For instance, Mechem's sniffer dogs have displayed a 90% success rate on target filters, so statistically having three dogs per team perform cross-checking paths on the field should produce the 99.6% confidence level for the type of mines the dogs can detect.
I know nothing at all about this area, but presumably this only works if the 10% failure rate is random rather than based on some quality of the undetected mines?
But of course that isn't true in reality. There will be properties of the mines (burial depth, ground cover, degree of leakage of sniffable explosive material) that affect its chance of being found by all dogs searching for a mine.
If you wanted the parent poster's statistical chance, you'd want three 90% ways of detecting mines that are truly independent. Perhaps a dog, and infrared imaging sensitive enough to detect the explosive material (I have no idea if that's possible), and this article's mechanism.
Is that an actual official classification? It seems a bit low. "There were 1000 mines here, and we found 990 of them. Good enough, I say. Move in, and bring your kids!"