CN107766932A - Image processing method and device based on neutral net - Google Patents
Image processing method and device based on neutral net Download PDFInfo
- Publication number
- CN107766932A CN107766932A CN201710983501.XA CN201710983501A CN107766932A CN 107766932 A CN107766932 A CN 107766932A CN 201710983501 A CN201710983501 A CN 201710983501A CN 107766932 A CN107766932 A CN 107766932A
- Authority
- CN
- China
- Prior art keywords
- subgraph
- operation layer
- computing
- image
- layer
- 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.)
- Granted
Links
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of image processing method and device based on neutral net.Wherein, this method includes：Obtain pending image；Pending image is converted into various sizes of at least one subgraph；An at least subgraph is inputted to multilayer neural network and carries out computing；Wherein, multilayer neural network includes multiple operation layers, connected between multiple operation layers by least one communication interface, n-th of operation layer will obtain operation result to transmit to (n+1)th operation layer, n by communication interface being positive integer to the progress computing of at least one subgraph in multiple operation layers.The present invention is solved due to that must wait until that the operation result in local caches is transferred to internal memory in correlation technique, and new input data is from after internal memory is loaded local caches, the technical problem of the wasting of resources caused by processor just can proceed by the operation of next round.
Description
Technical field
The present invention relates to image processing field, in particular to a kind of image processing method based on neutral net and
Device.
Background technology
At present, it is necessary to just can be by the following group after the completion of substantial amounts of Primary Stage Data is shifted in the transmitting procedure of neutral net middle level
Data carry out buffered.For in general neutral net, there must be substantial amounts of data to shift between local caches and internal memory,
Such large-scale memory access is relatively time consuming, therefore the overall efficiency of artificial neural networks can be limited.Further, since it must wait
Operation result into local caches is transferred to internal memory, and new input data is from after internal memory is loaded local caches,
Processor can just proceed by the operation of next round.Because Neural Network Data is huger so processor needs to expend
Substantial amounts of time resource waits the loading of pending data and acquisition, causes the waste of resource.
For it is above-mentioned the problem of, not yet propose effective solution at present.
The content of the invention
The embodiments of the invention provide a kind of image processing method and device based on neutral net, with least solve due to
It must wait until that the operation result in local caches is transferred to internal memory in correlation technique, and new input data is added from internal memory
After carrying local caches, the technical problem of the wasting of resources caused by processor just can proceed by the operation of next round.
One side according to embodiments of the present invention, there is provided a kind of image processing method based on neutral net, including：
Obtain pending image；Above-mentioned pending image is converted into various sizes of at least one subgraph；Will be above-mentioned at least one
Subgraph inputs to multilayer neural network and carries out computing；Wherein, above-mentioned multilayer neural network includes multiple operation layers, above-mentioned multiple
Connected between operation layer by least one communication interface, n-th of operation layer will be to above-mentioned at least one in above-mentioned multiple operation layers
Individual subgraph carries out computing and obtains operation result to transmit to (n+1)th operation layer, n by above-mentioned communication interface being positive integer.
Alternatively, each operation layer includes computing circuit and local memory, and above-mentioned local memory is used to store above-mentioned computing
Above-mentioned at least one subgraph that circuit uses, above-mentioned computing circuit are used to obtain above-mentioned at least one subgraph progress computing
Operation result, wherein, the computing circuit of n-th of operation layer by obtained above-mentioned operation result by above-mentioned communication interface transmit to
Local memory in (n+1)th operation layer.
Alternatively, the computing circuit of n-th of operation layer is in m-th obtained of operation result, the computing of n-th of operation layer
Circuit is simultaneously transmitted the m-1 obtained operation result to the local in (n+1)th operation layer by above-mentioned communication interface
Deposit, 1<m<N, m are integer.
Alternatively, after above-mentioned pending image being converted into various sizes of at least one subgraph, the above method is also
Including：According to the size of above-mentioned at least one subgraph, above-mentioned at least one image information is extracted；By above-mentioned at least one image
Information inputs to above-mentioned multilayer neural network and carries out computing.
Alternatively, the above-mentioned size according to above-mentioned at least one subgraph, extracting above-mentioned at least one image information includes：
If the size of above-mentioned subgraph is less than preset value, profile information is extracted from above-mentioned subgraph as above-mentioned image information；On if
The size for stating subgraph is more than or equal to above-mentioned preset value, and detailed information is extracted from above-mentioned subgraph as above-mentioned image information.
Alternatively, above-mentioned multiple operation layers include at least n chip；Above-mentioned multiple operation layers include convolutional layer, wherein, on
Stating convolutional layer includes multiple chips in parallel；Above-mentioned multiple operation layers include articulamentum, wherein, above-mentioned articulamentum includes in parallel
Multiple chips.
Another aspect according to embodiments of the present invention, a kind of image processing apparatus based on neutral net is additionally provided, wrapped
Include：Acquiring unit, for obtaining pending image；Conversion unit, it is various sizes of for above-mentioned pending image to be converted into
At least one subgraph；Arithmetic element, computing is carried out for above-mentioned at least one subgraph to be inputted to multilayer neural network；Its
In, above-mentioned multilayer neural network includes multiple operation layers, is connected between above-mentioned multiple operation layers by least one communication interface,
In above-mentioned multiple operation layers n-th operation layer above-mentioned at least one subgraph will be carried out computing obtain operation result pass through it is above-mentioned
It is positive integer that communication interface, which is transmitted to (n+1)th operation layer, n,.
Alternatively, each operation layer includes computing circuit and local memory, and above-mentioned local memory is used to store above-mentioned computing
Above-mentioned at least one subgraph that circuit uses, above-mentioned computing circuit are used to obtain above-mentioned at least one subgraph progress computing
Operation result, wherein, the computing circuit of n-th of operation layer by obtained above-mentioned operation result by above-mentioned communication interface transmit to
Local memory in (n+1)th operation layer.
Another aspect according to embodiments of the present invention, additionally provides a kind of storage medium, and above-mentioned storage medium includes storage
Program, wherein, said procedure perform with above-mentioned arbitrary characteristics the image processing method based on neutral net.
Another aspect according to embodiments of the present invention, a kind of processor being additionally provided, above-mentioned processor is used for operation program,
Wherein, the image processing method based on neutral net with above-mentioned arbitrary characteristics is performed when said procedure is run.
In embodiments of the present invention, using the pending image of acquisition；By pending image be converted into it is various sizes of at least
One subgraph；An at least subgraph is inputted to multilayer neural network and carries out computing；Wherein, multilayer neural network includes more
Individual operation layer, connected by least one communication interface between multiple operation layers, n-th of operation layer will be to extremely in multiple operation layers
Few subgraph progress computing obtains operation result and transmitted by communication interface to (n+1)th operation layer, and n is the side of positive integer
Formula, by the way that pending image is converted into various sizes of at least one subgraph, and transported simultaneously by multiple operation layers
Calculate, reached the purpose for improving operation efficiency, it is achieved thereby that the technique effect to economize on resources, and then solve due to related skill
It must wait until that the operation result in local caches is transferred to internal memory in art, and new input data is loaded local from internal memory
After buffer, the technical problem of the wasting of resources caused by processor just can proceed by the operation of next round.
Brief description of the drawings
Accompanying drawing described herein is used for providing a further understanding of the present invention, forms the part of the application, this hair
Bright schematic description and description is used to explain the present invention, does not form inappropriate limitation of the present invention.In the accompanying drawings：
Fig. 1 is a kind of flow signal of optionally image processing method based on neutral net according to embodiments of the present invention
Figure；
Fig. 2 is a kind of structural representation of optional multilayer neural network according to embodiments of the present invention；
Fig. 3 is a kind of structural representation of optionally image processing apparatus based on neutral net according to embodiments of the present invention
Figure.
Embodiment
In order that those skilled in the art more fully understand the present invention program, below in conjunction with the embodiment of the present invention
Accompanying drawing, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is only
The embodiment of a part of the invention, rather than whole embodiments.Based on the embodiment in the present invention, ordinary skill people
The every other embodiment that member is obtained under the premise of creative work is not made, it should all belong to the model that the present invention protects
Enclose.
It should be noted that term " first " in description and claims of this specification and above-mentioned accompanying drawing, "
Two " etc. be for distinguishing similar object, without for describing specific order or precedence.It should be appreciated that so use
Data can exchange in the appropriate case, so as to embodiments of the invention described herein can with except illustrating herein or
Order beyond those of description is implemented.In addition, term " comprising " and " having " and their any deformation, it is intended that cover
Cover it is non-exclusive include, be not necessarily limited to for example, containing the process of series of steps or unit, method, system, product or equipment
Those steps or unit clearly listed, but may include not list clearly or for these processes, method, product
Or the intrinsic other steps of equipment or unit.
Embodiment 1
According to embodiments of the present invention, there is provided a kind of embodiment of the method for the image processing method based on neutral net, need
It is noted that can be in the computer system of such as one group computer executable instructions the flow of accompanying drawing illustrates the step of
Middle execution, although also, show logical order in flow charts, in some cases, can be with different from herein
Order performs shown or described step.
Fig. 1 is the image processing method according to embodiments of the present invention based on neutral net, as shown in figure 1, this method bag
Include following steps：
Step S102, obtain pending image.
In the application above-mentioned steps S102, by being divided into different region units to be input to buffer area, meeting entire image
It is substantial amounts of to expend the loading time, while accuracy can also have a certain impact.The present embodiment is N number of different by the way that image is divided into
Resolution ratio level by N number of operation layer come simultaneous transmission data, can reach faster transmission rate.
Step S104, pending image is converted into various sizes of at least one subgraph.
In the application above-mentioned steps S104, after pending image is got, pending image is converted at least one
Individual subgraph, wherein, the size of at least one subgraph is different.
Step S106, an at least subgraph is inputted to multilayer neural network and carries out computing.
In the application above-mentioned steps S106, multilayer neural network includes multiple operation layers, by extremely between multiple operation layers
A few communication interface connects, and n-th of operation layer will carry out computing at least one subgraph and obtain computing in multiple operation layers
As a result it is positive integer to be transmitted by communication interface to (n+1)th operation layer, n.Wherein, the communication interface is high speed communication interface.
In the present embodiment, high speed communication interface can include high-speed serializer-deserializer, radio frequency interface.It is appreciated that
, the transmission rate of high speed communication interface is faster, and the time that data are transmitted between operation layer is shorter.
As shown in Fig. 2 each self-contained independent hardware configuration of each operation layer, wherein, each operation layer includes computing electricity
Road and local memory, local memory (are stored with for storing at least one subgraph that computing circuit uses in local memory
The input data that is used for computing circuit and can learning parameter), computing circuit is used to carry out computing at least one subgraph
Operation result is obtained, wherein, the computing circuit of n-th of operation layer transmits obtained operation result to n-th by communication interface+
Local memory in 1 operation layer.
Alternatively, the computing circuit of n-th of operation layer is in m-th obtained of operation result, the computing of n-th of operation layer
Circuit is simultaneously transmitted the m-1 obtained operation result to the local memory in (n+1)th operation layer by communication interface, and 1<
m<N, m are integer.
Alternatively, input layer is also included in multilayer neural network, the input layer is the hidden of front end in multilayer neural network
Hide layer, the input data that stores in local memory and can learning parameter be from the input layer.
By above-mentioned steps, by the way that pending image is converted into various sizes of at least one subgraph, and by more
Individual operation layer carries out computing simultaneously, has reached the purpose for improving operation efficiency, it is achieved thereby that the technique effect to economize on resources, enters
And solve due to that must wait until that the operation result in local caches is transferred to internal memory, and new input number in correlation technique
After local caches are loaded from internal memory, the wasting of resources caused by processor just can proceed by the operation of next round
Technical problem.
As a kind of optional implementation, by pending image be converted into various sizes of at least one subgraph it
Afterwards, method also includes：According to the size of at least one subgraph, at least one image information is extracted；At least one image is believed
Breath input to multilayer neural network carries out computing.
For example, the size of pending image is L1 × L2, by pending image is gradually reduced turn into a Li1 ×
Li2 subgraph, by the difference of detailed information in various sizes of subgraph and different size subgraphs, while input more
In layer Artificial neural, the quick load convergence of information is reached.
Alternatively, according to the size of at least one subgraph, extracting at least one image information includes：If the chi of subgraph
It is very little to be less than preset value, profile information is extracted from subgraph as image information；If the size of subgraph is more than or equal to preset value,
Detailed information is extracted from subgraph as image information.
Alternatively, detailed information is extracted from subgraph includes：The image that extraction subgraph is sampled using same class two-dimensional is believed
Breath；The profile information of at least one subgraph is rejected from image information, to obtain detailed information.Wherein, from image information
Rejecting the profile information of image includes：The profile information of subgraph is rejected from image information by wave filter, and to rejecting
The position of each pixel of the subgraph of profile information is recorded with detailed information, mapped.
By the way that pyramid of the pending image Jing Guo N grades is changed, N number of various sizes of son can be gradually transformed into
Image, (picture position, image information) is extracted in the sampling that two-dimensional signal is then carried out to different subgraphs, due to small size
Subgraph carried out profile information extraction, the profile of image can be rejected in for large-sized subgraph, and
Details is retained, large-sized subgraph sampled with class two-dimensional and extracted.In the present embodiment, it is with class two-dimensional sampling
Refer to the method that the position of current pixel point is recorded, mapped with detailed information.
Specifically, sampled with class two-dimensional：The detailed information or profile information that the image of general different resolution is included be
It is different, it is relatively more for large-sized image (image of i.e. big resolution ratio) detailed information, and for small size image (i.e.
The image of small resolution ratio) general profile information is more comprehensive, such as leaf, the train of thought of the image of big resolution ratio generally for leaf
Details is clearer, and the information that the image of small resolution ratio contains to the profile of leaf is relatively more.For different resolution ratio
Image can be stored by being sampled to image detail to generate a two-dimentional function f (x, y), wherein x, y representative images
Position, f (x, y) represent detailed information.
Alternatively, multiple operation layers include at least n chip；Multiple operation layers include convolutional layer, wherein, convolutional layer includes
Multiple chips in parallel；Multiple operation layers include articulamentum, wherein, articulamentum includes multiple chips in parallel.Wherein, Duo Geyun
Calculating layer can realize that, by greater number of chip of connecting, multilayer neural network is wrapped using multiple identical chips
The quantity of the operation layer contained can arbitrarily increase.
The convolutional layer is used to judge by the edge to pending image and details correlation function, if details and wheel
Wide relevance is very high, it can be determined that the relation of image and profile, can more preferable basis when convolution mask is selected
Strength of association carries out changing convolution mask in real time.
So faster conversion rate can be realized by carrying out convolution transform to different piece, when hardware can be synchronous
When handling different grades of conversion, the lifting of speed can be realized to the conversion rate of image, simultaneously because to different profiles
The feature of image can be more accurately extracted using different convolution masks.
You need to add is that above-mentioned chip can be arranged in same IC wafer or be arranged at same encapsulation
In different IC wafers, or it is arranged in different encapsulation, that is to say, that same crystalline substance can be passed through between adjacent operation layer
In piece, across chip, the high speed communication interface interconnection across encapsulating or across circuit board.
Embodiment 2
The embodiment of the present invention additionally provides a kind of image processing apparatus based on neutral net.It should be noted that the reality
The image processing method based on neutral net that can be used for performing the embodiment of the present invention is put in the pond makeup for applying the image information of example.
Fig. 3 is a kind of image processing apparatus based on neutral net according to embodiments of the present invention.As shown in figure 3, the base
Include in the image processing apparatus of neutral net：Acquiring unit 30, conversion unit 32 and arithmetic element 34.
Acquiring unit 30, for obtaining pending image；
Conversion unit 32, for pending image to be converted into various sizes of at least one subgraph；
Arithmetic element 34, computing is carried out for an at least subgraph to be inputted to multilayer neural network；
Wherein, multilayer neural network includes multiple operation layers, is connected between multiple operation layers by least one communication interface
Connect, n-th operation layer will carry out computing at least one subgraph and obtain operation result passing through communication interface in multiple operation layers
It is positive integer to transmit to (n+1)th operation layer, n.
Alternatively, each operation layer includes computing circuit and local memory, and local memory is used to store computing circuit use
At least one subgraph, computing circuit be used at least one subgraph carry out computing obtain operation result, wherein, n-th
The computing circuit of operation layer transmits obtained operation result to the local memory in (n+1)th operation layer by communication interface.
Alternatively, the computing circuit of n-th of operation layer is in m-th obtained of operation result, the computing of n-th of operation layer
Circuit is simultaneously transmitted the m-1 obtained operation result to the local memory in (n+1)th operation layer by communication interface, and 1<
m<N, m are integer.
Alternatively, the image processing apparatus based on neutral net also includes：Processing unit, for according at least one subgraph
The size of picture, extract at least one image information；At least one image information is inputted to multilayer neural network and carries out computing.
Alternatively, processing unit is used to perform size of the following steps according at least one subgraph, and extraction is at least one
Image information：If the size of subgraph is less than preset value, profile information is extracted from subgraph as image information；If subgraph
Size be more than or equal to preset value, from subgraph extract detailed information as image information.
Alternatively, multiple operation layers include at least n chip；Multiple operation layers include convolutional layer, wherein, convolutional layer includes
Multiple chips in parallel；Multiple operation layers include articulamentum, wherein, articulamentum includes multiple chips in parallel.
In embodiments of the present invention, using the pending image of acquisition；By pending image be converted into it is various sizes of at least
One subgraph；An at least subgraph is inputted to multilayer neural network and carries out computing；Wherein, multilayer neural network includes more
Individual operation layer, connected by least one communication interface between multiple operation layers, n-th of operation layer will be to extremely in multiple operation layers
Few subgraph progress computing obtains operation result and transmitted by communication interface to (n+1)th operation layer, and n is the side of positive integer
Formula, by the way that pending image is converted into various sizes of at least one subgraph, and transported simultaneously by multiple operation layers
Calculate, reached the purpose for improving operation efficiency, it is achieved thereby that the technique effect to economize on resources, and then solve due to related skill
It must wait until that the operation result in local caches is transferred to internal memory in art, and new input data is loaded local from internal memory
After buffer, the technical problem of the wasting of resources caused by processor just can proceed by the operation of next round.
Embodiment 3
The embodiment of the present invention additionally provides a kind of storage medium, and storage medium includes the program of storage, wherein, program performs
The image processing method based on neutral net with above-mentioned arbitrary characteristics.
The embodiment of the present invention additionally provides a kind of processor, and processor is used for operation program, wherein, program is performed with upper
State the image processing method based on neutral net of arbitrary characteristics.
The embodiments of the present invention are for illustration only, do not represent the quality of embodiment.
In the above embodiment of the present invention, the description to each embodiment all emphasizes particularly on different fields, and does not have in some embodiment
The part of detailed description, it may refer to the associated description of other embodiment.
In several embodiments provided herein, it should be understood that disclosed technology contents, others can be passed through
Mode is realized.Wherein, device embodiment described above is only schematical, such as the division of the unit, Ke Yiwei
A kind of division of logic function, can there is an other dividing mode when actually realizing, for example, multiple units or component can combine or
Person is desirably integrated into another system, or some features can be ignored, or does not perform.Another, shown or discussed is mutual
Between coupling or direct-coupling or communication connection can be INDIRECT COUPLING or communication link by some interfaces, unit or module
Connect, can be electrical or other forms.
The unit illustrated as separating component can be or may not be physically separate, show as unit
The part shown can be or may not be physical location, you can with positioned at a place, or can also be distributed to multiple
On unit.Some or all of unit therein can be selected to realize the purpose of this embodiment scheme according to the actual needs.
In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, can also
That unit is individually physically present, can also two or more units it is integrated in a unit.Above-mentioned integrated list
Member can both be realized in the form of hardware, can also be realized in the form of SFU software functional unit.
If the integrated unit is realized in the form of SFU software functional unit and is used as independent production marketing or use
When, it can be stored in a computer read/write memory medium.Based on such understanding, technical scheme is substantially
The part to be contributed in other words to prior art or all or part of the technical scheme can be in the form of software products
Embody, the computer software product is stored in a storage medium, including some instructions are causing a computer
Equipment (can be personal computer, server or network equipment etc.) perform each embodiment methods described of the present invention whole or
Part steps.And foregoing storage medium includes：USB flash disk, read-only storage (ROM, Read-Only Memory), arbitrary access are deposited
Reservoir (RAM, Random Access Memory), mobile hard disk, magnetic disc or CD etc. are various can be with store program codes
Medium.
Described above is only the preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, under the premise without departing from the principles of the invention, some improvements and modifications can also be made, these improvements and modifications also should
It is considered as protection scope of the present invention.
Claims (10)
- A kind of 1. image processing method based on neutral net, it is characterised in that including：Obtain pending image；The pending image is converted into various sizes of at least one subgraph；At least one subgraph is inputted to multilayer neural network and carries out computing；Wherein, the multilayer neural network includes multiple operation layers, is connect between the multiple operation layer by least one communication Mouth connects, and n-th of operation layer will carry out computing at least one subgraph and obtain operation result in the multiple operation layer It is positive integer to be transmitted by the communication interface to (n+1)th operation layer, n.
- 2. according to the method for claim 1, it is characterised in that each operation layer includes computing circuit and local memory, institute State local memory to be used to store at least one subgraph that the computing circuit uses, the computing circuit is used for described At least one subgraph carries out computing and obtains operation result, wherein, the computing that the computing circuit of n-th of operation layer will obtain As a result transmitted by the communication interface to the local memory in (n+1)th operation layer.
- 3. according to the method for claim 2, it is characterised in that the computing circuit of n-th of operation layer is in m-th obtained of fortune When calculating result, the computing circuit of n-th of operation layer is simultaneously transmitted the m-1 obtained operation result by the communication interface Local memory into (n+1)th operation layer, 1<m<N, m are integer.
- 4. according to the method for claim 1, it is characterised in that by the pending image be converted into it is various sizes of at least After one subgraph, methods described also includes：According to the size of at least one subgraph, at least one image information is extracted；At least one image information is inputted to the multilayer neural network and carries out computing.
- 5. according to the method for claim 4, it is characterised in that the size according at least one subgraph, carry At least one image information is taken to include：If the size of the subgraph is less than preset value, profile information is extracted from the subgraph as described image information；If the size of the subgraph is more than or equal to the preset value, detailed information is extracted from the subgraph as the figure As information.
- 6. method according to any one of claim 1 to 5, it is characterised in thatThe multiple operation layer includes at least n chip；The multiple operation layer includes convolutional layer, wherein, the convolutional layer includes multiple chips in parallel；The multiple operation layer includes articulamentum, wherein, the articulamentum includes multiple chips in parallel.
- A kind of 7. image processing apparatus based on neutral net, it is characterised in that including：Acquiring unit, for obtaining pending image；Conversion unit, for the pending image to be converted into various sizes of at least one subgraph；Arithmetic element, computing is carried out at least one subgraph to be inputted to multilayer neural network；Wherein, the multilayer neural network includes multiple operation layers, is connect between the multiple operation layer by least one communication Mouth connects, and n-th of operation layer will carry out computing at least one subgraph and obtain operation result in the multiple operation layer It is positive integer to be transmitted by the communication interface to (n+1)th operation layer, n.
- 8. device according to claim 7, it is characterised in that each operation layer includes computing circuit and local memory, institute State local memory to be used to store at least one subgraph that the computing circuit uses, the computing circuit is used for described At least one subgraph carries out computing and obtains operation result, wherein, the computing that the computing circuit of n-th of operation layer will obtain As a result transmitted by the communication interface to the local memory in (n+1)th operation layer.
- A kind of 9. storage medium, it is characterised in that the storage medium includes the program of storage, wherein, described program right of execution Profit requires the image processing method based on neutral net described in any one in 1 to 6.
- A kind of 10. processor, it is characterised in that the processor is used for operation program, wherein, right of execution when described program is run Profit requires the image processing method based on neutral net described in any one in 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710983501.XA CN107766932B (en) | 2017-10-19 | 2017-10-19 | Image processing method and device based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710983501.XA CN107766932B (en) | 2017-10-19 | 2017-10-19 | Image processing method and device based on neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107766932A true CN107766932A (en) | 2018-03-06 |
| CN107766932B CN107766932B (en) | 2021-06-04 |
Family
ID=61269886
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710983501.XA Active CN107766932B (en) | 2017-10-19 | 2017-10-19 | Image processing method and device based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107766932B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110287767A (en) * | 2019-05-06 | 2019-09-27 | 深圳市华付信息技术有限公司 | Can attack protection biopsy method, device, computer equipment and storage medium |
| CN111008982A (en) * | 2020-03-10 | 2020-04-14 | 光子算数(北京)科技有限责任公司 | Data processing method based on photonic neural network chip and related device or equipment |
| CN111832714A (en) * | 2019-04-19 | 2020-10-27 | 上海寒武纪信息科技有限公司 | Operation method and device |
| CN112286694A (en) * | 2020-12-24 | 2021-01-29 | 瀚博半导体(上海)有限公司 | Hardware accelerator memory allocation method and system based on deep learning computing network |
| WO2021018199A1 (en) * | 2019-07-31 | 2021-02-04 | 瀚博半导体(上海)有限公司 | Convolutional neural network-based image processing method and apparatus |
| WO2021135572A1 (en) * | 2019-12-31 | 2021-07-08 | 深圳云天励飞技术股份有限公司 | Neural network convolution implementation method and convolution implementation apparatus, and terminal device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070104393A1 (en) * | 2005-11-10 | 2007-05-10 | Honeywell International Inc. | Method and apparatus for propagating high resolution detail between multimodal data sets |
| CN103679195A (en) * | 2013-12-02 | 2014-03-26 | 北京工商大学 | Method and system for classifying texture images on basis of local edge pattern |
| CN104537393A (en) * | 2015-01-04 | 2015-04-22 | 大连理工大学 | Traffic sign recognizing method based on multi-resolution convolution neural networks |
| CN105979174A (en) * | 2016-05-05 | 2016-09-28 | 清华大学 | Filtering network and image processing system |
| US20170069108A1 (en) * | 2015-04-23 | 2017-03-09 | Sungwook Su | Optimal 3d depth scanning and post processing |
-
2017
- 2017-10-19 CN CN201710983501.XA patent/CN107766932B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070104393A1 (en) * | 2005-11-10 | 2007-05-10 | Honeywell International Inc. | Method and apparatus for propagating high resolution detail between multimodal data sets |
| CN103679195A (en) * | 2013-12-02 | 2014-03-26 | 北京工商大学 | Method and system for classifying texture images on basis of local edge pattern |
| CN104537393A (en) * | 2015-01-04 | 2015-04-22 | 大连理工大学 | Traffic sign recognizing method based on multi-resolution convolution neural networks |
| US20170069108A1 (en) * | 2015-04-23 | 2017-03-09 | Sungwook Su | Optimal 3d depth scanning and post processing |
| CN105979174A (en) * | 2016-05-05 | 2016-09-28 | 清华大学 | Filtering network and image processing system |
Non-Patent Citations (3)
| Title |
|---|
| ROJAS: "《Neural Networks-A Systematic Introduction》", 《SPRINGER-VERLAG》 * |
| 杨眷玉: "《基于卷积神经网络的物体识别研究与实现》", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 谭贞刚: "《基于多分辨率卷积网络的交通标志检测与识别》", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111832714A (en) * | 2019-04-19 | 2020-10-27 | 上海寒武纪信息科技有限公司 | Operation method and device |
| CN111832714B (en) * | 2019-04-19 | 2023-11-17 | 上海寒武纪信息科技有限公司 | Operation method and device |
| CN110287767A (en) * | 2019-05-06 | 2019-09-27 | 深圳市华付信息技术有限公司 | Can attack protection biopsy method, device, computer equipment and storage medium |
| WO2021018199A1 (en) * | 2019-07-31 | 2021-02-04 | 瀚博半导体(上海)有限公司 | Convolutional neural network-based image processing method and apparatus |
| WO2021135572A1 (en) * | 2019-12-31 | 2021-07-08 | 深圳云天励飞技术股份有限公司 | Neural network convolution implementation method and convolution implementation apparatus, and terminal device |
| CN111008982A (en) * | 2020-03-10 | 2020-04-14 | 光子算数(北京)科技有限责任公司 | Data processing method based on photonic neural network chip and related device or equipment |
| CN112286694A (en) * | 2020-12-24 | 2021-01-29 | 瀚博半导体(上海)有限公司 | Hardware accelerator memory allocation method and system based on deep learning computing network |
| CN112286694B (en) * | 2020-12-24 | 2021-04-02 | 瀚博半导体(上海)有限公司 | Hardware accelerator memory allocation method and system based on deep learning computing network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107766932B (en) | 2021-06-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107766932A (en) | Image processing method and device based on neutral net | |
| CN103020947B (en) | A kind of mass analysis method of image and device | |
| Yu et al. | Towards efficient and scale-robust ultra-high-definition image demoiréing | |
| CN105320965B (en) | Sky based on depth convolutional neural networks composes united hyperspectral image classification method | |
| CN104834933B (en) | A kind of detection method and device in saliency region | |
| CN103247036B (en) | Many exposure images fusion method and device | |
| CN110070101A (en) | Floristic recognition methods and device, storage medium, computer equipment | |
| CN108416327A (en) | A kind of object detection method, device, computer equipment and readable storage medium storing program for executing | |
| CN109903221A (en) | Image oversubscription method and device | |
| CN107644209A (en) | Method for detecting human face and device | |
| CN110378235A (en) | A kind of fuzzy facial image recognition method, device and terminal device | |
| CN108197532A (en) | The method, apparatus and computer installation of recognition of face | |
| CN107749044A (en) | The pond method and device of image information | |
| CN110084161A (en) | A kind of rapid detection method and system of skeleton key point | |
| CN107403141A (en) | Method for detecting human face and device, computer-readable recording medium, equipment | |
| CN112464784A (en) | Distributed training method based on hybrid parallel | |
| CN104504086B (en) | The clustering method and device of Webpage | |
| CN109472199A (en) | A kind of method and device of image co-registration classification | |
| JP7358117B2 (en) | Image automatic classification method and system | |
| CN104484383B (en) | JS document handling methods and device | |
| CN108664933A (en) | The training method and its sorting technique of a kind of convolutional neural networks for SAR image ship classification, ship classification model | |
| CN110162657A (en) | A kind of image search method and system based on high-level semantics features and color characteristic | |
| CN107563357A (en) | Live dress ornament based on scene cut, which is dressed up, recommends method, apparatus and computing device | |
| CN111861867A (en) | Image background blurring method and device | |
| CN108230269A (en) | Grid method, device, equipment and storage medium are gone based on depth residual error network |
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 |