CN108921182A - The feature-extraction images sensor that FPGA is realized - Google Patents
The feature-extraction images sensor that FPGA is realized Download PDFInfo
- Publication number
- CN108921182A CN108921182A CN201811123578.0A CN201811123578A CN108921182A CN 108921182 A CN108921182 A CN 108921182A CN 201811123578 A CN201811123578 A CN 201811123578A CN 108921182 A CN108921182 A CN 108921182A
- Authority
- CN
- China
- Prior art keywords
- neural networks
- convolutional neural
- image
- layer
- fpga
- 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.)
- Pending
Links
- 238000000605 extraction Methods 0.000 title claims abstract description 34
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 76
- 230000005540 biological transmission Effects 0.000 claims abstract description 18
- 230000005284 excitation Effects 0.000 claims abstract description 10
- 238000000034 method Methods 0.000 claims description 19
- 230000008569 process Effects 0.000 claims description 18
- 239000000203 mixture Substances 0.000 claims description 4
- 238000012549 training Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 2
- 238000013135 deep learning Methods 0.000 claims 1
- 230000004927 fusion Effects 0.000 claims 1
- 238000012545 processing Methods 0.000 abstract description 22
- 238000012544 monitoring process Methods 0.000 abstract description 10
- 230000008901 benefit Effects 0.000 abstract description 5
- 238000001514 detection method Methods 0.000 abstract description 3
- 238000005265 energy consumption Methods 0.000 abstract description 3
- 210000002569 neuron Anatomy 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000001444 catalytic combustion detection Methods 0.000 description 4
- 239000000284 extract Substances 0.000 description 4
- 238000012546 transfer Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000036651 mood Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 230000029058 respiratory gaseous exchange Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- 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
- G06N3/045—Combinations of networks
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N25/00—Circuitry of solid-state image sensors [SSIS]; Control thereof
- H04N25/70—SSIS architectures; Circuits associated therewith
- H04N25/71—Charge-coupled device [CCD] sensors; Charge-transfer registers specially adapted for CCD sensors
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N25/00—Circuitry of solid-state image sensors [SSIS]; Control thereof
- H04N25/70—SSIS architectures; Circuits associated therewith
- H04N25/76—Addressed sensors, e.g. MOS or CMOS sensors
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Signal Processing (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention discloses a kind of FPGA feature-extraction images sensor, is made of universal CCD/cmos image sensor, ARM controller, FPGA convolutional neural networks, image buffer storage unit, wireless data transmission mould group and network transmission mould group.One layer of convolutional layer, one layer of ReLU excitation layer and one layer of pond layer building part convolutional neural networks are disposed on the FPGA convolutional neural networks, convolutional layer is used for the dimension-reduction treatment of characteristic image for extracting characteristics of image, pond layer.One complete convolutional neural networks is divided into two parts, and another part is deployed in cloud front end, constitutes complete convolutional neural networks together with the convolutional neural networks of FPGA deployment.The complete convolutional neural networks can be trained to, and be used for picture control, detection object and target tracking.The feature-extraction images sensor that this FPGA is realized has low in energy consumption, the fast advantage of processing speed, when for special scenes monitoring, can reduce the demand to transmission bandwidth, can also avoid directly transmitting privacy brought by image or problem of divulging a secret.
Description
Technical field
The present invention relates to image procossing, convolutional neural networks field, the feature-extraction images realized more particularly to FPGA
Sensor.
Background technique
Image characteristics extraction, on the one hand can reduce the data volume for the image of being transmitted, and on the other hand can solve direct prison
The problem of control image, caused privacy is exposed and divulges a secret.Especially in family endowment, monitoring to senior activity, to patient's
Respiration monitoring control, mood monitoring etc., direct monitoring characteristics of image, rather than direct monitoring clear image, application are easier.
FPGA（Field-Programmable Gate Array）As Special Purpose Programmable circuit, all in many fields
Extensive use has been arrived, advantage and huge prospect are had more especially in terms of Digital Signal Processing and image procossing.
Existing convolutional neural networks scheme based on CPU or GPU extracts characteristic image, cannot be considered in terms of real-time, power consumption with
And the requirement of portability, and FPGA has the advantages that powerful parallel processing capability and super low-power consumption, part or whole can be rolled up
Product neural network is deployed on FPGA.
Summary of the invention
The present invention discloses a kind of FPGA feature-extraction images sensor, is controlled by universal CCD/cmos image sensor, ARM
Device, FPGA convolutional neural networks, image buffer storage unit, wireless data transmission mould group and network transmission mould group composition.Described
One layer of convolutional layer, one layer of ReLU excitation layer and one layer of pond layer building part convolutional Neural net are disposed on FPGA convolutional neural networks
Network, convolutional layer are used for the dimension-reduction treatment of characteristic image for extracting characteristics of image, pond layer.One complete convolutional neural networks
It is divided into two parts, another part is deployed in cloud front end, constitutes together with the convolutional neural networks of FPGA deployment complete
Convolutional neural networks.The complete convolutional neural networks can be trained to, and be chased after for picture control, detection object and target
Track.The feature-extraction images sensor that this FPGA is realized has low in energy consumption, the fast advantage of processing speed, is being used for specific field
When scape monitors, the demand to transmission bandwidth can be reduced, can also avoid directly transmitting privacy brought by image or divulging a secret asking
Topic.
One of preferably, the feature-extraction images sensor is by universal CCD/cmos image sensor, ARM
Controller, FPGA convolutional neural networks, image buffer storage unit, wireless data transfer module and network transmission module composition.
Preferably, the general image sensor is the CCD and cmos image sensor generally used on the market at present,
Its image definition determines the size of image, is determined by concrete application demand, for example, when acquiring vision signal, size
For D1：720x576；CIF：352x240；HDTV：1920x1080 etc..
Preferably, the described ARM controller control data input, caching, FPGA and output wireless data mould group or
It is flowed in the form of data flow in network modules, data is transferred to FPGA kernel by DDR in the form of data flow, controlled
Data flow shuttle with each CNN layers, using its parallel organization and high-performance assembly line, complete entire feature-extraction images at a high speed
The process flow of sensor.
Preferably, the FPGA convolutional neural networks unit is an image characteristics extraction convolutional neural networks, the volume
Product neural network is a part of complete convolutional neural networks.We are two parts, a part of portion a complete CNN points
Administration is in the FPGA convolutional neural networks unit, and another part is deployed in cloud front end, which can lead to
Big data training is crossed, image special characteristic is completed and extracts, meets specific image monitoring demand.
Preferably, after the complete CNN can have been trained, then the CNN is divided into two parts, by front end input, convolution
Processing part is directly deployed on FPGA convolutional neural networks unit, does not need retraining.
Preferably, the image buffer storage unit is used to cache intermediate image and character image data.
Preferably, the wireless data transfer module, for the character image data for handling convolutional neural networks well
It is for further processing again by wireless network transmissions to front end convolutional neural networks.The wireless data transmission mould group, can be with
It is 4G/5G mobile data transfer mould group, character image data will be transmitted by mobile data network；It is also possible to WiFi data biography
Defeated mould group, character image data will be transmitted by WiFi network.
Preferably, the network transmission module can be Ethernet transmission module, for handling convolutional neural networks
Good character image data is for further processing by network transmission to front end convolutional neural networks again.
Preferably two, described at least one convolutional layer of FPGA convolutional neural networks, a ReLU excitation
Layer and at least one maximum pond layer.
Preferably, the convolutional layer is used to handle by the picture signal of the general image sensor acquisition, passes through
Process of convolution extracts its characteristic image.
Further, the excitation layer will treated that characteristic image does Nonlinear Processing to the convolutional layer, and increases
The neuron number being activated.
Further, the maximum pond layer generallys use 2 × 2 convolution kernel processing, its maximum value is taken to indicate convolution knot
Fruit achievees the purpose that the characteristic image dimension after coming out reducing half.
Preferably three, the complete convolutional neural networks can choose existing classics in specific application
Convolutional neural networks, e.g., Alex-Net, Googl-Net, ResNet, DenseNet and its these exemplary convolution neural network framves
The convolutional neural networks for prolonging Shen and complementary building of structure.
Preferably, it when object identification and target tracking task are handled, can use such as, R-CNN（Convolution feature）,
SPP-NET/Fast R-CNN（Convolution feature）,Faster R-CNN（Convolution feature）,YOLO (v1 & v2)（Convolution feature）,
SSD（Convolution feature）Deng complicated convolutional neural networks.
Preferably, the complete convolutional neural networks are divided into two parts, front end processing portion, i.e., simple part, portion
The FPGA convolutional neural networks unit in the FPGA feature-extraction images sensor is affixed one's name to, another part is deployed in front end, due to
Front end can handle complicated image classification and object using the convolutional neural networks processing system for having supercomputing capability
Physical examination survey and tracking task.
Further, since the complicated part of the completely convolutional neural networks is deployed in front end, cloud can be regarded as
Service system, for all feature extractions for deploying same part convolutional neural networks in FPGA convolutional neural networks unit
Imaging sensor is used in conjunction with cloud part convolutional neural networks, completes the classification monitoring to image, the detection to object and tracking
Task.Data are transmitted through network implementations.
The beneficial effects of the invention are as follows：
Divide convolutional neural networks in FPGA achievement unit, e.g., one layer of convolutional network makees feature extraction, one layer of excitation layer and one layer of pond
Change layer and does dimension-reduction treatment.Another part convolutional neural networks are then disposed beyond the clouds, two parts convolutional neural networks have been constituted
Whole convolutional neural networks.Real-time processing difficulty is greatly reduced, the simplification of feature-extraction images sensor, Yi Shi have both been met
Existing requirement, is also equipped with low in energy consumption, and processing speed is fast, the good advantage of real-time.Meanwhile by disposing complicated convolution beyond the clouds
That a part of neural network constitutes complete convolutional neural networks with that a part of feature-extraction images sensor side deployment, can
With application typical convolutional neural networks processing monitoring image classification, object identification and target tracking intensive tasks can also be created
New increasingly complex integrated treatment framework, provides treatment effeciency and precision.
Specific characteristics of image is extracted from the original image of general image sensor acquisition signal, to reduce biography
Defeated, storage entire image bandwidth and memory space.Privacy is exposed for being unwilling, or is unwilling to be seen the spy of clear image
Different monitoring demand, provides a kind of special characteristics of image monitor mode.
Detailed description of the invention
Fig. 1 is the FPGA feature-extraction images sensor structure of the preferred embodiment of the present invention；
Fig. 2 is the convolutional neural networks that the preferred embodiment of the present invention is disposed in FPGA convolutional neural networks unit；
Fig. 3 is that the Alex-Net convolutional neural networks of the preferred embodiment of the present invention are divided into two parts deployment example；
Fig. 4 is the process flow of the feature-extraction images sensor of the preferred embodiment of the present invention.
Specific embodiment
Presently in connection with attached drawing and preferred embodiment, the present invention is described in further detail.These attached drawings are simplified
Schematic diagram, the basic structure of the invention will be illustrated schematically only, therefore it only shows the composition relevant to the invention.
Fig. 1 is the FPGA feature-extraction images sensor structure of the preferred embodiment of the present invention, including：101 universal CCDs/
Cmos image acquisition unit, here, it is preferred that cmos image acquisition unit；102 be ARM controller, in this as Data flow direction
Control, storage control, output control etc. handle each step program；106 be FPGA convolutional neural networks feature extraction unit,
A part of convolutional neural networks function of completing a complete convolutional neural networks, for example, doing characteristics of image with one layer of convolutional layer
It extracts, makees image Nonlinear Processing with one layer of ReLU excitation layer, make image dimension-reduction treatment with one layer of maximum pond layer；105 be figure
As cache unit, input picture or intermediate features image data are in the unit caches；104 be Ethernet interface unit, after processing
Character image data give front end by the network of the network interface connection；103 be wireless transport module, treated characteristic image
Data give front end by the wireless transport module, which can be 4G/5G mobile data module, are also possible to
WiFi wireless transport module, the former sends the data to front end by mobile data network, and the latter is sent data by WiFi network
Toward front end.
Fig. 2 is the convolutional neural networks that the preferred embodiment of the present invention is disposed in FPGA convolutional neural networks unit, and 201 be
One layer of convolutional layer, in the implementation column, taking input picture size is 227 X 227, with 11 X 11 of convolution kernel, step-length S=4, volume
Product core carries out process of convolution, and output characteristic pattern size is 55 X 55；202 be excitation layer, is 55 X by the characteristic pattern size of input
55 signal carries out Nonlinear Processing, and the model after realization is sparse can preferably excavate correlated characteristic, is fitted training data；
203 be maximum pond layer, and in the implementation column, pond layer uses 3 X, 3 convolution kernel, and the filter of step-length S=2, it is defeated that treated
Characteristic pattern size is 27 X 27 out, significantly reduces the dimension of output characteristic pattern.
Fig. 3 is that Alex-Net convolutional neural networks are divided into two parts deployment examples, Wo Mencong by the preferred embodiment of the present invention
One typical Alex-Net convolutional neural networks is divided into two parts by 310 points, left-hand component, 301,302 be deployed in it is described
On feature-extraction images sensor, right-hand component, 303,304,305,306,307,308,309 are deployed in cloud front end.301 be defeated
Entering size is 227 X, 227 X, 3 input picture；With 11 X 11, after 4,96 filters of step-length do process of convolution, output
For 55 X, 55 X, 96 characteristic images；Again with core size be 3 X 3, step-length 2, filter do pondization processing, feature will be exported
Figure is reduced to 27 X, 27 X, 96 characteristic patterns, as illustrated at 302, described 301,302 part convolutional neural networks is deployed in described
Feature-extraction images sensor on.
The another part of the Alex-Net convolutional neural networks of described Fig. 3 contains, and 303 be the 2nd convolutional layer, including makees
5 X 5, the process of convolution that step-length is 1 meet relu2, then with 3 X 3, the pondization that step-length is 2 is handled, and finally exports 13 X, 13 X 256
A characteristic pattern；304 be the 3rd convolutional layer, 3 X 3, the process of convolution that step-length is 1, output 13 X, 13 X, 384 characteristic patterns；305 are
4th convolutional layer, 3 X 3, the process of convolution that step-length is 1 export 13 X, 13 X, 384 new characteristic patterns；306 be the 5th convolutional layer, 3
X 3, the process of convolution that step-length is 1 export 13 X, 13 X, 384 characteristic patterns, handle using pond layer, dimensionality reduction to 6 X, 6 X
256 characteristic pattern outputs；307 be full articulamentum FC6, is first exported with 4096 neuron connection 6 X, 6 X, 256 characteristic patterns, then
Dropout loses some nodal informations at random, obtains 4096 new neurons；308 be FC7, is the full connection similar with FC6
Layer, there is 4096 neurons；309 be also full articulamentum, completes classification task, 1000 nerves with 1000 neurons
Member is connect with 4096 neurons of FC7, is then passed through Gaussian filter, is obtained the value of 1000 float types, that is, that predicts can
It can property.
Fig. 4 is the process flow of the feature-extraction images sensor of the preferred embodiment of the present invention；Initially enter process S401
Image is acquired, process S402 is entered back into and caches acquired image；Then process S403 is entered back into, the image of caching is given
FPGA unit does process of convolution, which is to make feature extraction to original image；Then process S404 is entered back into, to finishing feature
Characteristic pattern after extraction carries out Nonlinear Processing, the processing of ReLU excitation layer；Process S405 is entered back into, that treated is special to excitation layer
Sign figure carries out maximum pondization processing；Into process S406 by the data buffer storage handled well to cache unit；Finally enter back into data
Wireless data transmission or wired data transfer may be selected, before the characteristic pattern Jing Guo feature extraction is sent in transmission flow S407
The another part convolutional neural networks of end administration carry out further image classification, object identification or target tracking processing.
It is enlightenment with above-mentioned embodiment according to the present invention, through the above description, relevant staff completely may be used
Without departing from the scope of the technological thought of the present invention', to carry out various changes and amendments.The technical model of this invention
It encloses and is not limited to the contents of the specification, it is necessary to which the technical scope thereof is determined according to the scope of the claim.
Claims (8)
1. the feature-extraction images sensor that a kind of FPGA is realized, which is characterized in that it is controlled by general image sensor, AMR
The composition such as device, FPGA image characteristics extraction unit, image buffer storage unit, wireless transport module, network transmission module：
The general image sensor is CCD or cmos image sensor, is recommended using cmos image sensor convenient for integrated one
Bodyization and reduction volume：
Described image sensor output digital image signal；
The AMR controller is direct in input, buffer, FPGA and output for controlling image and character image data stream
Flowing, controls entire process flow by runs software；
Described image cache unit is used to keep in digital picture and characteristic image signal；
The wireless transport module is used to through the wireless network transmissions character image data and control signal；
The network transmission module is used to through the gauze network transmission feature image data and control signal.
2. image characteristics extraction unit as described in claim 1, which is characterized in that use FPGA as image characteristics extraction unit
Hardware, the hardware operation convolutional neural networks are as feature extraction.
3. convolutional neural networks as claimed in claim 2, which is characterized in that be a complete depth study convolutional neural networks
A part, the part include but are not limited to, one layer of convolutional layer, one layer of ReLU excitation layer, one layer of maximum pond layer.
4. complete depth as claimed in claim 3 learns convolutional neural networks, which is characterized in that it is divided into two parts, one
Divide on the FPGA for being deployed in the feature-extraction images sensor, another part convolutional neural networks are deployed in the operation of cloud front end
On platform.
5. complete depth as claimed in claim 3 learns convolutional neural networks, which is characterized in that the convolutional neural networks
It is the convolutional neural networks of typical deep learning convolutional neural networks or its deformation and fusion, is used for image classification, object
Identification or target tracking.
6. complete depth as claimed in claim 3 learns convolutional neural networks, which is characterized in that according to task and data training
Afterwards, just it is divided into two parts to dispose respectively.
7. complete depth as claimed in claim 3 learns convolutional neural networks, which is characterized in that, can after the completion of disposing respectively
To be trained as a whole.
8. another part convolutional neural networks as claimed in claim 4, which is characterized in that be deployed as cloud system, can give more
A feature-extraction images sensor for being deployed with same section convolutional neural networks in its image characteristics extraction unit is used in conjunction with.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811123578.0A CN108921182A (en) | 2018-09-26 | 2018-09-26 | The feature-extraction images sensor that FPGA is realized |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811123578.0A CN108921182A (en) | 2018-09-26 | 2018-09-26 | The feature-extraction images sensor that FPGA is realized |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108921182A true CN108921182A (en) | 2018-11-30 |
Family
ID=64409201
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811123578.0A Pending CN108921182A (en) | 2018-09-26 | 2018-09-26 | The feature-extraction images sensor that FPGA is realized |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108921182A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109447056A (en) * | 2018-12-07 | 2019-03-08 | 苏州米特希赛尔人工智能有限公司 | The feature-extraction images sensor that DSP is realized |
| CN109871939A (en) * | 2019-01-29 | 2019-06-11 | 深兰人工智能芯片研究院（江苏）有限公司 | A kind of image processing method and image processing apparatus |
| CN110400250A (en) * | 2019-07-29 | 2019-11-01 | 杭州凝眸智能科技有限公司 | Intelligent image preprocess method and system based on AI |
| CN110636221A (en) * | 2019-09-23 | 2019-12-31 | 天津天地人和企业管理咨询有限公司 | System and method for super frame rate of sensor based on FPGA |
| CN111669527A (en) * | 2020-07-01 | 2020-09-15 | 浙江大学 | Convolution operation framework in CMOS image sensor |
| CN111860540A (en) * | 2020-07-20 | 2020-10-30 | 深圳大学 | Neural network image feature extraction system based on FPGA |
| CN112101178A (en) * | 2020-09-10 | 2020-12-18 | 电子科技大学 | Intelligent SOC terminal assisting blind people in perceiving external environment |
| CN112330524A (en) * | 2020-10-26 | 2021-02-05 | 沈阳上博智像科技有限公司 | Device and method for quickly realizing convolution in image tracking system |
| CN113727014A (en) * | 2020-05-25 | 2021-11-30 | 西克股份公司 | Camera and method for processing image data |
| EP4231200A1 (en) * | 2022-02-17 | 2023-08-23 | Plantronics, Inc. | Distributed machine learning inference |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106250939A (en) * | 2016-07-30 | 2016-12-21 | 复旦大学 | System for Handwritten Character Recognition method based on FPGA+ARM multilamellar convolutional neural networks |
| CN107944548A (en) * | 2017-11-30 | 2018-04-20 | 睿视智觉（深圳）算法技术有限公司 | A kind of FPGA realizes the design method of convolutional layer |
| CN207458128U (en) * | 2017-09-07 | 2018-06-05 | 哈尔滨理工大学 | A kind of convolutional neural networks accelerator based on FPGA in vision application |
-
2018
- 2018-09-26 CN CN201811123578.0A patent/CN108921182A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106250939A (en) * | 2016-07-30 | 2016-12-21 | 复旦大学 | System for Handwritten Character Recognition method based on FPGA+ARM multilamellar convolutional neural networks |
| CN207458128U (en) * | 2017-09-07 | 2018-06-05 | 哈尔滨理工大学 | A kind of convolutional neural networks accelerator based on FPGA in vision application |
| CN107944548A (en) * | 2017-11-30 | 2018-04-20 | 睿视智觉（深圳）算法技术有限公司 | A kind of FPGA realizes the design method of convolutional layer |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109447056A (en) * | 2018-12-07 | 2019-03-08 | 苏州米特希赛尔人工智能有限公司 | The feature-extraction images sensor that DSP is realized |
| CN109871939A (en) * | 2019-01-29 | 2019-06-11 | 深兰人工智能芯片研究院（江苏）有限公司 | A kind of image processing method and image processing apparatus |
| CN110400250A (en) * | 2019-07-29 | 2019-11-01 | 杭州凝眸智能科技有限公司 | Intelligent image preprocess method and system based on AI |
| CN110636221A (en) * | 2019-09-23 | 2019-12-31 | 天津天地人和企业管理咨询有限公司 | System and method for super frame rate of sensor based on FPGA |
| CN113727014A (en) * | 2020-05-25 | 2021-11-30 | 西克股份公司 | Camera and method for processing image data |
| US11941859B2 (en) | 2020-05-25 | 2024-03-26 | Sick Ag | Camera and method for processing image data |
| CN111669527B (en) * | 2020-07-01 | 2021-06-08 | 浙江大学 | Convolution operation circuit in CMOS image sensor |
| CN111669527A (en) * | 2020-07-01 | 2020-09-15 | 浙江大学 | Convolution operation framework in CMOS image sensor |
| CN111860540A (en) * | 2020-07-20 | 2020-10-30 | 深圳大学 | Neural network image feature extraction system based on FPGA |
| CN111860540B (en) * | 2020-07-20 | 2024-01-12 | 深圳大学 | Neural network image feature extraction system based on FPGA |
| CN112101178A (en) * | 2020-09-10 | 2020-12-18 | 电子科技大学 | Intelligent SOC terminal assisting blind people in perceiving external environment |
| CN112330524A (en) * | 2020-10-26 | 2021-02-05 | 沈阳上博智像科技有限公司 | Device and method for quickly realizing convolution in image tracking system |
| EP4231200A1 (en) * | 2022-02-17 | 2023-08-23 | Plantronics, Inc. | Distributed machine learning inference |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108921182A (en) | The feature-extraction images sensor that FPGA is realized | |
| TWI746865B (en) | Deep vision processor | |
| US12062158B2 (en) | Image denoising method and apparatus | |
| US10785745B2 (en) | Scaling multi-core neurosynaptic networks across chip boundaries | |
| Farabet et al. | Comparison between frame-constrained fix-pixel-value and frame-free spiking-dynamic-pixel convnets for visual processing | |
| CN106529517B (en) | Image processing method and image processing equipment | |
| CN109871510B (en) | Two-dimensional convolution operation processing method, system, equipment and computer storage medium | |
| CN108898191A (en) | Convolutional neural networks feature-extraction images sensor | |
| Wu | Neuromorphic vision chips | |
| CN108416436A (en) | The method and its system of neural network division are carried out using multi-core processing module | |
| CN108154494A (en) | A kind of image fusion system based on low-light and infrared sensor | |
| DE102020108640A1 (en) | Voice-controlled camera with AI scene detection for precise focusing | |
| WO2021056974A1 (en) | Vein recognition method and apparatus, device, and storage medium | |
| CN105243399B (en) | A kind of method and apparatus that realizing image convolution, the method and apparatus for realizing caching | |
| CN109447056A (en) | The feature-extraction images sensor that DSP is realized | |
| CN208922294U (en) | The feature-extraction images sensor that FPGA is realized | |
| WO2022179599A1 (en) | Perceptual network and data processing method | |
| CN108184062B (en) | High-speed tracking system and method based on multi-level heterogeneous parallel processing | |
| US20190325250A1 (en) | Reconfigurable 3D Pixel-Parallel Neuromorphic Architecture for Smart Image Sensor | |
| Iturbe et al. | Nimbleai: Towards neuromorphic sensing-processing 3d-integrated chips | |
| CN107609646A (en) | A kind of residual error network implementation approach, system, equipment and computer-readable storage medium | |
| CN109542513B (en) | Convolutional neural network instruction data storage system and method | |
| Hou et al. | Real‐time defect detection method based on YOLO‐GSS at the edge end of a transmission line | |
| Liu et al. | Real-time object recognition based on NAO humanoid robot | |
| CN115050076A (en) | Mask detection system design method for embedded improved Yolov5s |
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 |