Methods

This page is best approached sequentially as presented below. It will take 20-30 minutes to review the text and videos. Once reviewed you will have a complete method for gathering, handling, and processing qualitative spray data at high resolution.

Collectors

When selecting a collector, there are intrinsic characteristics of the collector that will determine the success of the collection and analysis. A properly chosen collector will exhibit no color or texture that would be interpreted by the software as a droplet stain, and will bond with the spray reliably so that residue does not inadvertently flake or dust off in handling, confounding the data.

Considerations

Surface roughness**: The surface should be as smooth as possible. While higher roughness will increase collection efficiency of very fine droplets, it can also create texture in a 4800 DPI scan. Evaluate this by scanning blank cards and observing the scans. A blank collector scan should be thresholded in imageJ and analyzed to determine where “zero” is. This procedure is defined in the Calibrating Black Collectors video below.

Absorptiveness**: How will the collector interact with the spray? Hydrophobic surfaces, such as the styrene petri dishes are constructed of, polypropylene or other plastic films, and metal plates may cause spray to bead, bounce, or possibly even roll off, confounding results. The dried residue may also not bond to the substrate, making handling and processing difficult. A good collector should make some degree of bond with the spray so that reasonable handling will not result in deposit loss or transfer. Too much absorptiveness, such as one would find with alpha-cellulose filter papers, is also not good, as the wicking will result in dilution of the stain and irregular spreading, again, confounding analysis.

Color**: How pure is the background color of the collector? The more homogeneous, the better. Droplet analysis will involve setting the software to recognize a range of colors that are known to be the spray tracer (detailed more below). If the collector has color impurities, this will confound the data. The blank collectors should exhibit ZERO instances of color that would overlap with the color of the tracer.

Sampler stability**: Is your collector stable in the conditions that will be encountered? This is particularly applicable to water and oil sensing papers, and coated collectors such as magnesium oxide coated slides (Chaskopoulou et al. 2013). Soft substrates, high amounts of dust or non-spray material, wet or high humidity environments, can confound collectors that rely on an impact or reaction with water. Also, the method described here involves contact with the surface of the flatbed scanner, so the collector should be stable when physically contacted.

Sampler thickness**: Thinner is better for this method. Multiple collectors can be placed on a flatbed scanner for simultaneous processing; as thickness increases, shadowing at edges of the collector can affect the result.

1 side or 2?**: If the front and backside coverage need to be differentiated, assure that either 1) both sides are in fact the same or 2) you have adequate collector area to fold so that back and front can be collected and analyzed uniformly.

Recommended collectors for 4800 DPI scanning:

Kromekote/ Chromalux: Note: as of late 2012, Mohawk Paper Company ceased production of the venerable Kromekote C1S (1 sided) and C2S (2 sided). While there are still some supplies around on the internet, Chromalux C1S White is Mohawk’s future replacement for Kromekote. Kromekote and Chromalux are both cast-finish high-gloss printing papers. They have a very smooth, high gloss surface that is designed to bond with applied inks and minimize wicking. This has historically made them the collector of choice when using colored sprays (Matthews et al. 2014). Chromalux is offered as C1S only in quite a few weights, so 2-sided samplers must be made by cutting longer strips and folding the collector. This makes scanning easier and less confusing (see the section on using scanners on the website for more information).

http://www.mohawkconnects.com/products/paper/mohawk-chromolux

Water and Oil Sensing Paper (WSP): Produced by Syngenta and widely available through agricultural retailers, this is a common collector. WSP is a dual-layer finish paper, where moisture deposited on the surface dissolves and reacts the two layers to yield a blue stain. While an excellent ad hoc tool for quick and easy results, WSP should be used with caution in studies for two reasons. First, any moisture, including high humidity, will develop the blue tint. On humid days or in deep canopy conditions, this can lead to a blue shift in the background color. Unintended blue can lead to difficulty in thresholding (see below) and a subsequent confounding of data. If using, care should be taken to quickly dry and store collectors in a low humidity environment. Also, due to process control limitations with the film thicknesses, droplets under 20-30 microns may develop unreliably or at varying rates in relation to their true size. (Syngenta, pers. comm.)

Collector processing: Collectors should be bagged individually. The computer will be able to detect the thresholder (chosen) colors at intensities much lower than the human eye can detect. Don’t risk transference between collectors to save a little money on plastic sample bags

Tracers

The primary assumption of this method is that tracer color of any type equals a deposit, no matter how small, so being able to differentiate the tracer from the collector and other contaminants is critical. There are a myriad of tracer options; here, a few will be discussed. Some should not be used: the author has attempted to use the fluorescent tracers pyranine and pyrene tetrasulfonic acid (PSTA) (Hoffman et al. 2014), however tracer concentrations in the spray needed to be so high to be detected by visible light optical scanners as to not be practical. Choose a color that is unlikely to be encountered in the environment for the highest quality data. The following concentrations are suggestions based on experience. Experiment with your situation before going to the field and adjust as necessary to gain the proper intensity.

Rhodamine WT 20%: Used at .2% volume/volume in a spray solution, rhodamine WT makes a strong magenta stain that is easy to threshold. There are many concentrations of rhodamine available, take care to source the 20% grade or adjust your concentration accordingly. Using rhodamine WT also gives the option to do quantitative fluorescent tracer analysis along with the qualitative work.

FD&C Blue #1: Is a clear, bright, food grade blue available as powder or liquid. It is reasonably priced, and the blue tends to blend in with green foliage in application making the residue less obtrusive than red dyes. If food grade is not necessary, blue marker or pond dye can be good, easily available alternatives. One that the author has good experience with is Cygnet Select™ (Cygnet Industries). Depending on the form purchased, you may need to experiment with the final rate, but target .2% of whatever concentrate you get, or mix powder to a 20% mass/mass solution and try .2% of this liquid to start. FD&C Blue #1 and Cygnet Select can be mixed with PTSA fluorescent tracer to allow both qualitative and quantitative evaluation.

Black Food Grade Dye: Was used primarily in the mid 80’s to early nineties to do spray studies, but choosing black dye will eliminate your ability to threshold out contaminants and background texture in your analysis, so avoid black shades unless it is otherwise not feasible to use a more vibrant color.

DAY-GLO® Fluorescent Dyes: Are really ultra-fine grind wettable powder dispersions, often heavily surfacted to overcome the natural hydrophobicity of the vinyl powder that the pigment is compounded in. Most of the available colors make good, clear, readable marks that are easy to detect with a visible light scanner due to the naturally hydrophobic powder pigment that migrates to the edge of the stain as the liquid evaporates. As a powder, handling can be a consideration due to dust-off; If DAY-GLO pigments are to be used, processing should happen relatively soon after exposure and with as little disturbance to the surface as possible.

Flatbed Scanners

In this method, the scanning and the image analysis happen as two separate steps. A flatbed scanner is the tool of choice here to gather the digital images. Using a flatbed scanner in place of a microscope gives the combination of high resolution, large sample area, and the ability to capture multiple collector images in 1 setup, saving time. Up to 9600 DPI optical scanners are available, and at first glance it would be attractive to scan at 9600 DPI, yielding a pixel size of 2.7 microns. However in practicality, a 2 centimeter square scan at 9600 DPI becomes a file of over 2 gigabytes, and each scan can take up to 20 minutes to collect. 4800 DPI scans of 2 centimeters square yield a much more manageable 40-50 megabyte image with a pixel size of 5.3 microns. If you want to experiment with 9600 DPI, by all means do. ImageJ will process the images and all the procedures listed here will still apply. Note that you may want to use smaller sample areas, and even so you will need a sizable amount of RAM in your computer to process the images.

Unlike in sheet feed, business card scanners, or microscopy, the flatbed scanner has the advantage of being able to close the cover to the scanner, totally excluding ambient light, thereby eliminating a source of variability. New scanners come on the market all the time; regardless of what scanner you use, it should be able to do the following:

  • Resolution: Scan at a minimum of 4800 x 4800 OPTICAL resolution in at least 24 bit color (RGB or HSB are minimum 24 bit). Some scanners use software interpolation to create a finer resolution than they are mechanically able to. Avoid interpolated resolutions.

  • Scan multiple areas: The software should be able to define and scan multiple areas in one pass and save each area to a discrete file. This will save a lot of time compared to scanning individual images. High resolution scans take time. Fourteen or more samples may be loaded in a flatbed at one time and processed in the background, freeing the technician to perform other tasks.

  • Save to an uncompressed format: The scanner software should be able to output to either .BMP format or uncompressed TIFF (preferred format for ImageJ).

  • LED instant-on light source: Consistent light is important to consistent images. Avoid scanners that require a warm-up time.


Canon LiDE 210

The Canon LiDE 210 is a compact flatbed scanner that is powered off of the USB connection, reducing desk clutter. With a recent software driver update, it now saves to both .TIFF and .BMP files in uncompressed mode, eliminating the extra step in ImageJ to convert them to .TIFF files. It is important to run a calibration on the scanner at each new session (power-up) or whenever the scanner is moved. Failure to do so can result in significantly degraded and liney images.


Compared to the Epson, the Canon is a bit slower in use due to several extra steps in the image capture, and the processing of scans as batches of up to 12 images at once.

Epson Perfection V37

The Epson Perfection is a very easy to use scanner with a robust software interface. It is physically larger and requires an external power supply, but in return for those tradeoffs has several advantages over the LIDE 210. The V37 scans natively to TIFF and several other formats, and self-calibrates at each start. It has a heavier lid and wider border than the LiDE 210, so may hold folded or thicker collectors flatter against the glass. It has the particular advantage of saving files as it scans them, making them immediately available to process in ImageJ.


Due to the sequential scan and save, the Epson allows more collectors to be scanned per setup than the Canon, and each image is saved as it is scanned, allowing immediate processing by ImageJ or other image analysis software.

Analysis Considerations

Image Analysis - other options:

If earlier in the process it was determined that 4800 DPI resolution was unnecessary for the study at hand, and 600 DPI or lower would suffice, other image analysis programs may offer added analytical benefits. Other commercial software is also available with more advanced capabilities. A few examples commonly used within the spray community are offered below. They will not be further detailed on this site, visit their sites for further information This is not an exhaustive list, but includes several options.

Depositscan Freeware is based on an earlier version of ImageJ with a customized input procedure and analysis interface. The USDA has not supported it with recent software upgrades but it is still used throughout the spray analysis community. It can be downloaded here.

Stainalysis Freeware by REMSpC is a standalone Windows® application that can perform particle analysis on low resolution, 8-bit color or black & white images. It has more sophisticated statistical outputs and routines for aerial applications. http://www.remspc.com/Stainalysis

Commercial packages with more advanced analytical capabilities built-in that may handle the larger high resolution scans and automate many of the processes that are detailed in this website:

ImageJ Software

ImageJ has a diverse user community and due to the wide range of possible applications, can be daunting; there is a lot of information and plugins out there. Some users may want to explore the amazing diversity of options eventually, but it is not necessary to get started. By focusing only on the things needed for particle analysis, you can be analyzing scanned images within 15 minutes of downloading.

A few notes and suggestions: The number one rule before you get started: BACK UP YOUR DATA! Some functions and batch processes can overwrite existing data with astounding speed and efficiency, and once done cannot be undone. Even once you become comfortable using ImageJ, keep original copies of your scans somewhere you are not working on them.

After initial use, it will be obvious that there are a lot of options and checkboxes that a user should familiarize oneself with. All manuals are online, but can be helpful to print and bind the 198 page manual for use alongside the software. It is also helpful to have a computer equipped with dual screens. Many ImageJ routines run as separate windows; it is helpful to have the extra desktop real estate.

Downloading and running ImageJ the first time:

Go to ImageJ’s website, http://imagej.nih.gov/ij/download.html and download the version appropriate to your system.


If you do not have administrative rights to your computer, a standalone version called Fiji that incorporates the Java operating system files can be downloaded (Schindelin et al. 2012). Fiji looks and acts for all intents and purposes identical to ImageJ. It has the advantage of being able to operate off a thumb drive, allowing anyone to use it regardless of administrative rights. Fiji can be downloaded from the ImageJ developers’ sight, www.imagej.net/Downloads


On the Sample Data link at left, you will find links to several sample sets of scans representing various droplet spectra. If you haven’t generated your own yet, download these to work with.

.

Open ImageJ. You will see the following:

Figure 3. ImageJ Graphic User Interface (GUI).

Start by opening one picture. Instructions for opening a whole folder are shown in video tutorial on the Website. To open an image, FILE>OPEN>{IMAGEFILENAME.TIF} If you downloaded the example files, a good starter is found in the ‘40 Micron VMD’ folder. Use ‘4-2-M-T_0001’. Usually it is best to start with a blank, but for the first example it’s better to use a card with a good number of stains. The file will open in a separate window.

Next, set up the parameters that will apply to the image by going to Analyze>Set Measurements. Figure 4 below shows the resulting ‘Set Measurements’ menu. Once the boxes are checked they will remain checked through the rest of the session until they are changed by the user. Select all the boxes checked here:


Figure 4. Image file opened, Analyze>Set Measurements menu displayed.

Figure 5. Set Scale. One pixel equals 5.29 microns for a 4800 DPI scan.

Next, go to Analyze>Set Scale. This will tell ImageJ what a pixel value is in microns. At 4800 DPI each pixel is 5.29 microns. This could also be set as millimeters or centimeters. For this example Microns will be used. Again, once these are set they will remain through the session. If ImageJ is closed they will need to be re-set.


The next step is the most interesting and important. Thresholding the picture tells ImageJ what in the image is a particle, and what to ignore. There is a more detailed video on the website explaining this. It is a good idea to experiment with this a little with each new set of scans. To Threshold, go to Image>Adjust>Threshold. Use the values shown next to the slider bars below along with the example image. Write these down for use with future images.

After setting the slider bars to the values shown in the image to the right, click the ‘Filtered’ button and you will see the image transform in preparation for particle analysis.

Figure 6: Image prior to thresholding, example threshold values shown at right.

Figure 7. Image after thresholding. Note intensity of droplets that were not previously visible.

Each red dot will be measured by ImageJ. In this scenario, all non-magenta colors outside the range specified in "Hue" have been excluded, eliminating most of the potential dirt, dust, and contaminants. As a side note, when you zoom in you will see a lot of random pixels and incomplete drops. These are all spray tracer. By running a Blank Calibration first, you can be confident that if a pixel has turned red in the thresholding, there was dye stain in it. Some pixels may have had a stain so faint that the scanner did not detect it; this is evident in the donut shaped stains.

Now run the Particle Analysis. Select Analyze>Particle Analysis and the Screen in Figure 8 will appear. Set the minimum size to 28 microns square. This is just over the size of 1 square pixel, so any stain of 2 pixels or more will be measured. There is more discussion on this in the website. Choose Display Results for detail on each stain, Summarize for summary statistics, and Include holes to measure the areas inside stains that might not have registered in the thresholding. Circularity of 0-1 means that no matter how misshapen a droplet, it will be counted. Once this is done, hit OK.

Figure 8. The Analyze Particles window.

About this window:


Size(micron^2): will allow you to limit stains to a certain minimum size. Since its based on area, if you want to only measure stains larger than 1 pixel, set the range to 28-Infinity. Assuming your image was 4800 DPI, 5.29 microns per pixel squared equals 27.98 square microns, therefore any stain 28 square microns or larger is at least 2 pixels. Often, as your percent area cover gets greater, you will see there is an increasing amount of what appears to be "noise" at the edges of the droplets. It's unclear whether this is reflection in the scanner due to the higher intensity of color in the target range that causes false positive color perception, or if something else is going on. There is a little bit of art in this that is explained more in the video on thresholding.


Notes: There are some (very general non-absolute) rules of thumb for setting this range. You may find these useful later.

  • Very low coverage (less than 2%): 0-infinity is the place to start.

  • 2-5% coverage: Read stains greater than 1 pixel (28 microns^2) or larger.

  • 5-15% coverage: read stains 4 pixels (111.93 microns^2) or larger.

Circularity: Leave this at 0-1. This will count all droplet stains no matter how misshapen.


Show: Begin with the default, nothing. This command allows you to also create a graphical output for the image along with the tabular data. For beginning, leave this blank to avoid confusion. Once comfortable with the basic routine, Experiment with the different output options. An excellent technical description can be found in the Manual, section 30.2, on page 133.


Check Boxes:

Display Results (checked): creates a table of each particle found and its properties that you have specified in the "set measurements" routine above. Each subsequent image's data will be added to this table.


Clear results (NOT checked): will delete the data from your prior image scan. Do not check this box!


Summarize (checked): will create a separate table from the Results table with summary data for the entire image (area covered, number of hits, etc). Each subsequent image's data will be added to this table.


Add to Manager (unchecked)


Exclude on edges (unchecked): If checked, removes any stain touching the boundary of the image. Since the primary use of this analysis is percent area covered and hits per area, this should not be selected, as it will reduce the value of both.


Include holes (checked): Sometimes droplets make circular stains with holes in the middle. Checking this box will fill that "donut hole" area in the droplet size and percent area covered calculations.


Record Starts (unchecked): output not used in this type of analysis.


in situ Show (unchecked): Also not used in this analysis routine.

Figure 9. Desktop showing output Results and Summary tables after analyzing 1 collector.

Results Output:


Based on the boxes clicked above, ImageJ has now produced two tables.

  • Results will contain individual measurements for each droplet. You can see here that 6919 individual stains were found, and the measurements associated with each stain.

  • Summary will contain the number of hits, area, percent area covered, average size (in square microns) of each droplet, and Feret’s Diameters (also in microns). A Feret Diameter is the maximum distance across an irregular shaped object. MinFeret is the minimum distance across an irregular object. Other measurements are calculated but not useful for the work being done here, they can be deleted later.

In this analysis there is no attempt to reconstruct a hypothetical droplet spectrum from the stains. For small droplet sprays, especially aqueous ones, the spread factors would typically be highly dynamic due to evaporation and the subsequent change in solution concentration, as well as variable interaction with the paper sampling surface. Also, if the spray was in an air-assist condition, the droplets typically have higher kinetic energy than a droplet experiencing normal sedimentation, so the impacts tend to be more irregular.

The user can choose to save the tables as Microsoft Excel® files or leave them open for the addition of additional data and save them at a later date. The Results tables can get large fast; this table has nearly 7000 hits for a single collector.


Congratulations, you have just completed your first high resolution image analysis using ImageJ!


Continued Analysis:

To continue analyzing collectors, simply close the image file (NOT the Results or Summary windows, these can be ignored or minimized) and open the next image file (any one from the samples, you’re just practicing). The process from here is very simple to repeat. You will not need to set the Scale or the Measurements again.

After opening, Click Image> Adjust>Threshold. The same values you added last time should be there, if not, adjust them to the ones you wrote down earlier. Click ‘Filtered’ to activate the threshold, then go to Analyze>Analyze Particles.

If all the boxes are still the same, and they should be, click OK. There is now a second collector’s worth of information in the Results and Summary tables.

Continuing to analyze images should only require a minute or two of time per image.

Saving regularly is highly advisable. It is very easy to accidentally close a window and lose your data.

Good luck!

Video Resources:

Process your first image:

The YouTube video at right walks you through the simple process of processing a scanned collector image in ImageJ.

Calibrating Blank Collectors

Calibrating the blank collector is critical to this process; it allows you to remove the texture and color from your analysis that is not part of your tracer stains so that you can be confident that all measured stains are in fact, tracer.


The video below shows this simple process.

Thresholding Images:

Thresholding "teaches" ImageJ to recognize a wide range of tracers and separate the measured stains from dust, dirt, and other background contamination. There is a bit of both art and science in it. The video below shows the process step by step.

** Should be analyzed before going into the field.