As CriglCragl said, I think the key problem is with the idea of the scientist monitoring "whether these signals/images are in a state of superposition". Whenever a property of a system is measured, like the positions of particles in the system, the experimenter always gets a definite answer, up to the limits of the resolution of their measuring equipment. Superposition isn't directly measured, it's part of the mathematical model used to predict the results of later measurements, given known results of earlier measurements. The earlier measurement is used to define a quantum state for the system, which assigns "probability amplitudes" to all possible values of measurement variables like position and momentum. These amplitudes are then modeled as changing over time in a deterministic way governed by some fundamental quantum equation like the Schrodinger equation for non-relativistic quantum mechanics. Then if you perform another measurement later, the probability amplitudes at the later time would tell you the probabilities of getting different results.
If you initially measured a particle to be in some precise position at some time t0, then the quantum state at time t0 would assign all the probability amplitude to that position and zero amplitude to other positions, but over time the Schrodinger equation would cause this to evolve into a quantum state that assigns nonzero probability amplitudes to multiple different positions, and physicists would describe this by saying the particle is modeled as being in a "superposition" of different positions. But if you do another measurement at a later time t1, you will only find the particle at one specific position--again, the probability amplitudes at time t1 that you calculated by following this method after the first measurement give you the probabilities of finding the particle in different positions.
The evidence that these probabilities are correct comes from experiments where we have multiple particles or systems that start out in the same initial state, and then we can look at the statistics of different possible later states they end up in. A classic example is the double-slit experiment, where we have a bunch of particles that we know started out from a source at a particular fixed location, and then we can measure the frequencies that these particles are later detected at different positions on a screen facing the source, with a barrier containing two slits positioned between the source and the screen. If you use the method I outline to calculate how the quantum state evolves between the point the particle is emitted by the source and the time it's detected at the screen, there'll be some nonzero probability amplitude for the particle having passed through both of the slits in between, and this has consequences for the probability distribution for the particle to be detected at different positions on the screen due to a phenomenon called "quantum interference". But you only get this interference pattern if there is no measurement of which slit the particle actually went through--if you do measure which slit it went through, then according to the standard rules for calculating probabilities you have to model this as "collapsing" the quantum state at the moment of measurement, which changes the probabilities for finding the particle at different locations at the screen on a later measurement there, destroying the "interference pattern". If you're not familiar with the experiment, there's a short video on it here which is pretty good despite coming from an overall very New-Agey movie, and a more detailed explanation on this page.
Decoherence is a phenomenon that is modeled in terms of the deterministic evolution of the quantum state that happens between measurements, it doesn't involve any assumption of collapse in itself. The intriguing thing about it, though, is it can mimic some of the results of measurement when you just look at one subsystem of a larger more complex system. For example, in a double-slit experiment where you send an electron through the slits, you can suppose that the electron will interact with some other particles in the vicinity of the slits, like air molecules, and construct a larger quantum state for all these particles in combination (an 'entangled' system) to describe this. If you use this larger quantum state to predict the probabilities for just the electron at the screen, you find the same probability distribution that you would have got if you had performed a measurement at the slits. This despite the fact that you explicitly did not model any discontinuous collapse at the moment the electron was passing through the slits, and used the continuous Schrodinger equation to model the evolution of the whole quantum state between the electron leaving the source and the electron arriving at the screen. It's as if the interaction between the electron and its environment (the air molecules) was able to mimic the effect of a measurement in retrospect.
So we have two scenarios that produce exactly the same prediction about the statistical pattern of electrons on the screen; one scenario where the electron is measured going through the slits and this is modeled as collapsing the wavefunction, another where it interacts with air molecules as it goes through the slits but this is analyzed in terms of deterministic wavefunction evolution of an entangled state without a collapse at that moment. But to get back to your question, there is at least a theoretical type of experiment that would distinguish between these two ways of modeling things, illustrated by a variation of the double-slit experiment called the delayed choice quantum eraser, which I talked about in this answer on the physics stack exchange. Here, instead of an electron interacting with a bunch of air molecules as it goes through the slits, a photon going through the slits to a screen is entangled with a single other photon. The photon detected at the screen is known as the "signal" photon, and the other entangled photon is the "idler". The interaction guarantees that the total statistics of all the signal photons on the screen will not show interference, instead you'll get a pattern just like what you would have seen if the signal photon had actually been measured at the time it passed through the slits.
However, whether you can actually determine in retrospect which slit the signal photon went through depends on how you measure the idler. Referring to the diagram I posted on the physics stack exchange, if you direct the idler towards one pair of possible detectors D3 and D4, then a detection of the idler at D3 indicates the signal photon went through slit A, and a detection of the idler at D4 indicates the signal photon went through slit B. If on the other hand you direct the idler towards a different pair of detectors D1 and D2, then neither of those tells you anything about which slit the signal photon went through--this setup retroactively is said to "erase" the which-path information that you could have potentially measured with a different setup. And in that case if you do a large number of trials in the which-path-erasing setup, although the total pattern of signal photons on the screen won't show interference, if you just graph the positions of the subset if signal photos whose idlers went to one specific detector (either D1 or D2), you will see an interference pattern in that subset. This sort of interference pattern can only be seen if the which-path information has been erased this way, so if you modeled the initial interaction between signal and idler as collapsing the quantum state so the signal photon definitely went through one slit or the other, you wouldn't predict any interference pattern.
So although it would be totally unrealistic in practice, in principle you could imagine doing a similar experiment where instead of the particle traveling through the slits being entangled with a single other particle, it could be entangled with some complex intelligent being you believe to be conscious, and who is totally isolated from the outside world, like an AI running on a quantum computer at near absolute zero. If this AI makes a mental note of which slit it observed the particle to go through, but its memory is subsequently "erased" in such a thorough way that even an exhaustive measurement of its brain wouldn't allow outside observers to reconstruct what it had observed, then it should be possible to recover evidence of interference in a way analogous to the delayed choice quantum eraser. This has actually been proposed as a thought-experiment by David Deutsch, the physicist who first formalized the idea of a quantum computer, and who is also a strong advocate for the many-worlds interpretation (in which there is no real collapse of the wavefunction, whether due to consciousness or other proposed reasons in objective collapse theories). See this paper, which describes Deutsch's thought-experiment on page 15, where first a "quantum artificial intelligence" observes the spin of a silver atom, and then:
Step 3 – Having experienced this “state of consciousness,” the quantum observer makes a public record of whether it has observed a definite spin value or not without revealing what exactly it learned.
Step 4 – The next step is to undo, by reversing the dynamical evolution, Steps 2 and 1. This is in principle possible, because “the steps involve only the quantum computer which can effect any desired unitary transformation upon the state of a subsystem of itself” (p. 223)
Step 5 – Finally the horizontal component of the spin of the silver atom is measured
The paper also quotes Deutsch's comment that the quantum AI "experiences the splitting and remerging of its own consciousness by observing physical evidence for which there is no alternative realistic interpretation". (Deutsch originally proposed this thought-experiment in the 1986 paper "Three experimental implications of the Everett interpretation" which was published in Quantum Concepts of Space and Time edited by Roger Penrose and Christopher Isham)