Great Apes: A Geometric Morphometric Study (D Neaux, F Guy, E Gilissen, W Coudyzer, P Vignaud and S Ducrocq)

Readership: Students and researchers in human biology, genetics and genomics, plant science and agricultural science, evolution biology and dentistry and sports medicine.

1. Agricultural Crop Selection

Can Machine Vision Substitute for Plant Breeders’ Eye? A Case of Whole Crop Shape Selection in Soybean Breeding

SEISHININOMIYA¹

Abstract

Crop shape is often a target of selection in breeding programs, assuming that crop shape directly reflects their productivity, determines light intercepting efficiency, lodging resistance, machine harvest efficiency, etc. Because the functional relationship between shape and productivity has not yet been quantified, such shape selection is still empirical or based on the visual judgments of skilled breeders. This paper describes several discriminant models that have been examined to simulate such human visual judgments in the case of soybean plant shape. First, shape features were defined for whole plant images and then the shapes were differentiated into classes using linear discriminant functions, multilayer perceptron, fuzzy logic and decision trees with the shape features. The intent of these models was to act as substitutes for human judgments in the process of classifying soybean shape into classes. Second, models were also developed where images were directly used as input data without any shape feature extraction. Multilayer perceptron, simple perceptron and Hopfield models were adopted as models with direct image input. The discriminant models with shape features performed well indicating the possibility of substitution for human visual judgments. In particular, the decision tree achieved a practical level of discrimination with a low error rate. These models, however, required the definition of shape features, raising the issue of generalization to other crops. On the other hand, the models with direct image input performed as well as the fuzzy logic, multilayer perceptron models with shape features. Though their performance was still worse than that of the decision trees with shape features, it was concluded that the direct image input approach was useful and could be generalized to apply to other crops.

INTRODUCTION

Whole crop shape is often an important factor for determining light intercepting efficiency, lodging resistance, machine harvesting adaptability, etc., which directly reflect crop productivity [1]. Soybean is one such example and the shape is a target of selection in breeding programs [2]. Because the functional relationship between shape and productivity has not yet been quantified, such shape selection is still empirically based on visual judgments of skilled breeders. Such an evaluation process, however, requires long experience of the breeders and evaluation results can be inconsistent even by a same breeder, because of the dependence on human subjectivity.

To avoid such subjectivity, a series of studies to substitute for human visual judgments in soybean breeding programs have been conducted. These studies utilized biological image analysis and computational statistics [2, 3, 4, 5, 6, 7, 8]. The purpose of this paper is to summarize those studies, by comparing the capabilities and features of a number of different models, which simulate human visual evaluation in the selection of whole soybean crop shape. Because crossvalidation was not conducted in some of the original studies, a part of the results here were recalculated using cross-validation. The procedure in this paper utilizes the same discriminant models as in the original studies for the independent comparison among the different models.

MATERIALS AND SHAPE FEATURES

One hundred and seventy six soybean cultivars were grown in the summer season of 1989 at the Nagano Prefecture Chushin Agricultural Experimental Station under the ordinal cultivation management of the station. A total of 875 individual plants of the cultivars were sampled by dissecting them at the nodes of cotyledons in the beginning of the pod growing stage. This is when the upper-most leaves were fully expanded and the shapes of the plants were almost fixed except in the cases of a few indeterminate cultivars. Then, each sampled plant was placed on a white board and photographed (Fig. 1).

Fig. 1. Definition of the shape features (see also Table 1).

Because a soybean plant grows in a mostly two-dimensional fashion, expanding its branches and petioles roughly every 180 degrees, an image of each plant taken in this way can represent its shape [2]. Some examples are shown in Fig. 2.

An expert soybean breeder categorized the 875 shapes of the sampled soybean plants (Fig. 3) into three categories: GOOD, FAIR and POOR. GOOD ones represent those specimens that the breeder intended to leave for the further breeding program while POOR ones are discarded. The breeder gave a FAIR rating to intermediate shapes. This shape evaluation data set of 875 plants was used for the first step of the study. For the second step of the study, a data cleansing was conducted in order to make the discriminant models more general (see the Data cleansing and decision tree section for the details).

The soybean plant part of each image was extracted by binarization [9] and 18 shape features of the binarized crop shape were defined ([2], Table 1) for the further analyses.

Fig. 2. Shown here are some examples of binarized shape images. All the shapes were normalized to have a unit height.

Discriminant models with shape features

As an initial study, the discriminant power of fuzzy logic, linear discriminant function and neural network were examined in order to simulate human visual judgment on soybean plant shape, using the original 875 image data and the shape features defined for them (Table 1).

First, the discriminant power of a fuzzy logic model was examined. Fuzzy logic is an idea to simulate human ambiguous control quantitatively and was originally proposed by [10] and the idea was utilized to simulate human judgment of the soybean shape [5]. To develop such a fuzzy logic model was a rather empirical process, which involved three steps. These were: (1) the selection of proper shape features, (2) the determination of fuzzy ranges and (3) the adjustment of fuzzy rule sets. All the steps were repeated to achieve a set of satisfactory fuzzy rules to evaluate the plant shape (Fig. 3).

Fig. 3. Examples of soybean shape as classified into three categories by a breeder.

Table 1. Shape features of binarized soybean crop shape defined by [2]. (See also Fig. 1).

After a manually time-consuming iteration process, a set of seven fuzzy ranges was finally developed. This set contained the same triangle membership functions (Fig. 4) and was designed with the rule set as given in Table 2.

Fig. 4. Fuzzy ranges and the membership functions used in this study. VS, S, MS, M, ML, L and V...

Table of contents

• Front Cover
• Half Title
• Title Page
• Copyright
• Dedication
• Preface
• Content
• List of Symposium Participants
• 1. Agricultural Crop Selection
• 2. Entomological Studies
• 3. Human Morphological Studies
• 4. Primate Studies
• Index