The detection of fireballs streaks in astronomical imagery can be carried out by a variety of methods. The Desert Fireball Network uses a network of cameras to track and triangulate incoming fireballs to recover meteorites with orbits and to build a fireball orbital dataset. Fireball detection is done on-board camera, but due to the design constraints imposed by remote deployment, the cameras are limited in processing power and time. We describe the processing software used for fireball detection under these constrained circumstances. Two different approaches were compared: (1) A single-layer neural network with 10 hidden units that were trained using manually selected fireballs and (2) a more traditional computational approach based on cascading steps of increasing complexity, whereby computationally simple filters are used to discard uninteresting portions of the images, allowing for more computationally expensive analysis of the remainder. Both approaches allowed a full night’s worth of data (over a thousand 36-megapixel images) to be processed each day using a low-power single-board computer. We distinguish between large (likely meteorite-dropping) fireballs and smaller fainter ones (typical ‘shooting stars’). Traditional processing and neural network algorithms both performed well on large fireballs within an approximately 30 000-image dataset, with a true positive detection rate of 96% and 100%, respectively, but the neural network was significantly more successful at smaller fireballs, with rates of 67% and 82%, respectively. However, this improved success came at a cost of significantly more false positives for the neural network results, and additionally the neural network does not produce precise fireball coordinates within an image (as it classifies). Simple consideration of the network geometry indicates that overall detection rate for triangulated large fireballs is calculated to be better than 99.7% and 99.9%, by ensuring that there are multiple double-station opportunities to detect any one fireball. As such, both algorithms are considered sufficient for meteor-dropping fireball event detection, with some consideration of the acceptable number of false positives compared to sensitivity.

In biology, cell counting is a primary measurement and it is usually performed manually using hemocytometers such as Malassez blades. This work is tedious and can be automated using image processing. An algorithm based on Fourier transform filtering and the Hough transform was developed for Malassez blade grid extraction. This facilitates cell segmentation and counting within the grid. For the present work, a set of 137 images with high variability was processed. Grids were accurately detected in 98% of these images.

We have explored the possibility of measuring small changes of orientation within grains by electron backscattering diffraction (EBSD), in the scanning electron microscope. Conventional orientation maps (using EBSD) index the orientation of each position on the sample separately. This does not give accurate results for small differences of orientation. We have studied methods of measuring small changes in orientation by measuring the change from one EBSD pattern to the next directly, without indexing either. Previous workers have measured the change of position of a zone axis in the EBSD pattern. We have compared this with an alternative method, which we show to be superior, of measuring the shift of the peaks in the Hough transform from one diffraction pattern to the next. This means that we are measuring the change of orientation of sets of crystal planes within the grain, rather than measuring the change of orientation of zone axes. We show that it is possible, with a standard EBSD configuration, to measure the shift of the Kikuchi bands to a precision of about a 10th of a pixel, which corresponds to a change of orientation in the sample of about 0.1 mrad (0.006°).

Electron back scatter diffraction (EBSD) mapping and indexing has rapidly come into widespread use. However, inadequate attention has been paid to the details of the method. Many of the algorithms in current use were chosen because they were the first ones that were found to work, rather than because they were optimum. Results of systematic study are presented. It is shown that more than one method can successfully correct a sampling artifact, that there is an optimum binning ratio, that Gaussian filtering provides an alternative to “butterfly convolution,” that better alternatives for mapping image quality than those in current use are available, and that saving all the original patterns is practical and advantageous.

We study convergence in total variation of non-stationary Markov chains in continuous time and apply the results to the image analysis problem of object recognition. The input is a grey-scale or binary image and the desired output is a graphical pattern in continuous space, such as a list of geometric objects or a line drawing. The natural prior models are Markov point processes found in stochastic geometry. We construct well-defined spatial birth-and-death processes that converge weakly to the posterior distribution. A simulated annealing algorithm involving a sequence of spatial birth-and-death processes is developed and shown to converge in total variation to a uniform distribution on the set of posterior mode solutions. The method is demonstrated on a tame example.