Adiabatic Quantum Computers (AQCs) are special-purpose devices that promise to speed up the resolution of hard combinatorial optimization problems by exploiting quantum mechanical phenomena. Despite representing one of the most mature quantum computational paradigms, AQCs are often benchmarked on spin-glass problems that conveniently mimic their internal structure, and their performances are usually compared to those of suboptimal Simulated Annealing solvers. In this work, we evaluate the capabilities of AQCs to extract features from a dataset of low-resolution satellite images of airplanes, exploiting an approach based on matrix factorization. We assess the performance of three generations of quantum devices provided by the D-Wave company by analyzing their behavior via selected evaluation metrics as a function of the problem size. We also compare against classical results obtained with a commercial solver. Additionally, we outline a parameter-tuning procedure that allows for harnessing the full potential of AQCs. Although the quantum devices still tend to underperform their classical counterparts, we observe the two most recent AQCs exhibiting superior performance for the tested instances with the smallest size. Additionally, our results indicate incremental performance improvements across new AQCs generations. These observations suggest cautious optimism about the potential future applicability of such computational tools.