Currently, photovoltaic panels constitute an important part of renewable energy systems in urban areas of developed countries. They are expected to generate greener electrical energy from non-polluting solar resources. Therefore, government agencies, electricity grid operators and decision makers encourage their setting up by funding and tax reduction. To avoid frauds with these substitute energies, several organizations are interested in detailed information, including localization and energy production, about these solar systems. Field surveys constitute one of the methods to obtain the above information on photovoltaic panels. However, this approach consumes a lot of time and can be very expensive, which leads to the use of other less expensive and faster approaches. The integration of remote sensing data is an interesting alternative for automatic detection of photovoltaic installations and their localization. Some remote sensing-based methods use high spatial resolution airborne/spaceborne images with a limited number of spectral bands. Such data do not allow an effective detection of photovoltaic panels principally due to their material properties. Indeed, when the visual properties of these panels are altered by specular reflections, their detection becomes difficult. High spectral resolution hyperspectral data may be considered to overcome the above limitation. These data are acquired by sensors that gather images from hundreds of narrow and contiguous bands of the electromagnetic spectrum, which offer unique opportunities for precise material recognition. In the investigation reported here, a hyperspectral-unmixing based method is proposed to detect photovoltaic panels and to estimate their areas. This approach is based on a new multiplicative nonnegative matrix factorization (NMF) algorithm which exploits known panel spectra. The designed approach, which can be considered as a partial NMF method, is applied to real airborne hyperspectral data acquired over the urban region of Toulouse, France. The obtained results (detection and area estimation) are confirmed by using a very high spatial resolution ortho-image of the same region. Also, these results are compared with those obtained by the standard multiplicative NMF algorithm introduced by Lee and Seung. This comparison shows that the proposed method yields much better overall performance than the considered method from the literature.