Intelligent multi-stage sampling system
Through the intelligent multi-level sampling system, the feature extraction and recovery mechanisms of the downsampling module and the upsampling module are used to solve the problem of low data processing efficiency under the von Neumann architecture, and efficient multi-level data processing and storage optimization are achieved.
Patent Information
- Application Number
- CN202311830065.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-28
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2043-12-28
AI Technical Summary
Under the von Neumann architecture, the data processing progress is severely slowed down when facing massive data processing and multi-level data operations, especially in multi-level data upsampling and downsampling operations, frequent data processing and storage are required, resulting in Inefficient.
An intelligent multi-level sampling system was designed. Through k downsampling modules and k upsampling modules, the data processing was carried out step by step by step by step through feature extraction and feature recovery, and the data address was allocated during the downsampling process, avoiding the traditional The redundancy of data storage in the von Neumann architecture directly calls downsampled data during the upsampling process.
It effectively reduces storage space and system power consumption, improves data processing efficiency, and realizes a smooth multi-level sampling process.
Smart Images

Figure CN117785053B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to an intelligent multi-stage sampling system. Background Art
[0002] The basic architecture followed by the development of modern computers is the von Neumann architecture. This architecture is characterized by "program storage, shared data, and sequential execution", and requires the CPU to fetch instructions and data from the memory for corresponding calculations.
[0003] In other words, under the von Neumann architecture, instructions and data are stored in a specific space (ie, memory), while computations on the instructions and data are performed in another space.
[0004] The von Neumann architecture has been widely used following the development of modern computers. However, with the emergence of artificial intelligence, computers may need to process massive amounts of data, and data processing is multi-level, which makes the von Neumann architecture with separate storage and computing space often unable to cope with it, especially when facing multi-level data upsampling and downsampling operations, which requires data processing, data storage, and data processing again, and so on, which seriously slows down the progress of data processing. Summary of the invention
[0005] The present invention provides an intelligent multi-stage sampling system, which effectively solves the above technical problems existing in the prior art.
[0006] Specifically, the present invention provides an intelligent multi-stage sampling system, which includes k down-sampling modules and k up-sampling modules, k ≥ 3, wherein the initial data set M 0 First, the first-level downsampling module is input for feature extraction, thereby forming the first-level downsampling data set M 1 At the same time, the first level data address A is assigned 1 To store the first level down-sampled data set M 1 , then, the first stage downsampled data set M 1 Input the second level downsampling module for feature extraction, thereby forming the second level downsampling data set M 2 At the same time, the second level data address A is allocated 2 To store the second level down-sampled data group M 2 , and so on, until the k-1th level down-sampled data set M k-1 Input the k-th level downsampling module for feature extraction, thereby forming the k-th level downsampling data set M k At the same time, the k-th level data address A is allocated k To store the k-th level down-sampled data set M k , then, the k-th level downsampled data set M kInput the first level upsampling module for feature recovery, thereby forming the first level upsampling data group N 1 , then, from the k-1th level data address A k-1 The k-1th level down-sampled data group M is retrieved k-1 , thus, in the second-level upsampling module, based on the k-1th level downsampling data group M k-1 , for the first level upsampled data set N 1 Perform feature recovery to form the second level upsampling data group N 2 , and so on, the integer i takes values from 1 to k-1 in sequence, starting from the ki-th level data address A k-i The ki-th level down-sampled data set M is retrieved k-i , thus, in the i+1th level upsampling module, based on the kith level downsampling data set M k-i , perform feature recovery on the i-th level upsampled data group Ni to form the i+1-th level upsampled data group N i , until i takes the value k-1, the k-th level upsampling data group N is finally formed k .
[0007] Preferably, when each level of downsampling modules performs feature extraction, a pooling operation is used to perform feature extraction on the above-mentioned level of downsampling data groups.
[0008] More preferably, the pooling operation is implemented using a maximum pooling layer, or an average pooling layer.
[0009] Optionally, the initial data set M 0 is a 4-dimensional image data set.
[0010] Optionally, the 4-dimensional image data set is expressed as [N, 540, 960, 4], wherein the 4 dimensions in the image data set are the number of channels, width, height, and number of frames of the image data, respectively.
[0011] The intelligent multi-stage sampling system provided by the present invention does not specifically set up a memory to store each level of down-sampling data group when the down-sampling data group is applied to the up-sampling process, but allocates a data address to the down-sampling data group at each step node of the down-sampling. Therefore, in the subsequent up-sampling steps of each level, the original down-sampling data of the corresponding level is sequentially called from the allocated data address according to the previous level. In the process of orderly progress of the process, the original down-sampling data is sequentially called out and applied to up-sampling, thereby greatly reducing the storage space and reducing the power consumption of the system operation. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be discussed below. Obviously, the technical solutions described in conjunction with the drawings are only some embodiments of the present invention. For ordinary technicians in this field, other embodiments and their drawings can be obtained based on the embodiments shown in these drawings without paying creative work.
[0013] Figure 1 The working flow chart of the intelligent multi-stage sampling system according to the present invention is shown;
[0014] Figure 2 An exemplary flow chart of an intelligent multi-stage sampling system according to the present invention is shown. DETAILED DESCRIPTION
[0015] The technical solutions of various embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments described in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0016] The present invention provides an intelligent multi-stage sampling system, which integrates data up-sampling operation and data down-sampling operation, and finally forms output data after multi-stage up-and-down sampling.
[0017] Figure 1 The working flow chart of the intelligent multi-stage sampling system according to the present invention is shown. Figure 1 The left side is the downsampling process. Figure 1 The right side shows the upsampling process.
[0018] like Figure 1 As shown, the initial data set M 0 First, the first-level downsampling module is input for feature extraction, thereby forming the first-level downsampling data set M 1 At the same time, the first level data address A is assigned 1 To store the first level down-sampled data set M 1 .
[0019] Next, the first stage downsamples the data set M 1 Input the second level downsampling module for feature extraction, thereby forming the second level downsampling data set M 2 At the same time, the second level data address A is allocated 2 To store the second level down-sampled data group M 2 .
[0020] And so on, until the k-1th level down-sampled data set M k-1Input the k-th level downsampling module for feature extraction, thereby forming the k-th level downsampling data set M k At the same time, the k-th level data address A is allocated k To store the k-th level down-sampled data set M k .
[0021] This forms a total of k-level down-sampled data sets M 1 、M 2 …M k , each stored in data address A 1 , A 2 …A k middle.
[0022] Preferably, a pooling operation can be used to extract features from the above-mentioned data groups at each level. For example, a maximum pooling layer can be used to implement the pooling operation, and an average pooling layer can also be used to implement the pooling operation.
[0023] Next, the k-th level downsampled data set M k Input the first level upsampling module for feature recovery, thereby forming the first level upsampling data group N 1 .
[0024] Then, from the k-1th level data address A k-1 The k-1th level down-sampled data group M is retrieved k-1 , thus, in the second-level upsampling module, based on the k-1th level downsampling data group M k-1 , for the first level upsampled data set N 1 Perform feature recovery to form the second level upsampling data group N 2 .
[0025] Similarly, the integer i takes values from 1 to k-1, starting from the ki-th level data address A k-i The ki-th level down-sampled data set M is retrieved k-i , thus, in the i+1th level upsampling module, based on the kith level downsampling data set M k-i , perform feature recovery on the i-th level upsampled data group Ni to form the i+1-th level upsampled data group N i , until i takes the value k-1, the kth level upsampling data group N is finally formed k .
[0026] In the above process, downsampling is performed for k iterations and then upsampling is performed for k iterations. It can be seen from the above that the downsampling data groups at each level formed in the downsampling process are applied to the upsampling feature recovery in sequence according to the order of iterations in the upsampling process.
[0027] Compared with the traditional von Neumann architecture, the intelligent multi-stage sampling system provided by the present invention does not specifically set up a memory to store the down-sampling data groups at each level when the down-sampling data groups are applied to the up-sampling process, but allocates data addresses to the down-sampling data groups at each step node of the down-sampling. Therefore, in the subsequent up-sampling steps at each level, the original down-sampling data of the corresponding levels are sequentially called from the allocated data addresses according to the previous levels, and the original down-sampling data are sequentially called out and applied to up-sampling in the process of orderly progress.
[0028] Figure 2 FIG. 4 shows an exemplary flow chart of the intelligent multi-stage sampling system according to the present invention. Figure 2 As shown, the initial data set M 0 It is a 4-dimensional image data set [N, 540, 960, 4], where the 4 dimensions are the number of channels, width, height, and number of frames of the image data.
[0029] Figure 2 The middle left column is the downsampling process, and after the downsampling process from top to bottom in the left column ends, it turns to the upsampling process from bottom to top in the right column. The five arrows between the two columns indicate that the corresponding downsampling data group of each level is called from the corresponding allocation address in the upsampling operation.
[0030] The above description is only an exemplary embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. An intelligent multi-stage sampling system, It is characterized in that The intelligent multi-stage sampling system includes k down-sampling modules and k up-sampling modules, k≥3, wherein: Initial data set M 0 First, the first-level downsampling module is input for feature extraction, thereby forming the first-level downsampling data set M 1 At the same time, the first level data address A is assigned 1 To store the first level down-sampled data set M 1 , Next, the first stage downsamples the data set M 1 Input the second level downsampling module for feature extraction, thereby forming the second level downsampling data set M 2 At the same time, the second level data address A is allocated 2 To store the second level down-sampled data group M 2 , And so on, until the k-1th level down-sampled data set M k-1 Input the k-th level downsampling module for feature extraction, thereby forming the k-th level downsampling data set M k At the same time, the k-th level data address A is allocated k To store the k-th level down-sampled data set M k , Next, the k-th level downsampled data set M k Input the first level upsampling module for feature recovery, thereby forming the first level upsampling data group N 1 , Then, from the k-1th level data address A k-1 The k-1th level down-sampled data group M is retrieved k-1 , thus, in the second-level upsampling module, based on the k-1th level downsampling data group M k-1 , for the first level upsampled data set N 1 Perform feature recovery to form the second level upsampling data group N 2 , Similarly, the integer i takes values from 1 to k-1, starting from the ki-th level data address A k-i The ki-th level down-sampled data set M is retrieved k-i , thus, in the i+1th level upsampling module, based on the kith level downsampling data set M k-i , perform feature recovery on the i-th level upsampled data group Ni to form the i+1-th level upsampled data group N i , until i takes the value k-1, the k-th level upsampling data group N is finally formed k .
2. The intelligent multi-stage sampling system according to claim 1, It is characterized in that When performing feature extraction in each level of downsampling modules, a pooling operation is used to extract features from the above-mentioned downsampling data groups at each level.
3. The intelligent multi-stage sampling system according to claim 2, It is characterized in that The pooling operation is implemented using a maximum pooling layer, or an average pooling layer.
4. The intelligent multi-stage sampling system according to claim 1, It is characterized in that Initial data set M 0 is a 4-dimensional image data set.
5. The intelligent multi-stage sampling system according to claim 4, It is characterized in that The 4-dimensional image data set is expressed as [N, 540, 960, 4], wherein the 4 dimensions in the image data set are the number of channels, width, height, and number of frames of the image data, respectively.
Citation Information
Patent Citations
Floodlight effect rendering method and device, electronic equipment and machine readable storage medium
CN116637365A
Image processing method, neural network for image processing, equipment and medium
CN117095264A
