An image processing method for evaluating the quality of nanowire alignment in a microchannel is described. A solution containing nanowires flowing into a microchannel will tend to deposit the nanowires on the bottom surface of the channel via near-wall shear flows. The deposited nanowires generally form complex directional structures along the direction of flow, and the physical properties of these structures depend on the structural morphology, including the alignment quality. A quantitative analysis approach to characterizing the nanowire alignment is needed to estimate the useful features of the nanowire structures. This analysis consists of several image processing methods, including ridge detection, texton analysis and autocorrelation function (ACF) calculation. The ridge detection method improved the ACF by extracting nanowire frames 1–2 pixels in width. Dilation filters were introduced to permit a comparison of the ACF results calculated from different images, regardless of the nanowire orientation. An ACF based on the FFT was then calculated over a square interrogation window. The alignment angle probability distribution was obtained using texton analysis. Monte Carlo simulations of artificially generated images were carried out, and the new algorithm was applied to images collected using two types of microscopy.
Schematic diagram of the experimental setup with the optical microscopy.
Microscopy images of the nanowires accumulated in the microchannel, (a) upstream of the cross junction, in the vertical channel, (b) at the cross junction, (c) downstream of the cross junction in the vertical channel, (d) upstream of the cross junction in the horizontal channel, (e) immediately upstream of the cross junction, ( f ) immediately downstream of the cross junction, (g) downstream of the cross junction in the horizontal channel.
Image processing procedure used in the present study.
The vertically aligned nanowire and ridge images are shown in figures 4(c) and 7(c). The vertical branch in the ACF corresponded to a relatively low value of the ACF. Because ridge pixels are defined as binary values in the Cartesian coordinate system with an integer interval, the actual thickness of a ridge can vary with the nanowire orientation. The integer intervals between ridge pixels permit connections between neighboring pixels to be classified into 2 groupings: surface connections between horizontal and vertical pixels, or corner connections between diagonal pixels. The possible distributions of the ridge pixels are shown schematically in figure 8, and three clustering shapes, shown mainly in the ridge image, are illustrated in figures 8(b), (c) and (d), respectively. The orientations of each branch in the ACF were analyzed, and the maximum autocorrelation values for a given wire length were expected to be similar, regardless of the nanowire orientation; however, as shown in figures 8(b), (c) and (d), these values were inconsistent with one another due to variations in the connection type. To overcome these inconsistencies, two dilation filters were introduced: a cross filter Scross and a square filter Ssquare, which were convoluted with the ridge image. The refined ridge image can be expressed as
The original ridge image R and the filtered ridge images Rcross and Rsquare are shown, espectively. Note that the filtered ridge images are also defined as binary images. The effects of the dilation filter on the ACF can be seen in figure 8. Among overlapping wires, as shown in figure 7(c), branches with 90◦ angles were distinguished using the dilation filter, as shown in figures 10(b) and (c). These results indicated that the conditional treatment functioned satisfactorily.
Alignment directions at each interrogation window in the cross channel flow.
Alignment directions at each interrogation window in the SEM images.
The utility of the algorithm was tested by analyzing experimental images of nanowires. The alignment quality along the channel was analyzed by selecting seven points, as shown in figure 19. Real images for each test section are shown in figure 2. Figure 19 shows the orientational probability distributions. The results agreed well with the distributions expected based on the real images. Upstream of the cross junction, the flow was stable, and 60% of nanowires were aligned to within 10◦; however, the probability distribution was broadened at the cross junction due to the confluence of the two flows. Downstream from the cross junction, only 40% of the vertical flow and 30% of the horizontal flow of nanowires remained aligned within 10◦. The nanowire orientational distributions upstream of the cross junction were more anisotropic than the distributions downstream due to differences in the flow states. The confluence of the flows at the cross junction provided one approach to increase the isotropy of an orientational distribution. The dramatic changes in the alignment quality may be clarified by inspection of the graphical results of each interrogation window, as shown in figure 20. Almost all textons in images of nanowires upstream from the cross junction, as shown in figures 20(a), (d) and (e), revealed a single dominant orientation that coincided with the flow direction. In contrast, the images downstream from the cross junction, as shown in figures 20(c), ( f ) and (g), showed that the textons significantly increased along the nonflow direction. The alignment quality decreased downstream
of the cross junction, as a result of the confluence of the flows. SEM images of the nanowires were examined using the present method. Figures 21(a), (c) and (e) show SEM images with different imaging scales, and figures 21(b), (d) and ( f ) show the graphical analysis results. Because the imaging scales were varied, the interrogation window size w was varied. For each interrogation window, the graphical results revealed the dominant orientations.
PIV measurement of flow around an arbitrarily moving body
PIV measurement of fluids with moving interfaces,
e.g. fluid-structure interaction problems and two-phase flows, etc.
- Determination of the location and displacement of the interface.
Image processing techniques
An image processing technique based on user-defined textons that can be adapted for the accurate detection of interfaces.
A curvilinear interface in a PIV image is transformed into a straight interface in the regularized coordinates.
Rotating elliptic cylinder
Pendulum motions in a water
PIV Measurements of Flow around an Arbitrarily Moving Free Surface