An algorithm for preprocessing thermal imaging data from an unmanned aerial vehicle obtained by a non-radiometric camera with an uncooled bolometer is proposed. The problem of interpretation of survey data is considered. It is caused by significant differences in the exposure of adjacent frames. Because of this, there are stitching boundaries at forming a composite image. In addition, natural objects (river, wetlands, vegetation, etc.) have extreme thermal properties, which do not enable identifying less pronounced anomalies of the archaeological site occupation layer’s soil features. It is shown that the noted problems can be solved by preprocessing the original images, including noise removal and brightness normalization. Due to the described filtration, a 'seamless' mosaic of thermal radiation of the survey site, the medieval archaeological site Kushmanskoye III settlement, was constructed. This algorithm was applied to four original images. As a result of processing, the area of extremely 'warm' segments caused by natural objects significantly reduced (in total – from 44 to 10 %). This allowed not only to 'save' thermal anomalies associated with soil features of the occupation layer, but also to increase the proportion of these segments in the transformed image (in total – from 33 to 72%). In addition, the segments of the 'intermediate' classes emphasize the features of the thermal distribution. It is established that, noise removal and brightness normalization of the original images helps decreasing the influence of natural objects with extreme thermal properties and, due to this, to detail the distribution trends of the humus layer of the archaeological site and the surrounding area. The results are consistent with the data of interdisciplinary studies (geophysics, soil drilling and archeology), as well as with the landscape position of the identified thermal anomalies and survey data in the visible range (orthophotoplan)