CHAPTER 2 A NEW SCHEME FOR SATELLITE RAW DATA PROCESSING AND IMAGE REPRESENTATION

Transcription

1 40 CHAPTER 2 A NEW SCHEME FOR SATELLITE RAW DATA PROCESSING AND IMAGE REPRESENTATION 2.1 INTRODUCTION The Chapter-1 discusses the introduction and related work review of the research work. The overview of the proposed satellite image processing schemes is also discussed. The present Chapter deals with the development of new schemes for satellite RAW data preprocessing and image representation for meteorological applications. Due to the geometrical incompatibility, a raw satellite image may not exactly be superimposed on a standard map or another processed image. Multi-temporal analysis, data mosaicking, interpretation, and another manipulations require a comparative and combinative analysis of images of the same area that requires a compatible spatial coordinate system. Moreover the atmosphere, lying in between the sensor and the target has an impact on the radiation process and that ultimately affect the remotely sensed data. Therefore, to avoid the above distortion, and to eliminate the noise and to bring the image to a usable condition, data preprocessing is very essential. The intent is to remove any undesirable image characteristics that have crept in during the image acquisition process and to restore the image as close approximation of the original scene as possible. 2.2 BACKGROUND Meteorological image data from the imaging satellite is acquired and archived as per the operational imaging schedule. Processing of the data is conventionally taken up as offline activity as per the application requirements. In order to carry out the tasks, two kinds of systems are employed, namely the Data

2 41 Reception (DR) system and the Data Processing (DP) system. The Data Reception System is interfaced with the data acquisition hardware and is responsible for the scheduled reception of image data from the satellites and further archiving of this data into mass storage devices like NAS (Network Attached Storage) or SAN (Storage Area Networks) systems. Preprocessing of image data is a process, in which the raw data under study is subject to a sequence of processes that remove/correct the errors, noise and offsets in the raw data thereby improving the image quality, thus making it suitable for further processing of specific interest. The electrical signals generated by the coordinated functioning of the onboard imaging instrument sub-systems are recorded onboard into the stream of digital data. The data are formatted and transmitted to the ground using suitable radio frequency communication mode. 2.3 PROPOSED SCHEME The image raw data generated by VHRR scanning mechanism is voyaging many stages. Since image data transmitted with telemetry (both analog and digital), servo profile, attitude, space look, black body information and as well as with errors, it is always required to segregate the image data for each acquisition. The remotely sensed raw image data suffer from a variety of errors and deficiencies caused by satellite motion and the sensor system. The distortions would shrink the accuracy of the information extracted and thereby reduce the utility of the data, if the data are not precisely corrected. Also in order to enable comparison of features in the images for studying the temporal changes, geometric correspondence among all the images under study must be guaranteed. The set of processes to be carried out for segregation/correction of deficiencies and the removal of flaws present in the data is called preprocessing as shown in Figure 2.1. For correcting the image/data, the type and the source of distortions must be determined. The errors are due to sensor performances (internal) or due to platform perturbations and imaging geometry (external). The errors are categorized as systematic or predictable, and non-systematic. The systematic errors are constant and can be predicted and measured in advance.

3 42 The mirror scan velocity, over-sampling by design, scan skew, sensor offsets etc. are the different types of systematic errors. But the non-systematic distortions are not constant because the result from variation in the performance of the scanning mechanisms, spacecraft velocity, spacecraft attitude, and sometimes even due to ground processing systems. The errors are not predictable, but can be estimated by the detailed analysis of the received data. VHRR payload scan mechanism and modulator VHRR payload data transmission Data reception and acquisition at ground station Preprocessing of raw data Data product generation VIS Image TIR Image WVP Image Figure 2.1 Flow diagram of Satellite Raw data preprocessing and image representation Image Representation is a process that generates the final image product in which the required features of the image are highlighted and presented in a form and format with specified accuracies which can be readily usable by the users for specific themes of their interest. The weather satellites provide imageries in various spectral bands of meteorological importance, with a synoptic coverage on a periodic, repetitive basis. The periodic image updates are ideally suited to study

4 43 weather-related, dynamic atmospheric process on different scales. The meteorological parameters derived from satellite imagery play a critical role in improved definition of initial and boundary conditions for weather prediction models. The impact of satellite data is phenomenal in certain areas of meteorological applications such as cyclone monitoring, short-range forecast, aviation forecasts and aerospace forecast, especially in the tropics. It is desirable for the meteorologists to have a processed image from which the meteorological parameters can be readily referred to. In this regard image representation plays a very important role in highlighting the embedded features of the image that can be easily interpreted by the user. 2.4 SATELLITE DATA RECEPTION, FORMATS AND STREAMS The data from the INSAT VHRR payload (the source of data used in the research), is received at Kbps (BPSK modulation) data rate after differential encoding to NRZ S format to resolve the phase ambiguity at the receiving end. The main function of the VHRR data modulator is to BPSK modulate the data stream containing the Meteorological payload data entering from the VHRR payload electronics packages in to the imagery signal received at the satellite. The data modulator received the imaging signals in the form of digital bit streams, BPSK modulates and up converts the frequency to extended C Band. The scan mirror takes about seconds to scan in E-W direction followed by about 0.19 seconds to turn around and step 223 micro-radiance. The data update interval to the computer memory from the data acquisition hardware is seconds. The one scan line has been organized into 6500 data blocks. Each data block is made up of 10 words of 10 bit each. The data acquisition hardware converts the 10 bit words into 16 bit words by padding 6 zeros on the MSB side to make it easier to process on a computer. Out of these 6500 data blocks, the 5500 data blocks (and hence words of 16 bit) are exclusively dedicated to the image data and the remaining data comprise the instrument health parameters and other auxiliary

5 44 data. The following methodology is developed to carryout pre-processing on the data set. 2.5 GEOMETRIC AND RADIOMETRIC CORRECTIONS OF SATELLITE DATA The earth scene captured by VHRR payload transmitted with a variety of errors and deficiencies caused by satellite motion and the sensor system. The data with the kinds of errors may mislead to critical calculation of weather and it obviously reduces the reliability of the data. Therefore correction or checking for errors should always be the first step in the raw data processing, which is also represented always as pre-processing. The following parameters influence the generation of errors in the data: Systematic or predictable Errors The systematic errors in the INSAT image data stream can be estimated and measured in advance. They are due to imaging geometry, sensor offsets, detector response characteristics, oversampling by design, etc. Many of them are measured even while the instrument is on the ground before it is put into orbit. The systematic errors are constant and can be predicted and measured in advance as follows: Line Drop Errors: The common challenge of the data reception technology is loss of data during the transmission of data after scan. The loss is more during past decades, but it is unavoidable for present advanced technologies. The partial loss of data is normally known as Line Drop Errors and corrected by filling the portion by artificially generated mean value of the pixels of the nearby pixels. Errors by sensors: The non-uniformity is a practical characteristic of the sensors. Since an array of sensors are used to record the details of radiation of all wavelengths in Electro Magnetic spectrum are not same for the same time. This is also called stripping error. Pre-estimation of this error can be compensated during raw data processing. The limitations of the sensors cause random noises in the data.

6 45 Errors by atmosphere: Un-determinant signal value changes occurring due to atmospheric scattering of the radiation, thus increasing the signal value can be corrected by degenerating the values by known mathematical calculations. Errors by instrument: The sync between data scanned and mechanical mirror movement may not be always constant, which is called servo error. Servo error occurs leading or lagging of angular movement magnitude from the theoretically predicted position. The next error by instrument is sampling error. A pixel per area is the ratio determining the aspect ratio of the image. But when the detector detects more than one pixel per area which causes aspect ratio mismatch which can be corrected by averaging the oversampled data to the appropriate aspect ratio Non- systematic or non-predictable Errors The non-systematic distortions are not constant because they result from variation in the performance of the scanning mechanisms, spacecraft velocity, and spacecraft attitude and sometimes even due to ground processing systems. The errors are not predictable, but can be estimated by the detailed analysis of the received data. The error is identified only by satellite position. The change in the orientation of the spacecraft body like distance between earth and spacecraft (altitude) and up or down (attitude) will introduce distortion in the image. 2.6 SATELLITE RAW DATA PREPROCESSING The image preprocessing refers to the initial processing of raw image to correct the geometric distortion, calibrate the data radiometrically and eliminate the noise present in the data. After eliminating noise the data are suitable for further representation into images usable for precise study Line drop correction Algorithm Occasional drop of data being received from the satellite causes loss of information in the images (VIS, TIR and WVP). Conventionally the error is detected by identifying the discontinuity in the feature of the image content. If such line drops occur while the instrument is scanning the uniform illumination area, there may not be appreciable/detectable discontinuity in the features, making it difficult to identify as well as to estimate the number of line drops.

7 46 The parameter SLOW SCAN LINE COUNT, derived from the telemetry keeps track of the line number being scanned in real-time. If there is one or more drops in the received scan line, there will be a corresponding jump in this parameter, which readily indicates the line drops. Such line drops are corrected by replacing the dropped lines with the average of pixel values corresponding to the previous and the following line. The input for the proposed algorithm is satellite raw data. The corrections are carried out for all available image formats (VIS, TIR and WVP) and for full mode scan data. The line drop correcting algorithm is used for all mode of scans by modifying the total line numbers in the algorithm i.e for full mode scan, 1092 for normal mode scan and 351 for sector mode scan Attitude Error Correction Algorithm The attitude (orientation) variation of the satellite during the imaging operation will cause the distortion in the image in any one of the directions of east, west, north or south. Detection of such errors by conventional methods again calls for comparison of received image with respect to a standard map or a presumably processed image. The attitude errors as measured by the onboard attitude sensors called as earth sensor (ES) are extracted by the House Keeping Telemetry that are readily available for the image processing schemes. The magnitude and the direction (sign) of errors are derived along with the corresponding image updated for every telemetry frame. The Figure 2.2 represents the four possible types of satellite attitude error images. The line drop correction algorithm and attitude error correction algorithm are given in the following steps. When the METSAT satellite has zero attitude error then the earth image scanned by the VHRR payload is perfect which is illustrated in the Figure 2.2(a). Due to on board single event upsets and mirror scan offset position change the roll error and pitch errors are occur in the METSAT satellite. The positive roll error is illustrated in the Figure 2.2 (b), the negative roll error is illustrated in the Figure 2.2 (c). The positive pitch error is illustrated in the Figure 2.2 (d), the negative pitch error is illustrated in the Figure 2.2 (e).

8 47 Figure 2.2 Satellite Attitude error images. a) Zero attitude error image b) Positive Roll error image c) Negative Roll error image d) Positive Pitch error image e) Negative Pitch error image

9 48 Algorithm for Line drop corrections Step 1: Step 2: Input satellite raw data. Initialize i=0 and j=1. where i is the first line number and j is the second line number. Step 3: Check condition (j-i) is not equal to1. If true go to step 4 or else go to step 8. Step 4: Initialize array rms_ld [ ], k=i-2 and m=i+2. Step 5: Check condition i>1&& i<1560 go to step 6. Else go to step 7. Step 6: rms_ld[ ]=root mean square for line arrays k=i-2, k=i-1, m=i+1 and m=i+2 and go to step 8. Step 7: Copy next line for i=1 and copy previous line for i= Step 8: Step 9: Increment i and j by one and check i>1560 go to step 9 else go to step 3. End Algorithm for Attitude error corrections Step 1: Step 2: Step 3: Satellite raw data from Data Acquisition Hardware system as the input. Derive the House Keeping (HK) parameter from satellite Telemetry. Derive the magnitude and the direction (sign) of errors along with the fraction of the corresponding image updated for every telemetry frame. Step 4: Attitude error corrected Satellite data output.

10 Results and Discussion The existing methods of detection of errors in the image are found only on the image data. In the proposed approach, the paradigm of detection of errors in the image is totally changed. The house keeping parameters derived from the telemetry precisely indicate the missing lines, satellite attitude errors for every updates of the image data. The developed approaches and concepts, which are implemented in the image processing system, have yielded break through enhancements in the following two areas. (i) Detection of anomalies in the VHRR image data. (ii) The image pre-processing and representation to generate ready to use image product. Ever since the work on the system taking a shape, the image data are subjected for analysis and so far more than ninety thousand images have been processed. Various anomalies observed earlier are studied in detail to understand their specific signatures shown by means of the deriver parameters and their pattern. Based on the detailed study, the defects in the images are classified into the following three categories. (i) Minor Upsets: The minor upsets are the small image shifts whose magnitudes are less than 50 pixels. These shifts are corrected by the proposed approaches. (ii) Major Upsets: The major upsets are the image shifts, with high magnitudes (greater than 50 pixels) which cannot be corrected by the proposed methodology. (iii) High SD_FSSE: The error is the occasion when the parameters SD_FSSE (standard Deviation of Fast Scan Servo Error) have crossed the empirical threshold of 200, indicating the high distortion in the image data.

11 50 The Table 2.1 provides the data set for preprocessing and representation including the number of defects detected in each of the categories. Table: 2.1 Data Sets for Preprocessing and Representation YEAR NUMBER OF MINOR MAJOR HIGH VALUE OF IMAGES UPSET UPSET SD_FSSE PROCESSED , , , , , , , JAN MAY 8, TOTAL 91, The proposed methodology serves two specific applications. (i) (ii) For detecting the possible defects in the portion of the image being scanned. For detecting the anomalies in the onboard system, which is essential for taking over corrective action, and safeguarding the onboard instrument Line Drop Error Corrections Conventionally the line drop error is detected by identifying the discontinuity in the features of the image count. If such line drops occur while the instrument is scanning the uniform illumination area the proposed algorithm will correct the error. The VIS band line drop error input image is illustrated in the Figure 2.3 (a). The VIS band image is a full mode scan image with total number of scan line is

12 After applying the line drop error correction algorithm to the VIS band line drop input image, the line drop is corrected as shown in the Figure 2.3 (d). The TIR band line drop error input image is illustrated in the Figure 2.3 (b). The TIR band images are a full mode scan image with total number of scan line of After applying the line drop error correction algorithm to the VIS band line drop input image, the line drop is corrected as shown in the Figure 2.3 (e). The WVP band line drop error input image is illustrated in the Figure 2.3 (c). The WVP band image is full mode scan image with total number of scan line of 1560.After applying the line drop error correction algorithm to the VIS band line drop input image, the line drop is corrected as shown in the Figure 2.3 (f). As discussed in the chapter section the line drop error correction algorithm line factor kept constant at The algorithm is well suited for separate single line drop corrections. The image has showed good resolution and suitable for further analysis. Subsequent multi-line drop error correction by the proposed algorithm may lead to reduction in image resolution, thus the kind of error occurring is very rare and it was not considered for this proposed scheme.

13 52 Figure 2.3 Input Images for Line Drop Error correction algorithm: a) VIS Band b) TIR Band c) WVP Band and corresponding error corrected images: d) VIS Band e) TIR Band f) WVP Band

14 Attitude Error Corrections There are four types of possible satellite attitude errors that affect the VHRR SCAN images. The Negative Pitch error image of Kalpana satellite is illustrated in the Figure 2.4. The negative pitch error moves the earth scene image towards west, so the east portion information is not captured due to the negative pitch error of the satellite body. The negative pitch error is corrected by applying the derived house keeping parameters from the satellite telemetry systems. By applying the proposed attitude error correction algorithm the negative pitch error corrected image is represented in the Figure 2.5. Both negative pitch error and positive pitch error cause longitude shift in the scanned area. Figure 2.4 Negative Pitch attitude error image The existing methods of preprocessing, namely line drop corrections, satellite attitude and altitude error corrections, sampling corrections, etc. are extensively dependant on comparison analysis of the image under process with either a standard map or a presumably corrected/standard image.

15 54 Figure 2.5 Negative Pitch attitude error corrected image In the research work, the House Keeping (HK) parameters derived from the telemetry precisely indicate the missing lines, satellite attitude errors etc. for every update of the image data. Hence there is no need for indirect estimation of such errors from the image data. 2.7 NEW SCHEME FOR THERMAL INFRA RED IMAGE REPRESENTATION Image representation refers to usable image product. The problem that complicates representation of satellite image is that the human eye is poor at discriminating the slight radiometric or spectral differences that may characterize the features, which may be important for the theme of study. The required features of the image are essentially needed to be highlighted and presented in a form and format with specified accuracies that can be readily usable by the users (application scientists) for specific themes of their interest. Representation of image involves some mathematical operations that are to be applied to the input digital counts received from the satellite to improve the visual appearance of the output image for better interpretability either by visual

16 55 examination or by subsequent digital analysis that may involve further computations Primary Considerations of All band Images To start with TIR images covering the Indian sub-continental region spread approximately from 65 E to 95 E longitude in east-west direction and 8 N to 38 N in north-south direction were taken. The TIR band spatial resolution is 8km x 8km. Considering the above facts the image size has been worked out to be 400 pixels x 400 pixels for each spectral band Enhancement of Images The sensors employed onboard imaging instruments have been designed to detect upwelling radiance levels ranging from low (from oceans) to very high (from clouds and ice). The given image covering a particular area is unlikely that the full dynamic range of the sensor will be used. Frequently the original image received will be dull and lacking in contrast. Sensitivity of the detectors, weak signal of the objects present on the target (earth surface), and similar reflectance of different objects and environmental conditions at the time of imaging operation contribute to low contrast in the original image data. It is necessary to suitably enhance the input data as part of image representation task, essentially considering the separability (contrast) between the interested features of the image while representing the image. Each of the conventional images in the respective spectral bands, namely Visible, Thermal Infrared and Water Vapour are needed to be generated as grayscale images, by mapping the corresponding digital counts to image pixel gray values through empirically proven mapping functions. G ( ) v fv DCv (2.1) G f DC ) (2.2) t t ( t Gw fw( DCw) (2.3)

17 56 where, Gv, G t, G are the gray values corresponding to the digital counts w DC, v, DCt DCw, mapped by the mapping functions v t w f, f, f respectively. The subscripts v, t and w indicate the spectral bands Visible, Thermal Infra-Red and Water Vapour respectively. The minimum and maximum values of the digital count DC min and DC max from the set of images taken at different times of the day at different seasons are empirically obtained. The minimum digital count DC min is assigned to extremely black gray value (pixel brightness value = zero), and the maximum digital count DC max is assigned to extremely white gray value (pixel brightness value = 255). The intermediate values are interpolated between black and white (0 and 255) as illustrated in the Figure Figure 2.6 Illustration of contrast enhancement method The interpolation is obtained by following a linear relationship as given below: DC out m ( DC ) c (2.4) inp Where DC inp and DC out are the input and output digital counts of the same pixel, m and c are the gradient and intercept respectively, which are obtained as follows: Let DC and DC max be the minimum and maximum gray values of min the input image (obtained empirically from the set of images) and the DC out min and

18 57 DC out max be the corresponding minimum and maximum gray values of the output image, so that DC = m DC min + c (2.5) out min DC = m DC max + c (2.6) out min By solving the (2.5) and (2.6), parameter m and c are obtained and presented in equation (2.7) and (2.8) as: m = DC DC out max max DC DC out min min (2.7) and c = ( DC out min max ( DC ) ( DC max out max DC. DC min ) min ) (2.8) Hence the equation (2.8) as: DC out = ( DC ( DC out max max. DC. DC out min min ) ) ( DC inp ) ( DC out min. DC ( DC max max ) ( DC. DC out max min ). DC min ) (2.9) For DCout-min = 0 and DCout-max = 255, equation (2.9) reduces to: DCinp DC min DC out *255 (2.10) DC DC max min The intermediate values are obtained using the equation (2.10). By using the obtained values, the conventional images are generated. Though all the features in the image are visible, there is poor contrast in the lower gray values, especially the land features and sea to land transition points etc.

19 Results and Discussion The efficiency of the proposed Algorithm is demonstrated using computer simulations. The piecewise linear stretch with static coefficients are implemented, which are suitable for images taken at any time of the day and any season of the year Primary considerations of full mode VIS, WV and TIR band images The sensors employed onboard imaging instruments have been designed to detect upwelling radiance levels ranging from low (from oceans) to very high (from clouds and ice). The image covering a particular area is unlikely that the full dynamic range of the sensor is used. The original image received will be dull and lacking constrast. Sensitivity of the detectors, weak signal of the objects present on the target and similar reflectance of different objects and environmental conditions at the time of scanning operation contribute to low contrast in the original image data. It is necessary to suitably enhance the input data as the part of image representation task, essentially considering the separability between the interested features of the image while representing the image. Primarily the required subset isolated (400 X 400 pixels) from the full scale images of VIS is shown in the Figure 2.7, the TIR is represented in the Figure 2.8 and WVP is illustrated in the Figure 2.9. bands settle down by the latitude and longitude data which determine from satellite attitude data as 5ºS to 45ºN and 60ºE to 105ºE.

20 59 Figure 2.7 Primary considerations of full mode VIS band image (inset required subset of VHRR image) Enhancement of visible and water vapour imageries After isolating the required segment of the image data from the full-scale scanned data, the image is further processed to contrast enhancement, latitude, longitude grid alignment and in mapping techniques to provide ready-to-use images for manual analysis like cyclone tracking and weather forecasting. It is important to note that the contrast enhancement technique has visible changes only for sunlit VIS images that are illustrated in the Figure The enhancement scheme has not shown visible changes on TIR images that are represented in the Figure. 2.11, but it assists for in further processing and normally WVP isolated data is not processed for enhancement since the data are used for comparison or merging techniques.

21 60 Figure 2.8 Primary considerations of full mode TIR band image (inset required subset of VHRR image)

22 61 Figure 2.9 Primary considerations of full mode WVP band image (inset required subset of corresponding image) Figure 2.10 The VIS band images a) before and b) after contrast enhancement

23 62 (a) (b) Figure 2.11 The corresponding TIR band images a) before and b) after contrast enhancement Generally for enhancement of images, histogram stretching method is generally used, where the enhancement parameters vary from image to image. In the proposed approach, the piece-wise linear stretch with static coefficients are employed, which are suitable for images taken at any time of the day and any season of the year. The proposed approach enables the study of temporal changes in the image features. The Kalpana -1 satellite VIS band image for contrast enhancement is illustrated in the Figure 2.10 (a). By applying the proposed contrast enhancement scheme the VIS band image is illustrated in the Figure 2.10 (b). The Kalpana -1 satellite TIR band image for contrast enhancement is illustrated in the Figure 2.11 (a). By applying the proposed contrast enhancement scheme the TIR band image is illustrated in the Figure 2.11 (b). The validation of the look-up table-based method is carried out by comparing the images with similar imagery from different spacecraft. The image from Metosat- 5 of EUMETSAT (European Meteorological satellite) was chosen for

24 63 comparison. The Metosat 5 satellite is a spin-stabilized spacecraft unlike the 3- axis stabilized attitude control as similar to INSAT spacecrafts. This spacecraft was operated in India region for some duration to support a project named Indian Ocean Experiments (INDOEX) and hence its images were covered in Indian region as those INSAT. The image from this spacecraft was an excellent choice for comparing the results of the present research. There is a change in the satellite and image scanning techonologies of INSAT and METEOSAT satellites, and the spectral bands used for generating the images by METEOSAT -5 instrument and KALPANA VHRR instruments were the same. The METEOSAT-5 was pointed at 82.5 East longitude that covers Indian subcontinent also. The image product Enhance IR being processed at METOC center, Rota Spain, found to be suitable comparison that used only the thermal infrared channel. The commercial operators like METEOSAT, where two levels of data reception from the satellites, the interval between the time when the onboard instrument scans and the time when the final image products reach the analysts is considerable. In the proposed approach, when the image products are constructed, the products will be ready for use within 2 minutes after the last required line is scanned by the onboard instrument. As far as the specific aerospace requirements are considered in the research work, the Indian sub-continent region, the image products are available to the analysts well before completion of the onboard scanning. 2.8 SUMMARY The pre-processing and representation of satellite raw data are presented and discussed. The line drop correction, Attitude correction algorithms are discussed and implemented. The methodology used for the image representation is also presented. In the proposed research the improved image processing schemes are developed for investigating the VHRR data preprocessing and Image representation. The chapter -3 deals with the colour computing algorithm for weather forecasting systems.