Computer Vision for Green Plant Segmentation and Leaf Count

Introduction

Agriculture plays an important role in the economy of the country. The increase of population in a country leads to demand in agricultural products and hence, there is a need to increase the productivity of agricultural products. So, the agriculture related industries and the researchers are involving in research with great efforts, to continue agriculture for a prolonged period without any break. The important component of production statistics is yield rates. Most of the sampling operations for yield prediction are carried out manually. This may lead to possible occurrence of manual error and sampling error. To improve the productivity, these errors have to be reduced by making them as automation process. Computer vision techniques can be used to increase the productivity and reduce the production costs by automating the process and increasing the accuracy. This leads in arising of exciting problems in the field of computer vision. One of such problems is image-based plant phenotyping.

The Plant phenotyping refers to a quantitative description of anatomical, physiological, ontogenetical and biochemical properties of the plants. In recent times, several image-based methods for plant phenotyping are developed which are gaining more importance and, on par with growing commercial and scientific interest. The Plant phenotyping addresses the rural requirements without any limitation. One of the important rural requirements is to improve the crop yield, which requires an exceptional amount of research. The key plant phenotyping traits like plant yield, growth and development of plants can be characterized by total number of leaves in the plant and the leaf area (green plant region). The Plant image analysis can be used for predicting the crop yield. The Plant image analysis deals with plant measurements like anatomy, growth, surface, shape, etc. by analyzing the images of various plant organs such as root, leaves, etc.

Currently, the majority of the plant phenotyping systems depend on various custom-built or commercial solutions, ranging from the small-scale controlled environment to the field applications or automated large-scale greenhouses. The commercial solutions are more expensive and also require initial investment, which can be afforded by only few laboratories. Also, commercial solutions require appropriate analysis software and thus becoming vendor locked. Based on these limitations, many organizations embrace image-based plant phenotyping methods, which are localized for their environment, rather than using commercial solutions. Such image-based plant phenotyping methods are capable of addressing only specific phenotyping problems. Moreover, these image-based plant phenotyping methods cannot be easily implemented in different environments with various experimental settings, since many modifications in the design or in the complex image pipeline are necessary.

A vital amount of research have been performed in the field of plant phenotyping, which includes the research in leaf region segmentation, leaf counting, plants disease identification, and observing the development and growth of plant by analyzing the plant images. The image-based plant phenotyping is used to analyze the characteristics of plant growth and development, for estimating the crop yield. Plant growth depends on the total number of leaves and leaf area (Orlando et al., 2011), hence the leaf count measurement will be used for assessing the plant growth. The plant images may contain numerous leaves, branches, stems, and other objects in the background which meddle with the procedure. The leaf region must be isolated from the image, in order to count the leaves accurately. This chapter analyzes several methods existing in the field of image-based plant phenotyping and their limitations. Also, this chapter focuses on developing a new method which overcomes the limitations of existing methods.