Indian Journal of Radio&Space Physics
VO"~0;t.ober 1993, pp. 301-305
~,J!~ _.
CO.v--"'.c..itj~
.....·/J\a~
ro 0. g...L \.. 'fII\ '
T>r 0
"G'~ ..~~.-~~uter algorithm of n~ise remo~ in acoustic,r~d~r echogramy
~'~f~
~lectrorriCS and Co~unication SciencesUnit, I~dian Sta~stical Institute, Calcutta 700 0:;)
Rc;;<;c••. ..d 7Deeember-lQQ2~ •.•••••MN4Pt'Ld.M 1lttn.. 1991-
~e(J;kk-scattered echo signals recorded in an acoustic radar system provide information on the characteristic patterns of atmospheric boundary layer (ABL) within a height of 1 km from the ground. These acoustic imageries are composed of range and intensity information and usually con- taminated with varieties of noises. An algorithm based on image processing technique was developed to remove most of the noises and a method of global thresholding for extraction of patterns from the echograms has been discusse~
1 Introduction
There have been significant developments over the last three decades in the domain of instrumen- tation and technique for measuring and analysis of atmospheric weather, climate, and environmen- tal pollution. Objective of the atmospheric boun- dary layer (ABL) study may be enunciated as identification and recognition of its forms and structures, recording of their distribution and as- sessment of those distributions along with mathe- matical modelling of inherent physical signific- ances for different ABL processes. For that, we should have appropriate observational programme for data. to be obtained, to put into permanent forms such that they lend themselves for rapid an- alysis, comparison· and interpretations. They are commonly used in (i) prediction of forthcoming environmental events over which man has little or no control and (ii) execution and planning of envi- ronmental or resource management. lInplementa- tion of computer techniques in these may ease the human burden in decision-making.
The echo returns are recorded in a 3-D facsi- mile chart recorder. Apart from the atmospheric information, in case of ABL studies, these records contain various types of background noises. The facsimile records of ABL structures contaminated with noises were digitized by a video-scanner to convert imagery into some suitable gray level mapping!. By utilizing image processing tech- nique, a computer algorithm has been developed and discussed here for isolating the unwanted noises recorded in along with true ABL informa- tion. In this work our aim is to remove or reduce noises, as far as practicable, to make atmospheric information clear and distinct by suitable compu-
ter technique. With the understanding of typical characteristics of noises, tlUs algorithm can distin- guish the required atmospheric structure from the unwanted noises.
2 Sodar backscattered signal and its noise behaviour
Acoustic echo sounding system provides an im- portant tool for obtaining characteristics ABL patterns from atmosphere within the height of 1 km above ground. Short pulses of audible sound with 10 W acoustical power at 2.35 kHz opera- tional frequency for a duration of 50 ms are transmitted vertically upwards and just after blanking time of 100 ms, the returning/reflected/
scattered echo signals are received and interpret- ed as indications of atmospheric dynamics at that instant. Received in a facsimile chart recorder, so- dar echo re~rns depict instant atmospheric struc- ture, time of occurrence, vertical height from which they return back, and intensity (i.e., dark- ness) of the echo signal.
Our group has already developed some algo- rithms to study atmospheric patterns2, still the in- tensity information.may provide better results in the boundary estimation of ABL structures and noise distribution in associated echo returns.
Knowledge on those noise. types can help one to choose a preferred location for installation of acoustic radar and assist meteorologists to inter- pret meteorological information content in the echogram when noise is present. Different types of noises and their possible classification are shown in Fig. 1. Some work on the removal of echo noises associated with rain/ drizzle, re- corder and digitization are reported here.
302 INDIAN JRADIO &SPACE PHYS; OcrOBER 1993
Sodar Echogram Noise
Atmospheric and Natural
Fig. I-Sodar echo noise types and their classification.
~I
J,"
Facsim'lle Picture
Digitization
Digitized Image
Intensity Mopp\ng
Histogram Drawing &
Smoothening
Range/Fine Threshold Selection
Noise Cleaning
Fig. 2-Block diagram showing different steps involved in sodar echo noise cleaning.
3 Computer algorithm for separation of noise from signal
Separation of informative signal (image) from the noise (non-image) in an echogram structure is a challenging task, still taking intensity informa- tion into consideration along with height of re- turning,and time of occurrence of intensity values, one may be able to clean noises in the structures.
A block diagram showing different steps involved in noise cleaning is given in Fig. 2.
3.1 Digitization and intensity mapping
The digitization technique adopted here may provide solutions for (i) intensity quantification of sodar image which is not possible by visual checking, (ii) processing of the scene data (analog) for automatic machine perception, and (iii) im- provement of pictorial information for easy hu- man interpretations. The image segmentation is a required step to separate unwanted signals from informative signals. A global thresholding tech
nique (described later) has been used here.
Through intensity quantification technique a lot of atmospheric information can be extracted from echo pattern. Pixel intensity I depends on h, the vertical height where from echo signal returns back, temperature variance C~ at that height, and strength of transmitting signal S]' For a fixed ST and h, if we can estimate I then prediction of C}
(Ref. 3) required for estimation of heat budget H may be possible.
A Vidicon camera (video-scanner) system was utilized to convert analog signal of sodar pattern in some suitable digitized format for computer processingl. The digitization window corresponds to 500 m in vertical height within 2 h duration along time axis to make 256 x 256 grid points or pixels, the resolution of each of which is 2m x 28s only. The sampling logic (process) is used to ex- tract discrete set of real numbers ranging from 0 to 255 levels of intensity value from the given picture. Accordingly map of gray levels has been
Ijll'i1III.1 I
TRIPATHY et aL:COMPUTER ALGORITIIM FOR ACOUSTIC RADAR ECHO NOISE REMOVAL 303
formatted into an intensity distribution table with clustering of 256 gray levels into group of 16 equal ranges.
3.2 Histogram drawing and smoothening
Different techniques are available for segmenta- tion but most of these individual techniques are problem-oriented. For analysis of atmospheric da- ta we have chosen histogram method4 and select- ed. a cluster of 16 gray levels. The pixel with high intensity value has been allotted a low gray level and vice versa.A graph of gray values against the number of pixels of that picture having those gray values has been drawn in the histogram.
Entire background points should be mapped in- to a single gray level cluster but, in practice, fac- tors like non-uniformity in illumination, false print or graduation mark on the recording paper rather distribute gray values of that background around the peak. Moreover peak with higher gray value represents background or noise as here whitest part of the sodar echogram covers almost the background. It may be assumed that left-hand portion of the histogram mainly covers image pix- els as shown in Fig. 3. The moving average tech- nique was applied to filter out spurious peaks with very small amplitude over their bases so that nature of the histogram may be smooth and con- tinuous.
3.3 Range and/or fine threshold selection
From smoothened version of histogram we ob- served that in lower gray level range, most of the required information clustered together around the peak(s) in left half of the histogram, while background noises with higher gray levels flocked together around rightmost peak. Information on the atmospheric pattern is mostly covered within 30-40% of echogram region while rest part is void of information. Valley between two peaks, in case of multimodal histogram, represents neither Image pixels nor background distinctly; rather pix- els within the -valley are of transitory nature and are forming (i) the interior edge point (more than 50% covered with image pixels), (ii) the exterior edge point (less than 50% covered with image pixels), and (iii) feeble signal point (either coming from a region of very low temperature variance or returning from a longer height). We have to con- sider necessary criterion for selecting the range of threshold, so that crude separation of image and non-image pixels might be possible. Here we choose the cut-off zone at that region where slope of the histogram intensity distribution curve sharply rises. Obviously, with the selection of the
Fig. 3-Histogram of sodar echo.intensity value.
optimal cut-off zone the picture may be recon- structed with less background noises.
As for selection of range thresholding, one may.
have to consider 16 gray values (for each of the cluster of gray values), and may have possibility of either incorporating some of the background noise for higher threshold selection or sacrificing of some image pixel information. Again for the finer adjustment, regarding separation of image and background noise, some optimum threshold- ing is needed to enhance the quality of the pic- ture5• In this step we have to determine the opti- mal single gray value out of the pre-selected thresholding zone already chosen. At this optimal value of the single gray level, histogram intensity distribution curve should have steeper slopping.
The optimal thresholding line is shown in Fig. 4 to demarcate image pixels (on left-hand side) from the non-image pixels (on right-hand side).
3.4 Noise cleaning
Optimal thresholding of gray values makes clear-cut distinction between the desired atmos-
304 INDIAN JRADIO &SPACE PHYS, OcrOBER 1993
pheric information (image pixels) and the back- ground noise (non-image pixels). With the incor- poration of some logic, non-image pixels can be deleted from the overall picture so that noise- free atmospheric structure delineates ori it. Re- constructed pattern is shown in Fig. 5(a).
For comparison on the degree of noise cleaning by this method, original re.corded picture of noisy pattern has also been placed side by side, as shown in Fig. 5(b). It should be noted that we are using only the global thresholding to separate noise. Application of this technique, which has al- ready been used by us, dearly produces distinct shape of atmospheric plume for analysis.
4 Conclusion
The present approach for utilizing the compu- ter for prediction and analysis of sodar echograms provides several advantages over existing manual interpretation by experts in this field. It saves their time in interpretation of the routine atmos- pheric structures recorded in sodar echo gram and also enables to establish standardization in analy- sis techniques as it is, more logical and provides better resolution.
The computer algorithm applied here is able to Fig. 4-Smoothened histogram and its optimum thresholding clear off some tyy~s of global background noises
line. associated with atmospheric pattern structures.
(a) (b)
Fig. 5-(a) Sodar echo pattern after optimum noise removal; and (b) original record of sodar echo pattern (noisy).
TRIPAlHY et al.:COMPUTER ALGORITHM FOR ACOUSTIC RADAR ECHO NOISE REMOVAL 305
providing them funds to carry out this project at the Banaras Hindu University, Varanasi, under the National MONTBLEX Programme, and to Prof: D Dutta Majumder of the Indian Statistical Institute for his constant inspiration in the pursu- ance of this work. Thanks are also due to. Mr N C Deb of Electronics and ·Communication Sciences Unit, ISI, for technical assistance.
Acknowledgements
The authors express their gratitude to the De- partment of Science & Technology, Delhi, for There exist complex varieties of atmospheric noises to adversely affect and obscure atmospher- ic information, and as the behaviour of these at- mospheric noises is varying widely due to various sources, no single algorithm would be effective for total removal of all sorts of noises. Different types of local thresholding and filtering techniques are required for the removal of noises formed by wind speed, rain, chirping of birds, recorder
markings, and other such sources. On-line record- References
ing of the back scattered signal and application of /1 Trip~thy S, De A K&Das J, Indian J Radio &Space Phys,
advanced image processing techniques may keep .) 21 (1992) 321.
the noise level to its minimum. 2 Chaudhuri B B, De A K, Ganguli A &Das J, Indian J Ra- dio &Space Phys, 21 (1992) 123.
3 De AK &DasJ, Mausam(1ndia), 1993, in press.
4 Davis L S&Rosenfeld A, IEEE Trans Syst Man & Cybem (USA),8 (1978)300.
).-Rosenfeld A&Kak A C, Digital Picture Processing, Vol. I .. (Academic Pres, New York), 1982.
