Spectrum Tree Methods

introduction spectrum tree is a powerful post analysis tool that is uniquely available for amics, which can greatly enhance the material classification of a sample spectrum tree is made possible because amics acquires and saves the raw data set during data acquisition, i e bse images and x ray spectra this allows spectrum tree to perform various data clustering to either refine mixed phases or define undefined materials in the analysis samples the availability of the x ray spectrum data from every point analysis does allow for data clustering based on specific x ray energy, x ray energy ratios, and total x ray counts the availability of the bse data also allows for clustering based on bse recorded for each acquired x ray point cluster is also possible using the standard material classification dictionary, or based on specific sollected spectra step by step instructions standard listing this clustering method allows the user to pass all the selected spectra through a normal classification process but does allow the user to adjust the matching tolerance used for the classification additionally, there is an option to exclude certain materials during the reclassification process using this clustering method unlike other clustering methods, standard listing clustering is controlled manually using the right hand panel (shown below) this menu also allows users to adjust the matching tolerance specific for this clustering routine, rather than use the original tolerance used for the whole sample the user can then define the number of materials to be exlcued from the classification by adjusting the number in the "# of exlusions" row the image below shows an example uisng three exclusion the user can then select which material to exlcude using the drop down menu by clicking the row next to the exclusion number in the example below, barite, diaspore and epidote materials will be excluded from the process when the "split" button is pressed to execute the clustering the value having the ability to excute a classifcation process on a reinfe group of spectra, is to allow users the flexility to allow certain spectra type (e g all unknowns) to be reclassified with different tolerenace setting allow redistribute the selected spectra into different groups, or reduce the number of undefined spectra by combining with other clustering it can also be possible to include bse or texture before doing a classification to fine tune material classification schemes in the current menu high priority materials = high tolerance, normal priority materials = tolerance, and low priority materials = low tolerance example video on use the standard listing option in spectrum tree energy clustering this clustering method allows the user to specify one energy window and an energy window range, both in kev, as well as a minimum total x ray count in this window to create two clusters of spectra with one batch “passing” these filtering parameters and the other “failing” these parameters as the x ray energy is linked to characteristic x rays, this method can be used as a pseudo elemental filtering of spectra however, the user should be aware that there are many overlapping x ray peaks from different elements, such as the overlap seen for mo and s (2 309 kev) this should be taken into consideration when using this clustering method, as it may cause unexpected spectra to occur in either the passed or failed spectra batches after selecting the spectra data source, the user should choose the updated input data source by clicking the centerline window then, in the clustering tool option, select the “energy clustering” option when successfully set up, the user will be able to click on the displayed spectrum plot to select the target x ray energy and the count threshold the default window interval is always set to 20 ev, but can be changed aside from changing the energy window, the user can also manually set the target energy and count threshold in the same menu sidebar as the user adjusts and modifies the filtering parameters, a set of blue bars will appear on the spectra plot, graphically indicating the energy window being filtered additionally, the user will notice a percentage value shown next to these bars, which will provide a quick reference to the proportion of spectra that will pass the filtering parameters once the parameters have been set to achieve the desired clustering of the spectra, the user needs only to press the “apply” button on the sidebar menu once pressed amics will collect all spectra that contain an x ray peak at the target energy level window, with the required minimum total count threshold, and assign them to a passing cluster, while the remaining spectra are assigned to a failed cluster example video showing the use of energy clustering in spectrum tree bse clustering this clustering method allows the users to split selected spectra based on the measured bse value users can set multiple bse cut positions which will generate multiple clusters of spectra containing the different measured bse values when this clustering method is selected, amics will generate a histogram measure bse for all the spectra, providing an indication of bse groupings, which can provide natural break points users can then select bse breakpoints (cuts) by selecting them on the histogram, or manually on the right hand panel to manually set bse cuts, users can adjust the number of cuts (i e # of cuts), which will add additional cut lines, and then bse values can be entered into the following "cut" rows clustering will then be initiated using the "split" button example video showing the use of the bse clustering with spectrum tree area clustering this clustering method allows the users to group spectra based on the size of the particle the spectra are associated with as such this clustering method will only work on granulated type segmentation mapping methods, and trying to use this method on other measurements will not work however, for the segmentation mapping method (and also bright phase search) once the data source has been selected and the area clustering bining criteria selected, amics will provide a rough split showing the spectrum spread based on particle area as with all other clustering methods, the user can add area splits by clicking in the lower window or using the manual interface on the right hand panel the application for this clustering is unique and provides a step towards material classification not just related to chemistry but also material morphology a possible example would be to cluster a material that is known to occur either as a large or small particle during sample preparation (during comminution or in a thin section), this clustering would allow the user to reassign these spectra into a massive form and "fine grained" for of this material example video showing the set up and use of area clustering in spectrum tree ratio clustering this clustering tool is equivalent to clustering using elemental ratio, but as amics is a spectra driven toolbox, this clustering uses energy ranges rather than elements as with all other clustering methods, the first step is to select the input spectra data source once this is selected, users can then select the "ratio clustering" binning criteria users can then manually enter the numerator and denominator energy ranges for the ratio calculation this is done by entering the lower (" from") and upper (" to ev") energy ranges for range 1 and range 2, with range 1 being the denominator and range 2 being the numerator alternatively, the energy ranges can be selected directly on the spectra plot shown on the left to do this hold down the "shift" key while dragging over the desired energy range, with range 1 being selected the first time, and range 2 selected the second time once both ranges have been selected amics will generate a histogram plot of the spectra based on the calculated ratio of the two energy windows this histogram plot can indicate the natural break points of the calculated ratio, and the user can then select one or more cuts on how to cluster spectra this can be done by manually entering the number of desired cuts or clusters (# of cuts), and then manually entering the ratio value(s) to cluster example video showing the use of ratio clustering in spectrum tree manual clustering this clustering method is similar to automatic clustering, but in this case, a seed spectrum is used to generate a group based on a tolerance setting assigned by the user the user can either select a spectrum in the left hand panel by using the up and down arrow or allow amics to find the most optimal seed spectrum by changing the "using seed" option to "best automatic search" as with automatic clustering, the user can adjust the "tightness" of the clustering by adjusting the tolerance, with higher tolerance generating a tighter (more similar spectra) cluster an example video showing the use of manual clustering in spectrum tree automated clustering this clustering method is more of a traditional k means clustering using the relative similarity of spectra to group into groups, which can then be assigned to the material definition unlike the other clustering methods set up for this method is completed manually using the right hand panel the image below shows an example of a typical automatic clustering setup, with user input for tolerance, min population, seed search, and cluster limit for those less familiar with the k mean clustering method the image below shows a conceptional idea of the clustering process in this example, there is a total of 5 clustering groups, which can be set by the user adjusting the cluster limit this value will only set the upper limit of clusters, it could still be possible to have fewer clusters generated if selected spectra are very similar amics will also only assign a group of spectra into a cluster, if there is a minimum number of similar spectra, which is set by the user by adjusting the min population in the example below, there are spectra that are similar (greyed out), but do not meet the minimum number of 10 spectra, and so have not been assigned to a cluster finally, the user can define how similar spectra need to be y adjusting the tolerance, which is the measure of the required difference in the image example below the short red line represents the idea of the relative difference between spectra setting different tolerance levels will adjust the size of the grouping in the image below a higher tolerance setting (light blue and light green) creates a tighter cluster, compared to the low tolerance examples (light grey and light orange) the final setting is the seed setting, which is a random number used to initiate the clustering process by randomly setting the first center point to calculate the spectral difference users should be aware that automatic clustering is not an ideal method to be used on larger projects with multiple specimens this is due to the fact that each calculation of the cluster difference specimens will generally generate different clusters, due to the random initial difference calculation as such if used as a step in a reclassification process, the final result for each sample will not be consistent in general, this clustering is effective in grouping spectra together to create a new material definition which can then be used to reduce the number of undefined spectra example video of how to use automatic clustering in spectrum tree