SODAR Pattern Classification by Graph Matching

Abstract

The sound detection and ranging (SODAR) system is an effective and active remote sensing tool to probe, in round-the-clock basis, the lower planetary boundary layer (LPBL) of the earth. The observations obtained from such SODAR systems can prove to be extremely useful for LPBL studies if classified and interpreted automatically. Manual interpretation of such data by an expert in this field is very tedious and error-prone. Hence, in this letter, we attempt to classify such data obtained from monostatic SODAR system, which is represented as echogram, by graph matching method. We initially convert the time domain's sample intensity variation, which is recorded as height versus time scale on echogram, to frequency domain for obtaining higher discriminating power of each sample of each class (as for example, thermal plume, inversion, etc.) of the echogram. For discriminating the individual class of the echogram, we use a measure of similarity. Thus, an expert system using graph matching method based on the frequency domain representation of each SODAR sample of each class is developed.