Written by the TreasureGuide for the exclusive use of the Treasure Beaches Report.
 |
| Table Adapting Signal Detection Theory To Metal Detecting. |
How much discrimination should a detectorists use? That isn't an easy question to answer. In fact, there is no one answer. It depends upon many factors.
Today I'll just get a start and try to build a conceptual foundation that I'll build on in future posts. The eventual goal is to give you the tools to make decisions on a more scientific basis, but that will probably take a few more posts.
In World War II when radar was being developed signal detection theory was being developed. There would be signals on the radar screen but would also often be noise. You'll see parallels with metal detecting. Signals are almost never pure and have to be separated from noise.
On the radar screen, sometimes the signal was caused by an actual plane but sometimes apparent signals were produced by other things such as atmospheric conditions or sources of electrical inference. In metal detecting, you want to detect a a good - one produced by a good target rather black sand, junk, or other sources of "noise."
As detectors became more sophisticated, we moved from a binary signal (beep or no beep) to more complex readouts that show additional information. Nonetheless, thinking in terms like those of signal detection theory can be useful. You can say that detectorists are trying to make dig/no dig decisions based upon the indicators provided by the metal detector.
The first thing to know is that there are two basic types of errors. In academic settings those types of errors are often referred to as Type I and Type II errors.
Using the terminology of signal detection theory, a Type 1 error is when a signal indicates the presence of a good target when there is no good target present. The Type I error is described as a "false alarm" in the illustration above. A signal that looks like a good signal when there is not a good target can occur as a result of noise or, in the case of metal detecting, things like junk in the ground, mineralization or electrical interference in the environment. When detecting in black sand, for example, the detector might produce a signal when there is no target present. That is the equivalent of a Type I error - or a false alarm.
Today a good target might be indicated by the combination of an auditory tone, and a conductivity number and perhaps other readouts. Like radar operators, detectorists are always trying to find "good" signals in noise. We do not have perfect target ID, although it is getting better. Conductivity numbers, for example, are seldom read as a continuously repeatable number without and other numbers be produced by the same target.
The detectorist can use various decision strategies. He may use different decision strategies in different situations.
If you are getting a lot of false alarms, one thing you might do is reduce the detector's sensitivity setting. That might help you cut down on false alarms, but you should be aware that there are costs involved with that strategy. A reduced sensitivity can cause you to miss smaller of deeper good targets.
The best decision will consider the risks and rewards of the tradeoffs. It will also be based upon some reasonable knowledge or estimates of the type and number of targets likely to be found in that environment. It will also include consideration of the values of targets, as well as other factors. I plan to get into those more in the future.
There are also many user characteristics that should be considered. And different users will be prone to different strategies. Some will be most motivated by the maximum possible find. Some detectorists are more optimistic than others. To put it in more common terms, some detectorists just hate to dig junk. Others treasure hunters will do anything and everything possible to find that one huge find that can take them years and years and cost them everything. Then there is a good probabilistic long-term decision to maximize the average expected value. We can spend a post or two on different decision strategies as described in game theory.
The likelihood of a Type 1 error (false alarm) depends in part on the decision criterion adopted by the detectorist. A lenient criterion increases the absolute number of finds but also the number of "false alarms". To put it in other terms, if you regard more signals as "diggable," you will increase the number of good targets dug, but also the number of junk targets dug. Conversely, a strict criterion reduces false alarms but may lead to missing more good targets.
Type II errors (misses) can be increased by reducing Type I errors (false alarms). There is a relationship. There are trade-offs. The risks and rewards must be assessed by considering a variety of factors so the best decision strategy can be selected
All of this might sound like I'm making the obvious sound more complicated than necessary, but I'm trying to establish a foundation of conceptual tools for future discussions. There is a vast body of relevant academic knowledge to draw upon.
People like to simplify things and that is good, but sometimes you can not simplify without losing important detail. We like to think that a certain conductivity number or a certain tone means this or that, but things are usually more complex than that.
I'm not saying that our new readouts are no help. They are not perfect, but the additional information they provide can be used in many cases to narrow down the probable ID of a target. Again, we are talking about probabilities. That is the point I'm working towards.
I'll leave it there for today and hope to pick up from there some time in the future.
 |
| Surf Chart from SurfGuru.com. |
Looks like a couple bumps but nothing real exciting.