The second two chapters describe a novel high precision O2analyzer that was initially developed to measure AQ and a related general purpose data acquisition system that was developed alongside the O2analyzer.Automated image analysis techniques enable the non‐ destructive phenotyping of large plant diversity panels. The 1001 Genomes Project is one example of such a panel; it comprises 1135 sequenced natural accessions of Arabidopsis thaliana Heynh. sampled from a wide range of environments . Combining these high‐quality genetic resources with high‐throughput phenotyping methods enables powerful genome‐wide association studies. One technique for evaluating the developmental traits of such large diversity panels is growing the accessions in agar‐ filled culture dishes. This allows root traits to be quantified quickly using high‐throughput image analysis methods. The plants are not destroyed or contaminated in the process and can therefore be photographed at different stages of growth. One disadvantage of this approach is that the rosettes are askew, so rosette area is usually not assessed even when the leaves are visible in the photographs. Quantifying both root and shoot characteristics is usually preferable because many plant processes involve both organs; for example, nitrogen acquisition and allocation involves root uptake from the rhizosphere, assimilation into organic forms in both the roots and shoots, and translocation throughout the plant . Studying this process requires precise measurements of both the roots and shoots, plastic pot which has previously been technically difficult. Here, we show that leaf area measured from plate images is accurate even when the rosettes are somewhat askew and can therefore be used for rapidly phenotyping large image sets of Arabidopsis seedlings. As part of a larger study to examine the genetic basis of plant adaptation to different nitrogen forms and concentrations in the rhizosphere , we measured leaf area from more than 2000 images of Arabidopsis seedlings on agar plates.

To determine whether rosette area measurements taken from plate images are sufficient for shoot phenotyping, we compared them to both measurements from images of the rosettes photographed from directly overhead and seedling mass. To compare the overhead and plate image rosette area measurements, six different natural Arabidopsis accessions were planted on agar plates containing a base nutrient solution consisting of 2 mM CaCl2, 2 mM KH2PO4, 2 mM MgSO4, 1 mM KCl, 0.75 mM MES, 0.5 μM CuSO4, 2 μM MnSO4, 25 μM H3BO3, 42 μM FeNaDTPA, 2 μM ZnSO4, 0.5 μM H2MoO4, and 0.8% agar. Different concentrations of sucrose were added to the base media to ensure that there would be a variety of different‐sized seedlings. After planting, the plates were kept at 4°C for four days and then placed into a growth chamber with a 14‐h day/10‐h night cycle. After 12 days of growth, rosette area of the plants was measured in two ways, first from photographs of the seedlings in the plates and second from a photograph of the rosettes placed upright on paper. All photographs from this image set were taken with a Pixel 3A cellphone camera . A total of 58 seedlings were grown and measured this way. As part of a larger study investigating plant responses to different nitrogen forms and concentrations in the rhizosphere, we quantified the rosette area from plate images and compared it with seedling mass. A total of 148 Col‐0 seedlings were grown under 10 different nitrogen conditions with either nitrate or ammonium as the sole nitrogen source at concentrations ranging from 0.05 mM to 5 mM. After 12 days of growth, the plates were photographed and the seedlings, including both roots and shoots, were excised and weighed.As another part of the aforementioned study, more than 2000 images of Arabidopsis seedlings on agar plates were collected. This image set was generated from an experiment in which the 1135 natural accessions of the 1001 Genomes Project were grown under four different nitrogen conditions: 0.1 mM and 1 mM nitrate using KNO3 as the sole nitrogen source and 0.1 mM and 1 mM ammonium using NH4HCO3 as the sole nitrogen source.

The seedlings were grown under long‐day conditions . The closed plates were photographed 12 days after planting using an EOS Rebel digital camera fitted with an 18–55 mm EF‐S lens . The root traits, including primary root length and number of lateral roots, were estimated from the images using RootNav, image analysis software that allows the semiautomated quantification of complex root system architectures .Some of the image sets did not have a red two‐dimensional scale present, making them unsuitable for rosette area measurement using existing methods such as Easy Leaf Area . We developed our own image processing workflows in Python, which were able to use a scale if it was present or, alternatively, to detect the area of the agar plate to serve as a scale. These workflows use the PlantCV package for most of the image‐ processing functions. The general steps in the workflow are cropping the image to the plate region, leaf identification and pixel counting, and scale identification. Cropping the image to the region of interest was done to save processing time and eliminate background features that could be mistaken for objects of interest. This was done using binary thresholding or edge detection to separate the agar‐filled culture dish from the background . The choice to use edge detection to identify the plate versus binary thresholding was dependent on the image set used. The detection of the agar plate also allows for the rotation of the image if the plate is not correctly aligned within the image. Leaf identification was performed using binary thresholding and object detection . The specific color channel and threshold value used to identify the leaves varied between the different image sets due to different background and lighting conditions, but as long as the images within a set are taken against the same background and with the same lighting conditions then these values should remain consistent for processing the entire set. For the validation images, the “C” channel of CMYK color space was used to identify the leaves, whereas in the diversity panel image set the “B” channel of L*a*b* color space was used.

To determine the appropriate threshold values for an image set, we used the plot histogram function in PlantCV. This function is used to visualize the range of pixel intensity in the color channel of interest. For image sets with lower contrast, grow bag the histogram equalization function was used to make thresholding easier. To simplify leaf identification, an ROI was defined for the top section of the cropped image where the leaves are found. Objects detected within the ROI were grouped into six shoots using clustering. The image moment of each shoot in the binary image was used to calculate the number of pixels that made up the leaves in each seedling. Scale identification was performed either by using a reference scale that was placed within the image or using the plate itself as a scale. For dedicated reference scales, the same general process that was used to identify and count pixels of the leaves was used for the scale. Once the number of pixels in the scale or the number of pixels making up the plate were measured, the rosette area could be calculated. For the large image set of the diversity panel, we automated the workflow in a Python script, which took approximately four hours to process all 2000 images. We were able to estimate rosette area for over 90% of the seedlings that successfully germinated, resulting in 8964 individual measurements. Many of the seedlings that were not measurable had fallen below the middle of the plate and were not within the defined ROI. It is important to note that the parameters used for the various transformations, such as thresholding, grayscale conversion, and scale calculation, are specific to the image set. These parameters would need to be modified when using a different image set, but the general steps would still apply.The strong positive linear relationship between the rosette area measurements taken from the plate images and those taken from photographs of excised rosettes demonstrates that using plate images for shoot trait analyses can yield meaningful phenotype data with minimal effort. While the correlation between rosette area measured from plate images and seedling mass was not as strong, it was still sufficient to indicate that this is aviable method for estimating plant growth. A lower correlation between these measurements is also to be expected because the seedling mass includes both shoot and root mass and is therefore not as specific to shoot growth as is the rosette area. We were also able to apply this analysis to an image set generated for the purpose of root phenotyping, allowing us to obtain additional valuable phenotypic information. The rosette area measured using this technique across a large Arabidopsis diversity panel was found to be heritable and showed a significant response to rhizosphere nitrogen form and concentration. These results were in line with other developmental traits measured using established techniques, such as primary root length measured using RootNav . Agar plate images are widely used for the non‐ destructive measurement of Arabidopsis root traits. Here, we showed that useful shoot trait information can also be collected from these same images, enabling simultaneous root and shoot phenotyping. This can be done quicklyand is easily automated, making it suitable for large image sets. The images can be captured and analyzed without the need for specialized imaging equipment or dedicated phenotyping facilities. The agar plate itself can be used as a scale, enabling the analysis of image sets without dedicated two‐dimensional scales. With the procedures described here, image sets generated for root phenotyping in other studies might also provide data about shoot phenotypes without much additional effort.

The field of robot guidance has seen great advancement thanks to advances in Machine Vision and Machine Learning. Palletizer systems, comprised of vision guided pick-and place robots along a conveyor have become commonplace in manufacturing and logistics, reducing labor costs and handling heavier loads than humans are capable of handling . In the field of robotic surgery, neural networks have been developed to automate repetitive tasks based on input from cameras, reducing surgeon fatigue during long procedures. Permeant magnets could be a useful positioning aid in cases where clear line of sight is not available. For example, surgical robots have been incorporated in the insertion of pedicle screws during spinal fusion surgery, but only as far as aligning a surgical tool to the spine. The actual insertion of screws is highly dependent on the feel and experience of the surgeon. If some part of the screw could be magnetized, magnetometers could provide useful information about its position in the body. There has been some work on magnetic object tracking. Wahlstrom used an array of 4 magnetometers to track magnets from the opposite side of a piece of plywood using an Extended Kalman Filter, with RMS position error of 4.95mm and orientation error of 1.85 degrees. This work will attempt to calculate magnet positions with greater accuracy using a larger array of magnetic field readings. To avoid the increased cost of using a large number of magnetometers simultaneously, one magnetometer is positioned at different locations in 3D space. With readings taken from a large grid of points, existing nonlinear optimization algorithms can be used to compute the position and orientation of the magnets. In order to carry out this task, a system had to be designed and built to position a magnetometer in 3 dimensions. An alternate use that this system was created for was the characterization of magnetic devices fabricated by other members of the Magnetic Microsystems and Microrobotics lab. Measuring fields surrounding MMM lab devices will help in calculating magnetic forces and experimentally validating simulations.Agriculture is a key human activity in terms of food production, economic importance and impact on the global carbon cycle. As the human population heads toward 9 billion or beyond by 2050, there is an acute need to balance agricultural output with its impact on the environment, especially in terms of greenhouse gas production. An evolving set of tools, approaches and metrics are being employed under the term “climate smart agriculture” to help—from small and industrial scale growers to local and national policy setters—develop techniques at all levels and find solutions that strike that production-environment balance and promote various ecosystem services.