JPS58172769A - Demultiplex processing method for picture data - Google Patents

Abstract

PURPOSE:To perform demultiplexing at a high speed by compensating an interband offset while dividing it in the plural odds and shifts in the word-unit. CONSTITUTION:An address conversion part 19 performs address conversion regraring input data 17 according to an address conversion table and the number 18 of interband offset ends and stores the result in an intermediate buffer 20 or 21 through a buffer switching mechanism 23. A decision on how the data are chained in the buffer 20 and 21 is made on the basis of the value of the table 18. A last unit picture element, on the other hand, is stored separately in the intermediate buffers 22 and 20. The stored data of the buffer 22 is outputted during writing to the buffers 20 and 21. In this case, it is written in a computer memory 16 discretely on the basis of a preset interband offset table while the interband offset of every word is considered. Consequently, it is not necessary to refer to addresses in the bit-unit word, so the throughput is improved.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an input pre-processing device for satellite-1 image data, particularly image II! The present invention relates to providing a device suitable for performing data sorting and offset processing in real time.

In a computer system that receives an image taken by an artificial satellite on the ground and performs various corrections on the received data,
It is necessary to rearrange various image data based on the satellite imaging method, the method of transmitting the captured image to the ground, and the method of restoring it as riiJ image data after receiving it on the ground.

In particular, in all-digital systems, images are treated as a collection of large numbers of pixels (called pixels, usually treated as 8-bit or 1-byte data), and the system's challenge is how quickly they can be rearranged. This is an important factor in determining performance.

This rearrangement (S (series)/P (parallel) conversion) is called day multiplexing (Demulliplexin).
g), and as a conventional technology, it is executed as a preprocessing device when importing nine image data recorded on a high-density magnetic tape to an external storage device on the computer side, and it is high-speed without the need for sorting by a program on the computer side. It was already undergoing a transformation.

By the way, in conventional daytime plexing,
Band-to-band offset processing can be easily performed in a process called scatterwrite (a method of converting the cutting edge address to a computer storage device), so when rearranging image data, band-to-band offset processing is not considered at all. This is because the band-to-band offset between conventional sensors is regular and simple.

However, the band-to-band offset caused by the new sensor is
Because of the irregularity, if an attempt is made to perform interband offset correction during scatter write, address reference to the computer storage device becomes complicated, resulting in a reduction in processing speed.

Hereinafter, the problems encountered when the conventional day multiplexing device 1114c processes image data from a new sensor will be specifically explained.

FIG. 1 shows the overall configuration of the system.

The artificial satellite 1 rotates around the Earth 20 times and takes pictures of the Asahi Sun in a strip (generally using the Sukito method), serializes the data, and transmits it to the receiving antenna 3. (referred to as a downlink).

On the ground, high-density magnetic data 5 is recorded in real time via a satellite data receiving/recording device 4 coupled to an antenna. Next, in order to correct the 91g data, the recorded high-density magnetic tape 6 is reproduced and stored on a large-capacity disk 9 via the image input processing 8 in the image processing ffl device 7. Day multiplexing and interband offset correction are generally performed during single-channel input processing prior to storage on a large storage disk. After storing the large capacity disk, I
JIJJf* Image inside the processing device! An image correction process 10 will be performed.

The above flow will be explained in detail using a specific example, in which the problem caused by the difference in interband offset between the conventional sensor and the new sensor will be clearly explained.

Image taken by the satellite 1'J! If one scene is conceptually drawn, it will look like Figure 2 (a) 11. Now, suppose that the satellite's Senna instantly transmits 4 bands (Bl h B 2
hB3, Ba) 6 lines (Lt a Lx,
Lm a L4 ) can be imaged (in the future, the number of bands and lines that can be instantly imaged will increase).

With this pressure, the amount of data captured in one scan is 4x
It becomes 6Xn. In general, satellites do not have a large amount of memory, and the data is transmitted to the ground in real time. Since the downlink to the ground is usually serial, the satellite must perform P/S conversion based on certain rules.
This shows the transmission order of four pixels, and this order K
This means that the signals are repeatedly transmitted to the ground in the 13 formats (each band and line are mixed) as shown in FIG. 2(C) based on jIi.

10,000. Since image correction processing is performed on the ground, it is desirable that the data be arranged in the scanning direction for each band as shown below.

That is, it is necessary to rearrange the downlink format of FIG. 2(C) 13 as just described.

This procedure will be described later. It is also necessary to perform interband offset correction at the same time as the rearrangement, and these are collectively referred to as day multiplexing.

Figures 3 and 4 show the difference in interband offset between the conventional sensor and the new sensor.If the instantaneous captured image data 13 in Figure 83 corresponds to our position on the ground surface at the same time, then 1
4, and there are pixel shifts such as a, b, and c between each band.

The pixel deviation Vi of the conventional sensor was even and constant (9
11 For example, a = b = c = 2m elements (bytes)), so if the machine is a 16-bit machine, the interband offset correction is performed using the computer's own device IIL at the time of scatter write.
This could easily be done by shifting the (memory) address by a word (=16 bits).

However#! 4 As shown in FIG. 15, the interband offset of the new sensor is also accompanied by an l1iII accumulation of odd numbers (i.e. a=b=c=1 byte). Although the actual pixel shift is not constant but irregular, it is equivalent to a = b = C = 1 byte, so the explanation will be simplified for convenience of explanation. That is, when a=5 bytes, excluding the shift in word units, one word is 16 bits long at t', which is considered to be a=5-(2X2)=1 byte on the machine. If we consider a computer to be a machine with 1 word = 32 bits (which is the mainstream of current computers), a pixel is represented by 1 byte (8 bits), so if we exclude the deviation in word units, the θ byte deviation (""O+4* 8+
), 1 byte deviation (ae: 1, 5, 9... heat,
There are four possible cases: 2-byte deviation (" = 216.10.-), 3-byte deviation (a = 3a7a11, ...). In new sensors, interband offsets like these exist, but this When processed using a conventional day multiplexing device, the result is as shown in Figure 5 (Figure 5 shows 4
1J! (It is assumed that processing is performed in units of ij elements.) i.e. 7
If an attempt is made to process a 71-bit offset during scatter write, if the computer memory 16 is 32 bits, there will be four cases of byte-by-byte offsets, and word-by-word address reference will be impossible. Byte-by-byte address reference can be easily performed by adding dedicated hardware, but address reference takes a corresponding amount of time.

FIG. 6 shows a time chart of the operation in this case. Particularly problematic is the address reference of the computer memory, and no matter how fast day multiplexing is made, it is impossible to absorb the 1t-n machine memory output waiting time of Δt.

The computer memory input time for l scans is ΔT(Δ1XN
) minutes are spent referencing byte addresses.

In reality, the received data is sent unilaterally to the computer 9aK, so the idle time of Δt as shown in the figure is processed by slowing down the input speed of the high-density magnetic tape in accordance with Δt. The time required for this Δt depends on the dedicated hardware algorithm, but it is several times longer than when referring to an address in word units.

Therefore, even if the conventional demultiplexing device is used to deal with the band-to-band offset of the new sensor, no improvement in processing speed can be expected.

The purpose of the present invention is to maintain at least the conventional throughput by paying attention to the word unit on the computer side in performing interband offset correction, which is a processing bottleneck in conventional day multiplexing devices, without referring to byte addresses. The purpose of the present invention is to provide a new daytime multiplexing processing method.

The present invention shows that the interband offset by the new sensor is
The problem is that since a pixel is expressed in one byte, it does not correspond to a word unit on the computer side.

Therefore, a fractional deviation in byte units is absorbed when pixels are rearranged, and the remaining deviation in word units is absorbed at the time of output to the computer memory. In other words, by dividing the interband offset correction into fractional (byte unit deviation) and word unit deviation,
It is designed to perform day multiplexing at high speed.

FIG. 7 shows the configuration of a device that corrects fractional deviations in byte units when rearranging first-stage data, that is, the first stage 24 of the band-to-band offset correction division type daymultiplexing device. The front section consists of the following components:

(1) Band-to-band offset trend table (18) that stores the fraction (byte shift) of the band-to-band offset of input data (for example, 0, 1.2 on a 32-bit machine).
(4 cases of 3) (2) An address conversion unit (19) having a function of converting the input data to the address of the transfer destination based on the built-in address conversion table, including the offset fraction for each pixel.

It is usually composed of microprograms, etc.

(3) Intermediate buffers 1.2 and 3 (20, 21, 22) that temporarily store data after address conversion.

(4) Alternate buffer 7% between intermediate buffers 1, 2 and 3
+Ij go woh y 77 switch ale4111 (23)
.

The operation will be explained below.

FIG. 7 shows 14 instantaneous captured images f (P + @ PI*
Demultiplexing is performed in units of 1 a1'+*m*P++s).

1) In advance, from the evening link format 17, (P+, Lt, B weight) (P Otori 111
# Lla B1 ) (PI4 Tomie L1 #
Hl) (Pt*s a L1+ Bt)+(Pt,
An address translation table arranged in such order as Lt, Bm), . . . is created and set in the address translation section 19.

2) First, calculate the fraction of the interband offset according to the word length of the computer and set it in the interband offset fraction table 18 configured in ROM or the like.

3) The downlink format data 17 received and recorded from the high-density magnetic tape is transferred to the address conversion unit 19.
Load via.

4) The address conversion unit 19 converts the input data 17 to the intermediate buffer 120 or 221, which is the empty buffer at that time, for each pixel through the buffer switching mechanism 23 according to the address conversion table and the interband offset fraction table 18. Store in. Assuming that the interband offset follows FIG. 4, the data of band 1 is stored in the intermediate buffer 12'0 for four pixels, which is a unit of processing. Band 2 data is 3 pixels in intermediate buffer 1 20,
Intermediate buffer 2211CI pixels are stored. Similarly, bands 3 and 4 are stored across intermediate buffer k.

The value of the interband offset fraction table 18 determines how the band spans. At this point, each pixel is in a state where four pixels are arranged in the same band and on the same line. However, at the time of initial rearrangement, only four pieces of data of band l are lined up, and for the data of other bands, dummy data is lined up at the beginning by a fraction of the offset. If the capacity of the intermediate buffer is increased, the stored instant m1lli
This is determined by factors such as device size, data input/output speed, and buffer switching time.

5) On the other hand, the previous Ij111 elementary unit (Pt-4sP
t-maPs-msP+-1) has already been stored across the intermediate buffer 322 and IK in steps 3) and 4), and the data stored in the intermediate buffer 322 is transferred to the intermediate buffers 1 and 2 during writing. This will be output, but this will be controlled by the later stage of daymultiplexing, which will be described below.

FIG. 8 shows a part in which the interband offset is corrected in word units and outputted to the computer memory 16 by the latter part 23 of the interband offset division method, that is, the scatter write. The rear part consists of the following components.

O An interband offset table 26 that stores the shift in word units excluding a fraction of the interband offset of the input data. The operation of the write control unit 27 and the switching device @23° of the read source intermediate buffer at that time will be explained as follows, considering that it continues from the previous stage.

l) Read data via the buffer switching mechanism 23 from the intermediate buffer 322 that is not currently being written.

2) Based on the interband offset table 26 set in advance, the interband offset in word units is taken into consideration and written to the computer memory 16 in a discrete manner. Reference can be made to the computer memory in word units.

If the operations of the above-mentioned front-stage and rear-stage parts are performed continuously, a time chart for one scan as shown in FIG. 9 is obtained. Compared to the m6 diagram, the throughput is improved by 41 minutes because there is no need to refer to addresses in byte units.

Figure 10 shows how the band-related offset correction divided day multiplexing device shown in Figures @7 and @8 is actually used on a computer (C).
PU 28), inputs data on a high-density magnetic tape 6, performs day multiplexing and interband offset correction, and stores the data on a large-capacity disk 9. In the case of pot-density magnetic tape, the ff8IJ control method is different from the general magnetic tape for =+g machines, and data from the device is sent unilaterally and continuously to the computer, so it is difficult to transfer data from the computer memory to the large-capacity disk. Since the storage must be done in real time, the alternate buffer hand IQ (Nii machine memory l 29 and 230)
Is required.

The operation is indicated by dashed lines and will be explained below.

1) Data for one scan is read into the computer side memory 129 via the day multiplexing device 24, 25.

2) Store the data for the next scan in computer memory 2 in the same way.
30, and at the same time 111. Data in machine memory 1 is stored in large capacity disk 9.

3) In the procedure of 2), use computer memories 1 and 2 in reverse.

4) Repeat steps 2I and 3) in sequence until the required scan is completed.

The present invention is extremely flexible for processing daymultiplexing with interband offset correction for all strains. That is, regardless of the difference in the interband offsets of satellites (of all types), and regardless of whether the word unit on the computer side is 16 bits or 32 bits, the values in the interband offset (fraction) table are changed. This is because demarteplexing can be performed extremely easily.

According to the present invention, as is clear from FIGS. 6 and 9, 1
When data processing for a new sensor is performed using a conventional demarte brexing device, the address of the computer side memory during interband offset correction due to the difference between the word unit on the computer side and the size (byte unit) of the image data. Addressing in 1-byte units is now required for reference. As a result, the time required for byte addressing of ΔT per scan was added, and the throughput of the entire system was reduced. According to the band-to-band offset correction split type dmal deplexing device described in the present invention, ΔT in FIG.
Since the transfer time 31 for a 4-pixel rate is infinitesimal compared to the input time for one scan and can be ignored, it is the same as simply inputting the received data without performing daymultiplexing. The apparent daymultiplexing time is 0, and the effect is large.

[Brief explanation of the drawing]

Figure m1 is an overall beak configuration diagram of the ground station system that is the subject of the present invention, Figure m2 is a diagram explaining the downlink of ItJII I & Deday from the satellite, Figures 3 and 4 are:
Figure 5 shows the differences between the seven sets of bands between the conventional sensor and the new sensor, and Figure 5 shows the results (problems) when the conventional demultiplexing device is applied to the case of Figure 4.
Figure 6 shows the time chart of the operation of the m5 diagram, Figure 7,
FIG. 8 shows the front-stage and rear-stage parts of the band-to-band offset correction division type demultiplexing device according to the present invention.
The figure shows a time chart of the operation when the present invention is applied to demultiplexing of a new sensor, and FIG. 1O shows a conventional example of the system when the present invention is actually applied.

Claims (1)

[Claims] 1. Parallel/serial conversion of multiple nodes of data captured during navigation of the satellite is transmitted to the ground, where the received serial data is subjected to day multiplexing to create an image. In the reproduction method, the received serial data is successively received, and the positional deviation obtained by imaging the same position in different wavelength bands is corrected by 1) Fractional bits less than 1 bit are corrected, and 1 word of the computer that processes the image is corrected. A day multiplexing processing method for IjJJJ111 data, which is characterized by performing day multiplexing, loading the data into the computer, and performing image reproduction processing.