Sean
My expectation regarding identifying the mining areas is that you're going to need to digitize manually (as Dan suggests), but it IS possible that you MIGHT have luck with our segmentation tools to train it to find the appropriate symbology ("texture") - although best case I think you'll still have manual cleanup to do, and that may take just as long as manual tracing of the mining regions. I don't believe you'll have any luck with standard classification tools which focus on spectral content, not texture. (could you upload a screenshot of one of the areas showing "mining" and "non mining" so we can advise further?)
However, I want to redirect you with regard to the mosaicking. Please read about the mosaic dataset. You do not need to convert from SID to TIFF, and you should be able to create mosaics in minutes, not days. (they may not be perfect in the first 5 minutes, e.g. you will still have the collars to deal with, but if you can define a table of the easting & northing or lat/long of the collar on each file, that can be imported into the mosaic dataset to clip and hide the collar). This will be a "dynamic" mosaic you can use in ArcGIS to test the segmentation or manual digitizing.
Cody B
