CN109302614B - Video compression method based on third-order tensor self-coding network - Google Patents
Video compression method based on third-order tensor self-coding network Download PDFInfo
- Publication number
- CN109302614B CN109302614B CN201811168316.6A CN201811168316A CN109302614B CN 109302614 B CN109302614 B CN 109302614B CN 201811168316 A CN201811168316 A CN 201811168316A CN 109302614 B CN109302614 B CN 109302614B
- Authority
- CN
- China
- Prior art keywords
- network
- tensor
- video
- integer
- equal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/42—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals characterised by implementation details or hardware specially adapted for video compression or decompression, e.g. dedicated software implementation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/85—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using pre-processing or post-processing specially adapted for video compression
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention provides a video compression method based on a third-order tensor self-coding network. The self-coding network is widely used in image compression, but a large number of parameters need to be stored and a large number of pictures need to be used for training the network, so that a third-order tensor is used for replacing full connection parameters between layers in the self-coding network, a self-coding mechanism and a back propagation method are used for carrying out iterative solution on the parameters in the network to achieve convergence, a convergence result is coded, and finally a compressed video is obtained.
Description
Technical Field
The invention relates to a video compression method based on a third-order tensor self-coding network, and belongs to the technical field of video compression.
Background
In recent years, video technology has gained widespread use and growth. By 2020, 80% of internet traffic is video traffic. Uncompressed video, however, occupies a large amount of storage space. At present, there are mainly h.264 and h.265 video stream coding methods and neural network based methods. Based on H.264 and H.265 video stream coding methods, the compression rate is high, but the decompression speed is low; the neural network-based method has relatively high compression efficiency, but needs a large number of images for training, has more network parameters, and needs to occupy more memory space of terminal equipment.
Disclosure of Invention
The invention provides a video compression method based on a third-order tensor self-coding network, which aims to solve the problems in the prior art and has high compression rate and high decompression speed.
In order to achieve the purpose, the technical scheme provided by the invention is as follows: a video compression method based on a third-order tensor self-coding network comprises the following steps:
firstly, preprocessing a target video;
setting full connection parameters between layers in a target video self-coding network as third-order tensors and setting iteration ending conditions;
step three, forward transmission of a video network is carried out;
step four, judging whether iteration is terminated, if so, skipping to step six, and outputting a core tensor and decoding network parameters; otherwise, continuing the step five;
fifthly, performing back propagation of the video network;
sixthly, coding and compressing the core tensor and the decoding network;
and step seven, outputting the compressed video.
The technical scheme is further designed as follows: and in the second step, parameters of the network and iteration ending conditions are set according to the required compression ratio and the peak signal-to-noise ratio.
The third step comprises the following specific steps: performing modulo-1 modulo-2 modulo-3 multiplication by using three factor matrixes between a first layer network (input video) and a next layer network, and mapping the result by using a sigmoid function to obtain a new third-order tensor (the next layer network); and respectively carrying out matrix multiplication and tensor multiplication for 5 times in sequence to obtain the final third-order tensor (output video).
The sigmoid function formula is as follows:
and the iteration ending condition is that the error or the iteration number reaches an upper limit. When the error value is smaller than the set error value, the iteration is ended; or when the iteration times are more than the set iteration times, ending the iteration.
The back propagation step in the step five is as follows:
step 5.1, solving the gradient from the output layer to the hidden layer;
according to the chain rule we obtain:
wherein:
Etotalis a tensor self-encoding network loss function, YrealFor input of video, YoutIn order to output the video, the video is output,x is more than or equal to 1 and less than or equal to m, y is more than or equal to 1 and less than or equal to n, z is more than or equal to 1 and less than or equal to p, i is more than or equal to 1 and less than or equal to r, j is more than or equal to 1 and less than or equal to s, k is more than or equal to 1 and less than or equal to t, A, B and C are three factor matrixes; m, r represents the size of matrix A, n, s represents the size of matrix B, p, t represents the size of matrix C, x represents an integer between 1 and m, y represents an integer between 1 and n, z represents an integer between 1 and p, i represents an integer between 1 and r, j represents an integer between 1 and s, k represents an integer between 1 and t, m, n, p, r, s, t belong to positive integers,represents a real number and a real number,represents a derivation symbol;
the same can be obtained:
step 5.2, solving the gradient from the hidden layer to the hidden layer:
according to the chain rule we obtain:
wherein the content of the first and second substances,a is more than or equal to 1 and less than or equal to u, b is more than or equal to 1 and less than or equal to v, and c is more than or equal to 1 and less than or equal to w; u, v, W represent the size of the tensor W, u, v, W are positive integers, a represents an integer between 1 and u, b represents an integer between 1 and v, and c represents an integer between 1 and W.
Solving for the hidden layer to hidden layer gradient is similar to the output layer to hidden layer gradient, except thatThe solving formula is as follows:
wherein the content of the first and second substances,other solving steps are the same as the steps in 1);
the following can be obtained by the same method:
step 5.3, the gradient from all hidden layers to the hidden layer is obtained by using the method in the step 5.2
Step 5.4, the tensor self-encoding network parameters are updated using the ADAM (adaptive moment estimation) method and the gradients found in step 5.1, step 5.2 and step 5.3.
The concrete steps of coding and compressing the core tensor and the compression network in the sixth step are as follows:
step 6.1, extracting and separating the integer part and the decimal part of the obtained nuclear tensor and compressed network parameter;
6.2, coding and compressing the integer part by using Huffman coding;
step 6.3, multiplying the separated fractional part by a scalar alpha and rounding the result;
step 6.4, carrying out quantization compression on the integer obtained in the step 6.3 by using a beta bit binary system;
and 6.5, storing the compression results of the integer and the decimal part.
The alpha is 2043, and the beta is 11.
The decompression step in the step 7 is as follows:
step 7.1, inputting the compression result obtained in the step six;
7.2, performing inverse coding on the core tensor and the compression network;
performing inverse Huffman transformation on the integer part to obtain a core tensor and an integer part of a network;
taking out binary compression representation of the decimal part, and dividing the binary compression representation by alpha to obtain a core tensor and a network decimal part;
combining the core tensor and the decimal and integer part of the network to obtain the parameters of the core tensor and the network;
step 7.3, carrying out decompression by using tensor modular multiplication;
and 7.4, outputting the decompressed video.
The invention has the beneficial effects that:
the invention carries out video compression through a tensor self-coding network, and the tensor has the capacity of expressing multi-dimensional complex data, so that the invention has higher compression rate and higher decompression speed than the prior method under the same condition, and can be an algorithm for monitoring video compression.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Fig. 2 is a back propagation flow chart.
Detailed Description
The invention is described in detail below with reference to the figures and the specific embodiments.
As shown in fig. 1, the video compression method based on the third-order tensor self-coding network of the present invention specifically includes the following steps:
inputting a target video (three-dimensional) and carrying out normalization and graying processing;
replacing full-connection parameters of the self-coding network with the three factor matrixes and setting iteration conditions;
setting parameters of the network and iteration ending conditions according to the required compression ratio and the peak signal-to-noise ratio; the upper limit of iteration times is set to 10^5, the error value of each pixel value is set to 0.0012, the size of the middle five hidden layers is set according to the compression rate and the peak signal-to-noise ratio, fewer hidden layer nodes are set if a larger compression rate needs to be obtained, and more hidden layer nodes are set if a higher PSNR value is obtained. The number of nodes of the third hidden layer is the minimum, the number of nodes of the first hidden layer is less than that of nodes of the input layer and greater than that of nodes of the second hidden layer, the number of nodes of the first hidden layer is equal to that of nodes of the fifth hidden layer, the number of nodes of the second hidden layer is the same as that of nodes of the fourth hidden layer, and the number of nodes of the input layer is the same as that of nodes of the output layer.
Step three, forward propagation is carried out;
performing modulo-1 modulo-2 modulo-3 multiplication by using three factor matrixes between a first layer network (input video) and a next layer network, and mapping the result by using a sigmoid function to obtain a new third-order tensor (the next layer network); and respectively carrying out matrix multiplication and tensor multiplication for 5 times in sequence to obtain the final third-order tensor (output video). sigmoid formula is as follows:
step four, whether an iteration end condition is reached or not is judged, and when the error value is smaller than the set error value, the iteration is ended; or when the iteration times are more than the set iteration times, the iteration is ended; if so, jumping to the sixth step to obtain a coding network and a compression network; otherwise, continuing the step five;
step five, performing reverse propagation, and then returning to the step three; the back propagation step is shown in figure 2,
step 5.1, solving the gradient from the output layer to the hidden layer: the output layer is the last third-order tensor, and except the third-order tensors of the input layer and the output layer, other third-order orderings are all hidden layers.
According to the chain rule we obtain:
wherein:
Etotalis a tensor self-encoding network loss function, YrealFor input of video, YoutIn order to output the video, the video is output,x is more than or equal to 1 and less than or equal to m, y is more than or equal to 1 and less than or equal to n, z is more than or equal to 1 and less than or equal to p, i is more than or equal to 1 and less than or equal to r, j is more than or equal to 1 and less than or equal to s, k is more than or equal to 1 and less than or equal to t, A, B and C are three factor matrixes, m and r represent the size of the matrix A, n and s represent the size of the matrix B, p and t represent the size of the matrix C, x represents an integer between 1 and m, y represents an integer between 1 and n, and z represents 1 to pI represents an integer from 1 to r, j represents an integer from 1 to s, k represents an integer from 1 to t, and m, n, p, r, s, t are positive integers.Represents a real number and a real number,represents a derivation symbol;
the same can be obtained:
step 5.2, solving the gradient from the hidden layer to the hidden layer:
wherein the content of the first and second substances,a is more than or equal to 1 and less than or equal to u, b is more than or equal to 1 and less than or equal to v, c is more than or equal to 1 and less than or equal to W, u, v and W represent the size of tensor W, u, v and W are positive integers, and a represents an integer between 1 and uB represents an integer between 1 and v, and c represents an integer between 1 and w.
Solving for the hidden layer to hidden layer gradient is similar to the output layer to hidden layer gradient, except thatIs solved by the formula
Wherein the content of the first and second substances,other solving steps are the same as the steps in 1);
the following can be obtained by the same method:
step 5.3, the gradient from all hidden layers to the hidden layer is obtained by using the method in the step 5.2
Step 5.4, updating network parameters:
the tensor self-encoding network parameters are updated using the ADAM method and the gradients found in steps 5.1 to 5.3.
Sixthly, coding and compressing the core tensor and the compression network;
step 6.1, extracting and separating the integer part and the decimal part of the nuclear tensor and compressed network parameters obtained by separation
Step 6.2, using Huffman coding to code and compress the integer part
Step 6.3, multiply the fractional part obtained by separation by scalar alpha and round the result, alpha is 2043
Step 6.4, the integers obtained in the step 3 are quantized and compressed by using a beta bit binary system, wherein beta is 11
And 6.5, storing the compression results of the integer and the decimal part.
Step seven, outputting a compression result;
the decompression step is as follows:
step 7.1, inputting the compression result obtained in the step six;
7.2, performing inverse coding on the core tensor and the compression network;
performing inverse Huffman transform on the integer part in the result to obtain the integer part of the core tensor and the network
Taking out binary compression representation of the fractional part, and dividing the binary compression representation by alpha to obtain a core tensor and a fractional part of the network
Combining the nuclear tensor and the decimal and integer parts of the network to obtain the parameters of the nuclear tensor and the network
Step 7.3, carrying out decompression by using tensor modular multiplication;
and 7.4, outputting the decompressed video.
Table 1 shows some evaluation criteria of the compression result of the embodiment of the method of the present invention, Peak Signal to Noise Ratio (PSNR), Structural Similarity Index (SSIM), and compression Ratio are image quality evaluation criteria, where PSNR and SSIM range from 0 to 1, and the larger the values of the three are, the better the image quality is.
Video name | miss-america | bridge-close | bridge-far | claire | grandma |
PSNR | 33.05 | 37.68 | 45.48 | 33.39 | 33.47 |
SSIM | 0.93 | 0.97 | 0.98 | 0.93 | 0.90 |
Compression ratio | 112.35 | 79.35 | 88.73 | 183.64 | 134.59 |
Table 1 video compression results quantitative analysis table.
The technical solutions of the present invention are not limited to the above embodiments, and all technical solutions obtained by using equivalent substitution modes fall within the scope of the present invention.
Claims (6)
1. A video compression method based on a third-order tensor self-coding network is characterized by comprising the following steps:
firstly, preprocessing graying and normalization of a target three-dimensional video;
setting full connection parameters between layers in a target video self-coding network as third-order tensors and setting iteration ending conditions;
step three, forward transmission of a video network is carried out;
performing modulo-1 modulo-2 modulo-3 multiplication by using three factor matrixes between a first layer network and a next layer network, and mapping the result by using a sigmoid function to obtain a new third-order tensor; respectively and sequentially carrying out matrix multiplication 5 times and tensor multiplication to obtain a final third-order tensor; the input video is a first-layer network, the obtained new third-order tensor is a next-layer network, and the final third-order tensor is the output video;
step four, judging whether iteration is terminated, if so, skipping to step six, and outputting a core tensor and decoding network parameters; otherwise, continuing the step five;
fifthly, performing back propagation of the video network;
sixthly, coding and compressing the core tensor and the decoding network;
step 6.1, extracting and separating the integer part and the decimal part of the obtained nuclear tensor and compressed network parameter;
6.2, coding and compressing the integer part by using Huffman coding;
step 6.3, multiplying the separated fractional part by a scalar alpha, and rounding the result, wherein alpha is 2043;
step 6.4, carrying out quantization compression on the integer obtained in the step 6.3 by using a beta bit binary system, wherein beta is 11;
6.5, storing the compression results of the integer and the decimal part;
and step seven, outputting the compressed video.
2. The video compression method based on the third order tensor self-coding network as recited in claim 1, wherein: and in the second step, parameters of the network and iteration ending conditions are set according to the required compression ratio and the peak signal-to-noise ratio.
4. the video compression method based on the third order tensor self-coding network as recited in claim 2, wherein: the end condition of the iteration is as follows: when the error value is smaller than the set error value, the iteration is ended; or when the iteration times are more than the set iteration times, ending the iteration.
5. The video compression method based on the third order tensor self-coding network as recited in claim 1, wherein:
the back propagation step in the step five is as follows:
step 5.1, solving the gradient from the output layer to the hidden layer;
according to the chain rule we obtain:
wherein:
Etotalis a tensor self-encoding network loss function, YrealFor input of video, YoutIn order to output the video, the video is output,1≤x≤m,1≤y≤n,1≤z≤p,1≤r is more than or equal to i, j is more than or equal to 1 and less than or equal to s, k is more than or equal to 1 and less than or equal to t, A, B and C are three factor matrixes; m, r represents the size of matrix A, n, s represents the size of matrix B, p, t represents the size of matrix C, x represents an integer between 1 and m, y represents an integer between 1 and n, z represents an integer between 1 and p, i represents an integer between 1 and r, j represents an integer between 1 and s, k represents an integer between 1 and t, m, n, p, r, s, t belong to positive integers,represents a real number and a real number,represents a derivation symbol;
the same can be obtained:
step 5.2, solving the gradient from the hidden layer to the hidden layer:
according to the chain rule we obtain:
wherein the content of the first and second substances,a is more than or equal to 1 and less than or equal to u, b is more than or equal to 1 and less than or equal to v, and c is more than or equal to 1 and less than or equal to w; u, v, W represent the size of the tensor W, u, v, W are positive integers, a represents an integer between 1 and u, b represents an integer between 1 and v, and c represents an integer between 1 and W;
the same can be obtained:
step 5.3, the gradient from all hidden layers to the hidden layer is obtained by using the method in the step 5.2
Step 5.4, the tensor self-encoding network parameters are updated using the ADAM (adaptive moment estimation) method and the gradients found in step 5.1, step 5.2 and step 5.3.
6. The video compression method based on the third order tensor self-coding network as recited in claim 1, wherein: the decompression step in the step 7 is as follows:
step 7.1, inputting the compression result obtained in the step six;
7.2, performing inverse coding on the core tensor and the compression network;
performing inverse Huffman transformation on the integer part to obtain a core tensor and an integer part of a network;
taking out binary compression representation of the decimal part, and dividing the binary compression representation by alpha to obtain a core tensor and a network decimal part;
combining the core tensor and the decimal and integer part of the network to obtain the parameters of the core tensor and the network;
step 7.3, carrying out decompression by using tensor modular multiplication;
and 7.4, outputting the decompressed video.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811168316.6A CN109302614B (en) | 2018-10-08 | 2018-10-08 | Video compression method based on third-order tensor self-coding network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811168316.6A CN109302614B (en) | 2018-10-08 | 2018-10-08 | Video compression method based on third-order tensor self-coding network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109302614A CN109302614A (en) | 2019-02-01 |
CN109302614B true CN109302614B (en) | 2022-01-18 |
Family
ID=65161883
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811168316.6A Active CN109302614B (en) | 2018-10-08 | 2018-10-08 | Video compression method based on third-order tensor self-coding network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109302614B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106934458A (en) * | 2015-12-31 | 2017-07-07 | 中国科学院深圳先进技术研究院 | Multilayer automatic coding and system based on deep learning |
CN107516129A (en) * | 2017-08-01 | 2017-12-26 | 北京大学 | The depth Web compression method decomposed based on the adaptive Tucker of dimension |
CN107944556A (en) * | 2017-12-12 | 2018-04-20 | 电子科技大学 | Deep neural network compression method based on block item tensor resolution |
CN107967516A (en) * | 2017-10-12 | 2018-04-27 | 中科视拓(北京)科技有限公司 | A kind of acceleration of neutral net based on trace norm constraint and compression method |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10242313B2 (en) * | 2014-07-18 | 2019-03-26 | James LaRue | Joint proximity association template for neural networks |
-
2018
- 2018-10-08 CN CN201811168316.6A patent/CN109302614B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106934458A (en) * | 2015-12-31 | 2017-07-07 | 中国科学院深圳先进技术研究院 | Multilayer automatic coding and system based on deep learning |
CN107516129A (en) * | 2017-08-01 | 2017-12-26 | 北京大学 | The depth Web compression method decomposed based on the adaptive Tucker of dimension |
CN107967516A (en) * | 2017-10-12 | 2018-04-27 | 中科视拓(北京)科技有限公司 | A kind of acceleration of neutral net based on trace norm constraint and compression method |
CN107944556A (en) * | 2017-12-12 | 2018-04-20 | 电子科技大学 | Deep neural network compression method based on block item tensor resolution |
Non-Patent Citations (1)
Title |
---|
《BP算法的改进及其应用研究》;黄庆斌;《中国优秀硕士学位论文全文数据库 信息科技辑》;20101015;正文第12-16页 * |
Also Published As
Publication number | Publication date |
---|---|
CN109302614A (en) | 2019-02-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109996071B (en) | Variable code rate image coding and decoding system and method based on deep learning | |
CN109889839B (en) | Region-of-interest image coding and decoding system and method based on deep learning | |
CN111147862B (en) | End-to-end image compression method based on target coding | |
CN110248190B (en) | Multilayer residual coefficient image coding method based on compressed sensing | |
WO2020237646A1 (en) | Image processing method and device, and computer-readable storage medium | |
CN110602494A (en) | Image coding and decoding system and method based on deep learning | |
CN110677651A (en) | Video compression method | |
CN111246206B (en) | Optical flow information compression method and device based on self-encoder | |
CN110753225A (en) | Video compression method and device and terminal equipment | |
CN109903351B (en) | Image compression method based on combination of convolutional neural network and traditional coding | |
CN111247797A (en) | Method and apparatus for image encoding and decoding | |
CN112702600B (en) | Image coding and decoding neural network layered fixed-point method | |
CN108182712B (en) | Image processing method, device and system | |
US9948928B2 (en) | Method and apparatus for encoding an image | |
CN109302614B (en) | Video compression method based on third-order tensor self-coding network | |
Hasnat et al. | Luminance approximated vector quantization algorithm to retain better image quality of the decompressed image | |
US20220392117A1 (en) | Data compression and decompression system and method thereof | |
WO2023082107A1 (en) | Decoding method, encoding method, decoder, encoder, and encoding and decoding system | |
CN115361559A (en) | Image encoding method, image decoding method, image encoding device, image decoding device, and storage medium | |
CN111107377A (en) | Depth image compression method, device, equipment and storage medium | |
Yudin et al. | Data Compression Based on Coding Methods With a Controlled Level of Quality Loss | |
CN112200301B (en) | Convolution computing device and method | |
Kamal et al. | Iteration free fractal compression using genetic algorithm for still colour images | |
CN114882133B (en) | Image coding and decoding method, system, device and medium | |
CN116916033B (en) | Combined space-time video compression method based on random self-adaptive Fourier decomposition |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right |
Effective date of registration: 20221123 Address after: Building B1, Software Valley Science and Technology City, No. 34, Dazhou Road, Yuhuatai District, Nanjing City, Jiangsu Province, 210012 Patentee after: Jiangsu Xinshiyun Science and Technology Co.,Ltd. Address before: 210044 No. 219 Ning six road, Jiangbei new district, Nanjing, Jiangsu Patentee before: Nanjing University of Information Science and Technology |
|
TR01 | Transfer of patent right |