ISSN 2070-7401 (Print), ISSN 2411-0280 (Online)
Sovremennye problemy distantsionnogo zondirovaniya Zemli iz kosmosa


Sovremennye problemy distantsionnogo zondirovaniya Zemli iz kosmosa, 2019, Vol. 16, No. 2, pp. 29-41

Recognition of Earth surface categories based on correlation portraits and its use in modeling atmospheric pollution dispersion

B.M. Balter 1 , V.V. Egorov 1 , V.A. Kottsov 1 , M.V. Faminskaya 2 
1 Space Research Institute RAS, Moscow, Russia
2 Russian State Social University, Moscow, Russia
Accepted: 10.03.2019
DOI: 10.21046/2070-7401-2019-16-2-29-41
We describe a method for recognition of template objects in multi- and hyperspectral remote sensing data. The method uses the matrices of correlation between spectral channels and compares them to correlation matrices of template objects using the correlation between these matrices as a measure of similarity (so called double correlation, DC). Templates are recognized in remote sensing data using the maximum value of DC between a template and a fragment of data. The method is sensitive to spatial variations within the fragment analyzed; thus, it can serve as a complement to methods of classification based on averaged spectra, such as maximum likelihood (ML). We describe adding DC to ML in classification of multitemporal Landsat data stacked like a hyperspectral cube for recognition of surface types, which are important in modeling of air pollution dispersion. Such imitation of hyperspectral data is incomplete because it lacks spectral continuity and is vulnerable to temporal drift of target objects but it is the only type of data systematically available for industrial air pollution studies. We calculate the probabilities of recognition and false alarm as functions of ML threshold and DC weight. For three surface types, which are similar spectrally but have different spatial structures and so are potential targets for DC (industrial, dense residential and low intensity residential), the effect of DC measured by improvement of the sum of missed target and false alarm probabilities is between 2 and 14 %. The resulting effect for a real problem of modeling the maximal hourly industrial pollutant concentrations in a city is 2–3 % for yearly maxima and up to 30 % for specific dates. Although for some districts and dates, the effect of DC is negative, on the whole, DC improves the modeling accuracy.
Keywords: correlation matrix, hyperspectral and multispectral data, template objects, maximum likelihood, pollutant dispersion, probability of recognition, probability of false alarm
Full text


  1. Balter B. M., Egorov V. V., Kuzmin A. A., Chekalina T. I., Primenenie spektral’no-korrelyatsionnykh metodov i teorii katastrof v izuchenii prostranstvennoi neodnorodnosti zemnoi poverkhnosti (Using spectro-correlational methods and catastrophe theory for studies of spatial nonuniformity of Earth surface), Isslevovanie Zemli iz kosmosa, 1991, Vol. 10, No. 5, pp. 10–15.
  2. Balter B. M., Balter D. B., Kottsov V. A., Obrabotka giperspektral’nykh dannykh po Zemle i Marsu (Processing of hyperspectral data on Earth and Mars), Sovremennye problemy distantsionnogo zondirovaniya Zemli iz kosmosa, 2006, Vol. 3, No. 1, pp. 68–76.
  3. Balter B. M., Egorov V. V., Kottsov V. A., Stalnaya M. V., Korrelyatsionnye portrety giperspektral’nykh dannykh distantsionnogo zondirovaniya (Correlation portraits of hyperspectral remote sensing data), Vserossiyskaya nauchno-tekhnicheskaya konferentsiya “Sovremennye problemy opredeleniya orientatsii i navigatsii kosmicheskikh apparatov” (All-Russia Scientific and Technical Conf. “Current Problems of Orientation and Navigation of Spacecraft”), Proc., Russia, Tarusa, 22–25 Sept. 2008, Moscow, IKI RAN, 2009, pp. 510–518.
  4. Balter B. M., Egorov V. V., Kottsov V. A., Sposob preobrazovaniya izobrazheniya (Method of converting the image), Patent RU 2586405, Reg. 10.06.2016.
  5. Balter B. M., Egorov V. V., Kottsov V. A., Novye vozmozhnosti korrelyatsionnogo analiza dlya system tekhnicheskogo zreniya (New capabilities of correlation analysis for systems of technical vision), Tekhnicheskoe zrenie, 2017, No. 1, pp. 53–59.
  6. Popa A., Balter B. M., Ganzorig M., Egorov V. V., Osobennosti korrelyatsionnoi struktury spektra opticheskogo signala, voskhodyashchego ot zondiruemykh ob’ektov na primere morskoi poverkhnosti (Characteristics of spectral correlation structure of optical signal from remotely sensed objects exemplified by sea surface), Isslevovanie Zemli iz kosmosa, 1988, Vol. 7, No. 3, pp. 23–30.
  7. Biehl L., Landgrebe D., MultiSpec — a tool for multispectral–hyperspectral image data analysis, Computers and Geosciences, 2002, Vol. 28, No. 10, pp. 1153–1159.
  8. Chekalina T. I., Popova I. V., Balter B. M., Egorov V. V., Correlation portraits and neural networks for spaceborne high-resolution spectrometry, Proc. ISSSR Intern. Symp., Maui, Hawaii, 1992, Vol. 2, pp. 1137–1149.
  9. Lee C., Landgrebe D., Analyzing High Dimensional Multispectral Data, IEEE Transactions on Geoscience and Remote Sensing, 1993, Vol. 31, pp. 792–800.