CN119228882A

CN119228882A - Space debris centroid positioning method and system based on lightweight super-resolution

Info

Publication number: CN119228882A
Application number: CN202411765104.1A
Authority: CN
Inventors: 禹霁阳; 黄丹; 卢玲
Original assignee: Ordnance Science and Research Academy of China; Beijing Institute of Spacecraft System Engineering
Current assignee: Ordnance Science and Research Academy of China; Beijing Institute of Spacecraft System Engineering
Priority date: 2024-12-04
Filing date: 2024-12-04
Publication date: 2024-12-31
Anticipated expiration: 2044-12-04
Also published as: CN119228882B

Abstract

The invention discloses a space debris centroid positioning method and system based on light super-resolution, which relate to the technical field of centroid positioning and specifically comprise the steps of obtaining data to be processed; the method comprises the steps of inputting data to be processed into a deep convolution network for calculation, inputting calculation data into a first channel and a second channel respectively for three-level up-sampling processing, carrying out average summation on up-sampling data output by the first channel and the second channel to obtain average data, inputting the average data into the deep convolution network to obtain super-resolution data, and carrying out centroid solution on the super-resolution data to obtain positioning data. The invention designs a lightweight super-resolution depth convolution network Tpf-Net, which is characterized in that an up-down sampling dual-channel parallel fusion super-resolution network is established, a five-layer pyramid convolution network is combined, and 8 times super-resolution data is obtained through three times of up-sampling, so that high-precision positioning calculation can be realized.

Description

Space debris centroid positioning method and system based on light super-resolution

Technical Field

The invention relates to the technical field of centroid positioning, in particular to a method and a system for positioning a centroid of a space debris based on light super-resolution.

Background

With the increasing global space activity, the problem of space debris has become a key factor affecting spacecraft safety and space exploration sustainability. Space debris is mainly composed of spent satellites, rocket debris, and debris from collisions with small objects, which travel at high speeds in earth orbit, pose a serious threat to normal functioning satellites and other space assets. Once a collision occurs, expensive space assets may be lost, even threatening the life safety of astronauts. Therefore, effectively monitoring and managing space debris is an important task to ensure long-term safety of space environments.

Conventional imaging techniques are limited in resolution and it is difficult to accurately identify and locate these differently sized, morphologically distinct spatial patches. The super-resolution imaging technology can break through the resolution limit of a physical optical system through an advanced image processing algorithm, and provide a clearer and finer image than the original acquired data. This capability is critical for accurate tracking and analysis of spatial debris.

However, imaging of spatial debris faces a unique set of challenges:

Distance-the distance between the space debris and the viewing device is typically very far, resulting in a relatively small effective area for the target to reflect.

The reflectivity is low-many debris surfaces may not have good reflective properties, making their contrast in the image very low.

The background interference is large, and space background is complex and changeable, including stars, planets and other artificial or natural celestial bodies, can be used as interference sources.

Effective pixels are few-particularly for small patches, the number of effective pixels available under conventional imaging conditions is extremely limited, which directly affects centroid positioning accuracy.

The signal-to-noise ratio is low, and because the signal strength is weak and a large amount of noise exists, the realization of the centroid positioning with high precision becomes very difficult.

Therefore, how to improve the super-resolution processing efficiency and the imaging quality in the process of locating the center of mass of the space debris is a problem to be solved by those skilled in the art.

Disclosure of Invention

In view of the above, the present invention provides a method and a system for locating the centroid of a space debris based on light super resolution, which overcome the above-mentioned drawbacks.

In order to achieve the above purpose, the present invention adopts the following technical scheme:

a space debris centroid positioning method based on light super-resolution comprises the following specific steps:

Acquiring data to be processed;

The data to be processed is input into a deep convolution network for calculation, the calculated data is respectively input into a first channel and a second channel for three-level up-sampling processing, and the up-sampling data output by the first channel and the second channel are subjected to mean summation to obtain mean data;

Inputting the mean value data into the deep convolution network to obtain super-resolution data;

and carrying out centroid solving on the super-resolution data to obtain positioning data.

Optionally, the specific step of performing three-level upsampling processing in the first channel is to sequentially perform three upsampling on the calculated data to obtain first upsampled data, second upsampled data and third upsampled data respectively.

The method comprises the specific steps of carrying out three-level up-sampling processing on a second channel, wherein the specific steps of carrying out the down-sampling processing on the calculated data to obtain down-sampled data, carrying out the up-sampling processing on the down-sampled data to obtain fourth up-sampled data, carrying out average summation on the fourth up-sampled data and the first up-sampled data in the same dimension to obtain first average value data, carrying out the up-sampling processing on the first average value data to obtain fifth up-sampled data, carrying out average summation on the fifth up-sampled data and the second up-sampled data to obtain second average value data, and carrying out the up-sampling processing on the second average value data to obtain sixth up-sampled data, carrying out average summation on the sixth up-sampled data and the third up-sampled data to obtain third average value data.

Optionally, the data needs to be computed over the deep convolutional network before either the upsampling or downsampling process is performed.

Optionally, the deep convolutional network employs a pyramid convolutional network.

A lightweight super-resolution based spatial debris centroid positioning system comprising:

the data acquisition module is used for acquiring data to be processed;

The multi-stage up-sampling module is used for inputting the data to be processed into a deep convolutional network for calculation, inputting calculated data into a first channel and a second channel respectively for three-stage up-sampling processing, and carrying out average summation on the up-sampled data output by the first channel and the second channel to obtain average data;

the super-resolution data acquisition module is used for inputting the mean value data into the deep convolution network to acquire super-resolution data;

And the position acquisition module is used for carrying out centroid solving on the super-resolution data to obtain positioning data.

Optionally, the multi-stage upsampling module includes a first computing unit, a first channel processing unit, and a second channel processing unit;

The first calculation unit is used for calculating the data to be processed by utilizing the deep convolution network to obtain calculation data;

The first channel processing unit is used for carrying out three-level up-sampling processing on the calculated data to obtain first channel up-sampling data, wherein the first channel up-sampling data comprises first up-sampling data, second up-sampling data and third up-sampling data;

And the second channel processing unit is used for performing three-level up-sampling processing after performing down-sampling on the calculated data to obtain second channel up-sampling data, and performing average summation on the first channel up-sampling data and the second channel up-sampling data to obtain the average data.

Optionally, the second channel processing unit includes:

A downsampling subunit, configured to perform downsampling processing on the calculated data to obtain downsampled data;

a first up-sampling subunit, configured to perform up-sampling processing on the down-sampling data to obtain fourth up-sampling data, where the fourth up-sampling data is in the same dimension as the first up-sampling data;

a first average summing subunit, configured to average sum the fourth upsampled data with the first upsampled data to obtain first average data;

a second up-sampling subunit, configured to perform up-sampling processing on the first mean value data to obtain fifth up-sampling data;

A second average summing subunit, configured to average sum the fifth upsampled data with the second upsampled data to obtain second average data;

A third up-sampling subunit, configured to perform up-sampling processing on the second mean value data to obtain sixth up-sampling data;

a third average summing subunit, configured to average sum the sixth upsampled data with the third upsampled data to obtain third average data;

and the calculating subunit is used for calculating the data before up-sampling or down-sampling processing by using the deep convolution network.

Compared with the prior art, the invention provides a space debris centroid positioning method and system based on light super-resolution, designs a light super-resolution depth convolution network Tpf-Net, and can realize high-precision positioning calculation by establishing an up-down sampling double-channel parallel fusion super-resolution network, combining a five-layer pyramid convolution network, and obtaining 8 times super-resolution data through three times up sampling. The calculation processing process only comprises three kinds of calculation including up-down sampling, pyramid convolution and mean summation, the whole framework is simple and suitable for parallelization, each calculation process can adopt a mode of parallelization of a plurality of calculation operators or parallelization of a plurality of convolutions, and the data processing speed is greatly improved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present invention, and that other drawings can be obtained according to the provided drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic flow chart of the method provided by the invention;

FIG. 2 is a schematic diagram of a light-weight super-resolution deep convolutional network architecture provided by the invention;

FIG. 3 is a schematic diagram of a pyramid convolution structure provided by the present invention;

Fig. 4 (a) is a schematic diagram of low signal-to-noise ratio and tailing-free data in a test data set provided by the invention, fig. 4 (b) is a schematic diagram of high signal-to-noise ratio and tailing-free data in a test data set provided by the invention, fig. 4 (c) is a schematic diagram of low signal-to-noise ratio and tailing-free data in a test data set provided by the invention, and fig. 4 (d) is a schematic diagram of high signal-to-noise ratio and tailing-free data in a test data set provided by the invention;

fig. 5 is a schematic diagram of centroid positioning accuracy of different algorithms under different signal-to-noise ratios provided by the present invention.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

The invention discloses a space debris centroid positioning method and system based on light super-resolution, which can realize generation of 8 times super-resolution data through three times up-sampling by constructing a super-resolution network comprising up-down sampling double-channel parallel fusion and combining with a five-layer pyramid convolution network. On the basis, centroid solving is carried out on super-resolution data, so that the goal of high-precision positioning and solving is achieved, and as shown in fig. 1, the specific steps are as follows:

Step 1, obtaining data to be processed;

Step 2, inputting data to be processed into a deep convolutional network for calculation, inputting calculated data into a first channel and a second channel respectively for three-level up-sampling treatment, and carrying out average summation on up-sampling data output by the first channel and the second channel to obtain average data;

Step3, inputting the mean value data into a deep convolution network to obtain super-resolution data;

and 4, carrying out centroid solving on the super-resolution data to obtain positioning data.

Further, the method is realized based on a lightweight super-resolution deep convolutional network (Tpf-Net), and the structure of the method is shown in fig. 2, and the method comprises a deep convolutional network layer (conv), an upsampling layer and an accumulation summation layer. The whole method for realizing Tpf-Net is roughly divided into 3 steps, namely, inputting no more thanImage of (a)Performing deep convolutional network (conv) calculation, and simultaneously upsampling and downsampling to obtain respectivelyIs the first up-sampled data of (1)AndDownsampled data of (a)First up-sampled dataAndDownsampled data of (a)Respectively entering a deep convolution network (conv) for calculation, and carrying out average summation calculation on the same-dimension data output by the two channels to obtainThen up-sampling is carried out, and finally up-sampling is achievedWhen the mean sum resultsObtaining final super-resolution data through a deep convolution network (conv)。

Further, the first up-sampled dataAnd downsampling dataThe calculation process of (1) is as follows:

(1);

Wherein, AndThe up-sampling and down-sampling functions respectively,To image pairProceeding withIs calculated by the pyramid convolution of (a), in this embodiment,32.

In one embodiment, the specific step of performing three-level upsampling processing in the first channel is to sequentially perform three upsampling on the calculated data to obtain first upsampled data, second upsampled data, and third upsampled data, respectively.

In an embodiment, the specific steps of performing three-level upsampling processing in the second channel are that the first-level upsampling is that the calculated data is subjected to downsampling processing to obtain downsampled data, the downsampled data is subjected to upsampling processing to obtain fourth upsampled data, average summation of the fourth upsampled data and the first upsampled data in the same dimension is performed to obtain first average data, the second-level upsampling is that the first average data is subjected to upsampling processing to obtain fifth upsampled data, average summation of the fifth upsampled data and the second upsampled data is performed to obtain second average data, and the third-level upsampling is that the second average data is subjected to upsampling processing to obtain sixth upsampled data, average summation of the sixth upsampled data and the third upsampled data is performed to obtain third average data.

In one embodiment, before upsampling or downsampling, the data is calculated through a deep convolutional network, and the calculation process is as follows:

(2);

In the formula, For the first upsampled data; For the second upsampled data; Is a pyramid convolution network; upsampling data for a first time for a second channel; is downsampled data; For the third upsampled data; up-sampling the data for a fourth up-sample; For the fifth upsampled data; Is the first mean value data; Upsampling the data for a sixth upsampling; Is the second mean value data;

the average summation process of the up-sampling results of the two channels is as follows:

(3);

In the formula, Is the third mean value data; representing averaging.

The super-resolution data is:

(4);

in one embodiment, the deep convolutional network employs a pyramid convolutional network.

Further, the deep convolutional network (conv) is a five-layer pyramid convolutional network (Pyramid convolution,) The architecture is as shown in FIG. 3, and the same convolution computation structure is adopted for different input pixel sizes, specifically for the inputFirst 16 images are madeConvolution kernel computation, whereObtaining 16Subgraph, and then go through 32 16 groupsConvolving to obtain 32Subgraph, then pass through 16 32 groupsConvolving to obtain 16Subgraph, finally, through 16 groupsConvolution to obtain convolvedAn image. And a convolution network of pyramid and inverted pyramid cascade is adopted to calculate the up-down sampled image, so that the up-down sampled image is accurately restored.

Based on the above description, it is not difficult to see that Tpf-Net is particularly suitable for digital logic acceleration calculations, mainly:

I. the calculation amount is small, the pyramid convolution process is just a five-layer convolution, and the calculation amount is that The image convolution reaches the highest and only needs to be less thanWeight [ ]Convolution), the overall network weight does not exceedCompared with the traditional super-resolution network, the method has the advantages that the method is reduced by 90 percent;

The structure is fixed, the whole calculation processing process only comprises three calculation steps of up-down sampling, pyramid convolution and average summation, the whole framework is simple and suitable for parallelism, and each calculation process can adopt a mode of paralleling a plurality of calculation operators or paralleling a plurality of convolutions to improve the instantaneity;

And III, the parallelism is high, 8 pyramid convolution calculations are independent, the intermediate variable only has front and rear time sequence input and output, the precedence relationship of the convolution calculations in each pyramid convolution is not influenced, and the parallelism is improved.

In one embodiment, the method is verified by the following steps:

Constructing a data set, namely making a data set for super-resolution calculation by using the real space debris monitoring public data of the foundation wide-angle camera array (GWAC), wherein the data set comprises 800 training images and 200 test image pairs. Training the data by adopting a Tpf-Net model, and testing a test data set. Meanwhile, for test comparison, an improved least square estimation algorithm, a point diffusion model algorithm and an iterative adaptive window algorithm are selected as comparison algorithms of the embodiment, and the same test data set is tested. As three common mainstream centroid positioning algorithms, the centroid positioning accuracy performance is representative to a certain extent.

In the embodiment, three types of Tpf-Net are designed, and the test is performed according to the difference of the convolution kernel sizes. Wherein, The convolution kernel version is the Tpf-Net base version,The convolution kernel version is Tpf-Net +,The convolution kernel version is Tpf-net++. The convolution kernel size employs three versions for testing the impact of convolution kernel size on network effects.

Training configuration, namely processing the data sets of 1000 original images, and obtaining 8 times low-resolution data by adopting a sampling filtering method. Training 800 pairs of training images by using Pytorch frames, training a platform by using GPU 3060Ti of NVIDIA, and a processor by using I7-10700.

As shown in Table 1, the data set experimental results are compared with the previous 3-type space debris and star target centroid positioning method, and meanwhile, the accuracy and the calculated amount are compared with the previous 3-type super-resolution algorithm which is commonly used, and the algorithm comprises a traditional calculation model and a CNN-based algorithm. The traditional algorithm is mainly divided into two types based on a point spread function and interpolation filtering. Wherein EIWA achieves a positioning accuracy of up to 0.0056 pixels. All super-resolution results are compared through the tests of PSNR and SSIM.

Table 1 shows the quantitative comparison of Tpf-Net with other super-resolution algorithms and centroid location algorithms, and the resulting centroid location accuracy comparison.

Table 1 centroid positioning method under different signal-to-noise ratios and comparison based on super-resolution algorithm

The first three algorithms adopt a traditional calculation method to locate the mass center with high precision, so that the numerical values of PNSR and SSIM do not exist. The least square method is adopted to predict the centroid position, and the error is larger when the signal-to-noise ratio is low, and the pixel size is less than 0.2 pixel. The point spread function method can reach about 0.05 pixel at a lower signal-to-noise ratio, and the improvement of the signal-to-noise ratio is not high for the improvement of the positioning precision. The energy iteration-based adaptive window method reaches 0.02 pixel at low signal-to-noise ratio and reaches about 0.005 at high signal-to-noise ratio.

The middle three methods respectively adopt an iterative sum function reconstruction network, an enhanced eight-packet convolution network and a multi-model integration network to realize the super-resolution function, wherein EOctConv and MMSR are difficult to form training convergence when the signal to noise ratio is low, and the relationship with the network comprising multi-class architecture is larger. It can be seen that the previous algorithm based on CNN is difficult to achieve higher positioning accuracy, because the super-resolution intelligent method is mainly applicable to structured images or targets, and it is difficult to obtain better results for space fragment point target information.

The three network versions designed by the embodiment have similar centroid positioning accuracy obtained by training under low signal-to-noise ratio and have higher signal-to-noise ratioThe convolution kernel version accuracy is higher. But in view of comparison withThe weight of the convolution kernel is increased by 4 times, the centroid positioning precision is increased by less than 20%, most targets in the field of space debris are in a low signal-to-noise ratio state, and the efficiency of adopting the Tpf-Net version is higher.

The experimental results of the different types of data are compared, and for 200 pieces of test image data, 50 pieces of data can be divided into four types, wherein the 50 pieces of data comprise low signal-to-noise ratio without tail, high signal-to-noise ratio without tail, low signal-to-noise ratio with tail and high signal-to-noise ratio with tail, and the four types of data are shown in the figures 4 (a) -4 (d). The four types of data have different distribution characteristics, and meanwhile, whether the data are trailing or not can greatly influence the mass center positioning accuracy, so that the results of the four types of data are used for comparison and analysis, and are shown in table 2. Wherein a smear of less than 4 pixels is considered to be no smear, and the smear length of the smear data is distributed over 5-10 pixels, where the smear is not considered to be more than 10 pixels.

Where a low snr with no tail is the same as for snr <6, there are already test results in table 1. The test results of the smear data in table 2 are reduced by an order of magnitude compared to the accuracy of the no smear results, and the smear data positioning accuracy is reduced to 0.1 pixel for low signal to noise ratios. The method also shows that for the tailing data with low signal to noise ratio, the whole object is broken due to tailing, so that the object body is difficult to be well identified and super-resolution prediction is carried out in the super-resolution calculation process, and the centroid positioning precision is low. The morphological pretreatment can effectively improve the accuracy of tail fracture with low signal-to-noise ratio on the positioning of the mass center, and a single image is better in result through pyramid convolution of up-and-down sampling. The demarcation threshold between high and low signal to noise ratios is 6.

Table 2 comparison of centroid location method of different types of test data based on progressive preprocessing super-resolution algorithm

Under the condition of trailing, tpf-Net++ has higher centroid positioning accuracy compared with Tpf-Net, because the increase of the convolution kernel size can improve the perceptibility of trailing data when the trailing length is 5-10, thereby generating better super-resolution images and improving the positioning accuracy.

And comparing the data experimental results of different signal to noise ratios, namely testing the data of different signal to noise ratios in the data set to obtain centroid positioning accuracy result comparison under different algorithms, as shown in figure 5.

Overall, the increase of positioning accuracy tends to be stable when the signal-to-noise ratio of various algorithms is above 4. On one hand, because the dark and weak targets with larger influence on the precision have better algorithm adaptability when the signal to noise ratio is higher, the positioning precision can be synchronously improved according to the increase of the signal to noise ratio, and on the other hand, because most algorithms are that data with the signal to noise ratio below 3 are greatly influenced by noise, the target information is difficult to provide assistance for the precision. Among them, the least square method is a simple least square estimation of the target area and the periphery, and thus the overall accuracy is poor. The point spread function method is difficult to reach high-precision indexes when the signal-to-noise ratio is low. The EIWA algorithm obtains higher precision due to the window function iteration method. Compared with the lightweight super-resolution network Tpf-Net designed in the text, the precision difference is not large at high signal-to-noise ratio, and compared with EIWA, the precision of the Tpf-Net is improved by 40% at low signal-to-noise ratio.

The embodiment designs a lightweight super-resolution depth convolution network Tpf-Net, and 8 times super-resolution data is obtained through three times up-sampling by establishing an up-down sampling double-channel parallel fusion super-resolution network and combining a five-layer pyramid convolution network. And carrying out centroid solving on the obtained super-resolution data to realize high-precision positioning and solving. Compared with the prior algorithm, the calculation process is mainly characterized by up-down sampling calculation and pyramid convolution calculation before and after sampling, and the weight is only smaller than that of the prior algorithmThe architecture is simple, is suitable for digital logic acceleration, and is suitable for on-orbit lightweight computing application. During the course of the experiment, coverage tests were performed on data of different types, different signal to noise ratios in GWAC. Compared with the prior algorithm, the centroid positioning precision is improved by more than 40% when the signal-to-noise ratio is low, the positioning precision is improved by 10 times when the signal-to-noise ratio is high, and the effectiveness of Tpf-Net is illustrated by comparing the precision with the precision of the prior algorithm.

On the other hand, the embodiment also discloses a space debris centroid positioning system based on light super-resolution, and the method disclosed by the embodiment is applied and comprises the following steps:

the data acquisition module is used for acquiring data to be processed;

The multi-stage up-sampling module is used for inputting data to be processed into the deep convolutional network for calculation, inputting calculation data into the first channel and the second channel respectively for three-stage up-sampling processing, and carrying out average summation on the up-sampling data output by the first channel and the second channel to obtain average data;

The super-resolution result acquisition module inputs the mean value data into the deep convolution network to acquire super-resolution data;

In an embodiment, the multi-level upsampling module includes a first computing unit, a first channel processing unit, and a second channel processing unit;

The first calculation unit is used for calculating the data to be processed by using a deep convolution network to obtain calculated data;

and the second channel processing unit is used for carrying out three-level up-sampling processing after carrying out down-sampling on the calculated data to obtain second channel up-sampling data, and carrying out average summation on the first channel up-sampling data and the second channel up-sampling data to obtain average data.

In an embodiment, the second channel processing unit comprises:

the first up-sampling subunit is used for up-sampling the down-sampling data to obtain fourth up-sampling data, and the fourth up-sampling data and the first up-sampling data are in the same dimension;

The first average summing subunit is used for carrying out average summation on the fourth up-sampling data and the first up-sampling data to obtain first average data;

The second up-sampling subunit is used for up-sampling the first mean value data to obtain fifth up-sampling data;

The second average summing subunit is used for carrying out average summation on the fifth up-sampling data and the second up-sampling data to obtain second average data;

The third average summing subunit is configured to average sum the sixth up-sampled data and the third up-sampled data to obtain third average data;

And the second calculating subunit is used for calculating the data before up-sampling or down-sampling processing by using the deep convolution network.

In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A space debris centroid positioning method based on light super-resolution is characterized by comprising the following specific steps:

Acquiring data to be processed;

2. The method for locating the centroid of the space debris based on light super-resolution according to claim 1, wherein the specific step of performing three-level upsampling processing in the first channel is to sequentially perform three upsampling on the calculated data to obtain first upsampled data, second upsampled data and third upsampled data, respectively.

3. The method for positioning the centroid of the space debris based on light super-resolution according to claim 2 is characterized by comprising the specific steps of performing three-level upsampling in the second channel, wherein the first-level upsampling is to perform downsampling on the calculated data to obtain downsampled data, the downsampled data is to perform upsampling on the downsampled data to obtain fourth upsampled data, and average summation is performed on the fourth upsampled data and the first upsampled data in the same dimension to obtain first average data, the second-level upsampling is to perform upsampling on the first average data to obtain fifth upsampled data, average summation is performed on the fifth upsampled data and the second upsampled data to obtain second average data, and the third-level upsampling is to perform upsampling on the second average data to obtain sixth upsampled data, average summation is performed on the sixth upsampled data and the third upsampled data to obtain third average data.

4. A method of locating the centroid of a spatial debris based on light super resolution as claimed in claim 3, wherein the data is calculated by the depth convolution network before upsampling or downsampling.

5. The method for locating the centroid of a space debris based on light super resolution according to claim 1, wherein the depth convolution network is a pyramid convolution network.

6. A lightweight super-resolution based spatial debris centroid positioning system, comprising:

the data acquisition module is used for acquiring data to be processed;

7. The light-weight super-resolution based spatial debris centroid positioning system of claim 6, wherein the multi-stage upsampling module comprises a first computing unit, a first channel processing unit, and a second channel processing unit;

8. The lightweight super-resolution based spatial debris centroid positioning system as set forth in claim 7, wherein said second channel processing unit comprises: