The Exploration of Automated Image Processing Techniques in the Study of Scientific Argumentation

Introduction

Scientific argumentation is an important practice in which scientists generate and refine scientific knowledge of natural and man-made phenomena (Forbes, & Davis, 2008). Through scientific argumentation, scientists discover the extent of their knowledge, make sense of scientific phenomena, and critique their ideas and the ideas of their peers. Scientific argumentation also provides students opportunities to generate their own scientific knowledge by creating evidence via the collection and analysis of data, using this evidence in their reasoning, and persuading their peers (Bell, 1997). Because students engage in the aforementioned processes in producing written scientific arguments, analyzing their arguments can reveal their mental models of scientific phenomena and how they reason with data based on their mental models.

Even though scientific argumentation is primarily language based, those who engage in scientific argumentation can use visual representations of data as evidence to support their claims (Spiegelhalter, Pearson, & Short, 2011). This is reflected in the shift in the cultural significance attributed to the visual over the linguistic aspects of techno-scientific texts (Dimopoulos, Koulaidis, & Sklaveniti, 2003). This emphasis on the incorporation of images into knowledge generation and communication reflects the rapid and extensive spread of visualization technologies through our techno-centric society. Despite this technical advancement, however, little is understood about how students create, modify, and reason with visual images as they engage in scientific argumentation (Kerkhoven, Russo, Land-Zandstra, Saxena, & Rodenburg, 2016).

Although recent studies have examined the roles of visual representation in science learning environments (Bybee, 2014; European Commission & Education, Audiovisual & Culture Executive Agency [EACEA], 2012), in most of these studies, trained researchers employed qualitative image coding, which—like the qualitative methods used to analyze textual scientific argumentation—frequently relies on small samples (Spiegelhalter, Pearson, & Short, 2011; Evagorou & Erduran, 2015). Some of these studies explored the use of images in science textbooks (Spiegelhalter, Pearson, & Short, 2011), others analyzed visual artifacts produced by students in their study of science topics, and others examined students’ drawings of science and scientists. The literature on gender stereotypes also features several analyses of visual images (Sonka, Hlavac, & Boyle, 2014). As such studies indicate, image analysis can offer unique insights into students’ understanding of scientific phenomena.

The use of images has evolved with the emergence of new technologies—including drawings, advanced digital images, and three-dimensional models—which are used to produce tons of images every day. However, although image analysis and mining are now are used in science, engineering, and medical applications (Lehmann, 2005), few researchers have used automated image processing technologies to study science education (Xing & Gao, 2018; Xing & Du, 2018; Xing, Chen, Stein, & Marcinkowski, 2016). In this chapter, we use image processing methods to automatically extract features from images captured by students for use as evidence for their claims and then explore the relationships between these features and the students’ performance in argumentation. The primary research question of this study is, “Which features of images relate significantly to students’ performance in making scientifically valid claims and explaining the extent to which the data support their claims?” In this paper, we use a series of image processing methods to automatically segment specific regions of students’ images and to automatically identify several features of these regions.