Qualitative image analysis of spray deposition has always been an important tool in understanding the overall quality of a spray application. Getting quality images of small droplet deposition, however, has been problematic and so the true potential of imaging spray deposition has not been reached. While there are many issues to resolve in any image analysis situation, two persistent issues have been a lack of adequate resolution and a lack of affordable, sufficiently powerful software that is capable of providing reliable, repeatable results.
Figure 1. Repeated analysis of the same deposition card using progressively coarser resolution and the effect on hits per area and percent area covered.
To illustrate the problems more clearly, a typical 600 DPI (dots per inch) optical scan, such as one would use in the methods described in the USDA’s Depositscan software (Zhu et al. 2011), would normally be considered very high resolution. A scan at this resolution has a pixel size of 42 microns width, making each pixel represent a very significant 1764 square microns. Therefore a “small droplet” stain of only 3 or 4 pixels width would be 126 to 168 microns diameter. Assuming a spread factor of 1.7, that 4 pixel drop would represent a droplet nearly large enough to be considered “non-driftable” by current ASTM Standard E2798 (105 microns).
Figure 2. Visualization of the results of resolution on droplet distribution. Voronoi overlay (polygons) added to provide perspective on the relative area a droplet would have to influence.
For evaluation of herbicide sprays, focus on this larger spectrum is usually acceptable, but for three dimensional canopy sprays, low volume, or ultra-low volume (ULV) sprays, this inability to register smaller stains is a severe limitation to accurate interpretation of the spray deposition. Higher resolution analysis has typically been done using microscopy, which is relatively time consuming and results in a very small field of view to achieve micron-resolutions. Once a researcher had an image of sufficient resolution, the public software options such as Stainalysis (Mickle 2004), are not able to handle the high resolution. Other commercially available software packages such as ImagePro Plus (Media Cybernetics) are able to provide sophisticated analysis at high resolution, but the complexity and cost limit access to this type of software package. Consequently high resolution qualitative analysis is not used much.
In three dimensional canopy, low-volume, and ULV spraying, the deepest portions of the canopy often receive as little as 2-3 orders of magnitude less than the maximum deposition, (Roten et al 2013) and yet disease control is often achieved in spite of these low levels. The difference in these situations between success and failure may be a difference of a few percent coverage. Depending on the spectra, that coverage may represent hundreds or even thousands of hits per square centimeter. Better understanding of the fine details of canopy coverage should reduce the need for expensive and sometimes unpredictable field efficacy tests. Figure 1 Illustrates the problem of resolution and the loss of both hits and percent area covered due to exclusion of small stains.
Fortunately, the market has brought solutions forward that make gathering and processing this data within the grasp of nearly every scientist. High resolution, high quality optical scanners can now be purchased for fewer than one hundred dollars. A 4800 DPI scan of 2 centimeters square, yielding a resolution of 5.3 microns per pixel, can now be achieved in only a few moments. With regards to software, the health and biotech community had the same problems as the spray community, so the US National Institute of Health (NIH) created an open source software platform called ImageJ. It is a Java-based package specifically designed to analyze high resolution images of small structures (like cells or drop stains). The Java base makes it operating system independent, running on Windows® Macintosh®, and Linux® systems. Unlike Stainalysis and Depositscan, which were specific applications and are not supported, ImageJ is supported by a robust global user group. It can be customized extensively by knowledgeable users and to that end libraries of hundreds of open source customized macros are available. Most importantly for this project, it is able to provide useful analysis of large, high resolution scans “out of the box”, without a high degree of user training or modification of the core software.