In the next few years the classification of radio sources observed by the large surveys will be a challenging problem and spectral index is a powerful tool for addressing it. Here we present an algorithm to estimate the spectral index of sources from multiwavelength radio images. We have applied our algorithm to SCORPIO, a Galactic plane survey centred around 2.1GHz carried out with Australian Telescope Compact Array and found we can measure reliable spectral indices only for sources stronger than 40 times the rms noise. Above a threshold of 1mJy, the source density in SCORPIO is 20 per cent greater than in a typical extragalactic field, like Australia Telescope Large Area Survey because of the presence of Galactic sources. Among this excess population, 16 sources per square degree have a spectral index of about zero suggesting optically thin thermal emission such as HII regions and planetary nebulae, while 12 per square degree present a rising spectrum, suggesting optically thick thermal emission such as stars and UCHII regions.