CN110427821A - A kind of method for detecting human face and system based on lightweight convolutional neural networks - Google Patents
A kind of method for detecting human face and system based on lightweight convolutional neural networks Download PDFInfo
- Publication number
- CN110427821A CN110427821A CN201910567225.8A CN201910567225A CN110427821A CN 110427821 A CN110427821 A CN 110427821A CN 201910567225 A CN201910567225 A CN 201910567225A CN 110427821 A CN110427821 A CN 110427821A
- Authority
- CN
- China
- Prior art keywords
- neural networks
- convolutional neural
- lightweight
- network
- block
- 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
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention belongs to Computer Vision Detection Technique fields, and in particular to a kind of method for detecting human face and system based on lightweight convolutional neural networks.This programme is by model optimization into light weight model firstly the need of the critical issue of solution, it reduces calculation amount, improve arithmetic speed, secondly the accuracy that is light-weighted while needing to guarantee face detection model of network is realized, therefore balance network lightweight and accuracy rate, it would be desirable to be able to which studying on the basis of lightweight Face datection network model and how improving the accuracy rate of algorithm is critical issue that this programme solves.This case has some superiority in terms of detection accuracy, model size, detection speed synthesis, which ensure that certain precision compared with the Face datection algorithm based on VGG16, detection speed, in terms of it is more advantageous.
Description
Technical field
The invention belongs to Computer Vision Detection Technique fields, and in particular to a kind of based on lightweight convolutional neural networks
Method for detecting human face and system.
Background technique
Test problems are a traditional problems of computer vision, and wherein Face datection is most important one one again.
With increasingly mature, the development of Face datection algorithm in addition of computer algebra method, Face datection starts in society now
Play the part of more and more important role, for example airport security, company are checked card, public arena monitors, electronic equipment exempts from close entrance etc.,
Face datection has its ample scope for abilities.
Existing method for detecting human face can be roughly divided into two types.One kind by template and is needed based on template matching
The picture of detection compares, special to face Invariance feature such as contour feature, color to determine whether the picture is face
Sign, textural characteristics etc. establish template, judge whether the region includes people by the similarity of calculating input image and face template
Face, the algorithm based on template matching is dependent on established face template prior in static scene, in the big dynamic of dimensional variation
In scene, the effect is relatively poor.Another kind of is to be calculated the classifier in the feature and machine learning of manual construction based on characteristic statistics
Method such as artificial neural network and Adaboost etc. combine, and statistical learning are carried out in a large amount of face samples, in testing image
Whether a certain region can carry out correct classification by classifier to achieve the purpose that Face datection, and the detection based on characteristic statistics is calculated
Method is limited by the feature operator of engineer, cannot sufficiently get the characterization of higher in image.
Convolutional neural networks are used for Face datection problem after being succeeded in image classification problem quickly, in precision
On significantly surmount before AdaBoost frame, currently had some high-precisions, efficient algorithm.
Face datection network technology based on deep learning made breakthrough progress, but in real-time and accuracy
Balance on still there are many insufficient.With the increase of convolution layer depth, it can extract better feature, and detection accuracy can mention
Height, but calculation amount is also bigger, it is meant that detection slows, and such as SSD frame and thus derives various improved SSD
Model, although the SSD model based on VGG-Net is improved in detection accuracy, since its model calculation amount is larger, inspection
Degree of testing the speed is slower, it is difficult to meet real-time application demand.If you need to realize that real-time face detects, then need to carry out beta pruning to neural network
Optimization, this can sacrifice certain detection accuracy, therefore how accomplish to be able to maintain detection accuracy, while again can compression network, make to examine
Degree of testing the speed becomes faster, and is that current face detects the technical problem underlying faced.
Summary of the invention
In order to solve technological deficiency existing in the prior art, the invention proposes one kind to be based on lightweight convolutional Neural net
The method for detecting human face and system of network.
The invention is realized by the following technical scheme:
A kind of method for detecting human face based on lightweight convolutional neural networks comprising step:
Face database is established in S1, data processing, carries out image preprocessing, generates training sample；
S2, the sample based on input carry out feature extraction using lightweight convolutional neural networks；The lightweight volume
Product neural network includes the convolutional layer that convolution kernel size is respectively 3x3 and 1x1；
S3 is merged the different characteristic layer of lightweight convolutional neural networks based on Fusion Features module；
S4, anchor point frame are chosen；
S5 exports Analysis On Multi-scale Features figure；
S6, the mapping matching of face candidate area；
S7, face classification return；
S8 establishes non-maxima suppression constraint；
S9, output test result.
Further, in the step S2, the core network of the lightweight convolutional neural networks CNN includes: successively
The 1 A block and 5 B block of connection, wherein the port number of convolution is set as 16 in A block, each B block's
It is 16,32,64,128,128 that port number, which is set gradually,.
Further, the A block is 3 × 3 by convolution kernel size, and the convolutional layer realization that step-length is 2 is down-sampled,
Addition convolution kernel size is 1 × 1 afterwards, and the convolutional layer that step-length is 1 achievees the purpose that input feature vector dimensionality reduction, behind each convolutional layer
It is sequentially connected the Batch Normalization and nonlinear function ReLU of the nonlinear characteristic for improving network.
Further, the B block realizes down-sampled, step-length by the maximum pond layer that convolution kernel size is 2 × 2
It is 2, has then been sequentially connected the convolutional layer that convolution kernel size is 3 × 3,1 × 1,3 × 3, each convolutional layer step-length is 1, and below
It is sequentially connected the nonlinear characteristic Batch Normalization and nonlinear function ReLU for improving network.
Further, in the step S3, the Fusion Features module include convolution kernel size be respectively 3 × 3 and 1 ×
1 convolutional layer, and step-length is 1, is up-sampled using deconvolution, and by low-level feature figure by the way of element summation
It is fused together with transformed high-level characteristic figure.
The present invention also provides a kind of face detection systems based on lightweight convolutional neural networks, are with SSD detection framework
Basis, in feature extraction using designed light weight convolution grade neural network as core network comprising:
Data processing module carries out image preprocessing, generates training sample for establishing face database；
Light weight convolution grade neural network module, the sample based on input are carried out using lightweight convolutional neural networks
Feature extraction；The lightweight convolutional neural networks include the convolutional layer that convolution kernel size is respectively 3x3 and 1x1；
Fusion Features module merges for the different characteristic layer to core network, and exports Analysis On Multi-scale Features figure；
Anchor point frame chooses module, for the ratio of width to height of anchor point frame to be arranged.
Preferably, the lightweight convolutional neural networks include: sequentially connected 1 A block and 5 B block,
Wherein, the port number of convolution is set as 16 in A block, the port number of each B block sets gradually as 16,32,64,128,
128。
Preferably, the A block is 3 × 3 by convolution kernel size, and the convolutional layer realization that step-length is 2 is down-sampled, thereafter
Adding convolution kernel size is 1 × 1, and the convolutional layer that step-length is 1 achievees the purpose that input feature vector dimensionality reduction, behind each convolutional layer according to
The secondary Batch Normalization and nonlinear function ReLU connected for improving the nonlinear characteristic of network.
Preferably, the B block realizes down-sampled that step-length is by the maximum pond layer that convolution kernel size is 2 × 2
2, it is then sequentially connected the convolutional layer that convolution kernel size is 3 × 3,1 × 1,3 × 3, each convolutional layer step-length is 1, and below
It is sequentially connected the nonlinear characteristic Batch Normalization and nonlinear function ReLU for improving network.
Preferably, the Fusion Features module includes the convolutional layer that convolution kernel size is respectively 3 × 3 and 1 × 1, and step-length
It is 1, is up-sampled using deconvolution, and by low-level feature figure and transformed high-rise spy by the way of element summation
Sign figure is fused together.
Compared with prior art, the present invention at least has the following beneficial effects or advantage: this programme is from design lightweight
Face datection network set out, on the basis of original deep learning neural network model theoretical research, analysis Face datection mind
Key restriction factors through network model in terms of size and operation efficiency, by the analysis to key restriction factors, to network
It optimizes；Face datection is used for as core network using designed lightweight network in SSD frame, reduces model
Memory space, to the occupancy of calculator memory when reducing model foundation and initializing；Multiple convolution are realized on convolutional network
The fusion of layer feature, is used for lightweight Face datection network for the Fusion Features module of design, to allow algorithm in precision aspect
It improves；Compared with traditional SSD detection algorithm, it is contemplated that the particularity of face shape and ratio, this programme is detection
The ratio of width to height of frame is set as 1, reduces complicated network query function, improves the operation efficiency of deep learning neural network.
Detailed description of the invention
The present invention is described in further details below with reference to attached drawing；
Fig. 1 (a) is A block structure chart of the present invention；
Fig. 1 (b) is B block structure chart of the present invention；
Fig. 1 (c) is Face datection core network structure chart of the invention；Fig. 2 is Fusion Features module map of the invention；
Fig. 3 is Face datection flow chart of the invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are some of the embodiments of the present invention, instead of all the embodiments.Based on this hair
Embodiment in bright, every other implementation obtained by those of ordinary skill in the art without making creative efforts
Example, shall fall within the protection scope of the present invention.
The convolutional neural networks model hierarchy of mainstream is deep, parameter is big, and calculation amount is very huge, cannot achieve real-time detection,
It can be seen from the experiment that influencing time-consuming factor in convolutional calculation is broadly divided into two aspects: the dimension and convolution kernel of input feature vector figure
Number.The wide Gao Yue great of input feature vector figure, it is more that convolution kernel slides the number calculated needed in the horizontal and vertical directions；It is defeated
The port number for entering characteristic pattern is bigger, and convolution kernel needs linear combination to calculate parameter when locally connection calculates each time is more；
In forward direction operation, input feature vector needs to combine calculating with each neuron in convolutional neural networks, and the convolution kernel for including is got over
More, when this layer calculates and the iterative calculation amount of convolution kernel is bigger.
By being analyzed above it is found that convolutional neural networks model can be by increasing the depth and each layer of convolution kernel of model
Number improves model efficiency, but bring is that calculation amount increases problem；If merely use the image of small in resolution as
Input, information content is very little to make network poor fitting again.For this problem, this programme is based on SSD detection framework, in spy
Using designed lightweight neural network when sign is extracted, when guaranteeing certain accuracy of identification, it is possible to reduce network rule
Mould (parameter amount, calculation amount), the convolution kernel that the convolution kernel partial replacement of 3x3 is 1x1 by designed network make parameter amount contract in this way
It is 9 times small, but the validity of the feature of the extraction of convolutional network may be influenced in this way, in order to ensure recognition efficiency, only do
Partial replacement；Network reduces the channel sized of input data simultaneously, can reduce parameter amount in this way.Network backbone part knot
Shown in structure such as Fig. 1 (c), A block (shown in such as Fig. 1 (a)) is 3 × 3 by convolution kernel size, and the convolutional layer that step-length is 2 is realized
It is down-sampled, it is used to improve the nonlinear characteristic of network followed by Batch Normalization and nonlinear function ReLU, then
Achieve the purpose that input feature vector dimensionality reduction by adding 1 × 1 convolutional layer below, to optimize the calculating time of model, A block
The port number of middle convolution is set as 16.B block (shown in such as Fig. 1 (b)) realizes down-sampled, step by 2 × 2 maximum pond layers
A length of 2, it has then been sequentially connected the convolutional layer that convolution kernel size is 3 × 3,1 × 1,3 × 3, each convolutional layer step-length is 1, behind
Batch Normalization and nonlinear function ReLU are all connected with to improve non-linear expression's ability of model.Entirely leading
In dry network, 5 B block have been used altogether, it is 16,32,64,128,128 that port number, which is set gradually,.
Different detection branches are independent from each other in SSD structure, therefore are easy to appear same object by different size of inspection
Survey the problem of frame detected simultaneously.This case can increase the connection of different layers in the way of Fusion Features, reduce repeat block
Appearance, on the other hand, the Feature Semantics information of low layer is fewer, information of more paying attention to detail, and target position is accurate；High-rise spy
It is relatively abundanter to levy semantic information, but target position is relatively coarse, utilizes low-level feature high score simultaneously by Fusion Features module
The high semantic information of resolution and high-level characteristic, the feature by merging these different layers achievees the effect that prediction, and prediction is
It is individually carried out on each fused characteristic layer, the detection accuracy of small size face can be improved.
Fusion Features module is as shown in Fig. 2, the port number of Conv3 × 3, Conv1 × 1 is 256, and step-length is 1, is made
It is up-sampled with deconvolution, and by low-level feature figure and transformation in such a way that element sums (element-wise sum)
High-level characteristic figure afterwards is fused together.The different characteristic layer of core network is melted by designed Fusion Features module
It closes, the characteristic pattern of corresponding different scale can be exported, auxiliary convolutional layer that is last and adding after core network exports together
As the candidate frame of every class detection block, the purpose for improving detection accuracy is realized, while guaranteeing lesser model and higher fortune again
Line efficiency.
This case is to can achieve multiple scale detecting using detection block of different sizes by the different characteristic figure of single network
Effect, while calculation amount can be reduced with shared parameter, improve detection speed.Theoretically the receptive field of detection block is uniform,
In fact, influence of the intermediate input to output is heavier, a kind of similar Gaussian Profile.According to the inspection of above-mentioned theory and SSD
Frame design method is surveyed, the detection block for meeting multiple dimensioned Face datection is provided in this technology.And pass through the analysis to human face data,
Face is all largely square, and the characteristics of due to face shape, the ratio of width to height of detection block is set as 1 by this case, in this way can be with
The quantity that network query function obtains is reduced, to improve detection speed.
Face datection process is as shown in figure 3, this case based on SSD detection framework, uses designed in feature extraction
Lightweight neural network as core network, greatly reduce neural network parameter；Secondly, we roll up the lightweight of design
Multiple convolution features are merged in product network, and then the precision of algorithm are allowed to improve, and maintain that SSD's is multiple dimensioned pre-
Survey feature improves the accuracy rate of detection, so that it is guaranteed that the accuracy rate of network Face datection.The network and the face based on VGG16
Detection algorithm is compared, and has certain advantage in the composite factors such as arithmetic accuracy, detection speed, model size.In detection block
Setting on, the shape of general object is indefinite, so SSD is provided with the anchor point frame of different proportion, but due to face
Particularity, the ratio of Generic face are fixed.In order to reduce unnecessary calculation amount, this is set as by the ratio of width to height of detection block
1, the quantity of network query function acquisition can be reduced in this way, to improve detection speed.
The present invention also provides a kind of computer readable storage mediums, are stored thereon with computer program, which is characterized in that should
The step of method for detecting human face based on lightweight convolutional neural networks is realized when program is executed by processor.
The present invention also provides a kind of computer equipment, including memory, processor and storage on a memory and can located
The computer program run on reason device, wherein the processor is realized when executing described program is based on lightweight convolutional neural networks
Method for detecting human face the step of.
Particular embodiments described above has carried out further in detail the purpose of the present invention, technical scheme and beneficial effects
Describe in detail it is bright, it should be understood that the above is only a specific embodiment of the present invention, the guarantor being not intended to limit the present invention
Protect range.Without departing from the spirit and scope of the invention, any modification, equivalent substitution, improvement and etc. done also belong to this
Within the protection scope of invention.
Claims (10)
1. a kind of method for detecting human face based on lightweight convolutional neural networks, which is characterized in that comprising steps of
Face database is established in S1, data processing, carries out image preprocessing, generates training sample；
S2, the sample based on input carry out feature extraction using lightweight convolutional neural networks；The lightweight convolution mind
It include the convolutional layer that convolution kernel size is respectively 3x3 and 1x1 through network；
S3 is merged the different characteristic layer of lightweight convolutional neural networks based on Fusion Features module；
S4, anchor point frame are chosen；
S5 exports Analysis On Multi-scale Features figure；
S6, the mapping matching of face candidate area；
S7, face classification return；
S8 establishes non-maxima suppression constraint；
S9, output test result.
2. the method for detecting human face according to claim 1 based on lightweight convolutional neural networks, which is characterized in that in institute
It states in step S2, the core network of the lightweight convolutional neural networks CNN includes: sequentially connected 1 A block and 5 B
Block, wherein the port number of convolution is set as 16 in A block, the port number of each B block sets gradually as 16,32,
64、128、128。
3. the method for detecting human face according to claim 2 based on lightweight convolutional neural networks, which is characterized in that described
A block by convolution kernel size be 3 × 3, step-length be 2 convolutional layer realize it is down-sampled, thereafter add convolution kernel size be 1 ×
1, the convolutional layer that step-length is 1 achievees the purpose that input feature vector dimensionality reduction, is sequentially connected behind each convolutional layer for improving network
Nonlinear characteristic Batch Normalization and nonlinear function ReLU.
4. the method for detecting human face according to claim 3 based on lightweight convolutional neural networks, which is characterized in that described
B block is down-sampled by the maximum pond layer realization that convolution kernel size is 2 × 2, and step-length 2 has then been sequentially connected volume
The convolutional layer that product core size is 3 × 3,1 × 1,3 × 3, each convolutional layer step-length are 1, and are sequentially connected for improving net below
The nonlinear characteristic Batch Normalization and nonlinear function ReLU of network.
5. the method for detecting human face according to claim 1 based on lightweight convolutional neural networks, which is characterized in that in institute
It states in step S3, the Fusion Features module includes the convolutional layer that convolution kernel size is respectively 3 × 3 and 1 × 1, and step-length is
1, it is up-sampled using deconvolution, and by low-level feature figure and transformed high-level characteristic figure by the way of element summation
It is fused together.
6. a kind of face detection system based on lightweight convolutional neural networks, based on SSD detection framework, in feature extraction
Light weight convolution grade neural network is as core network designed by Shi Caiyong characterized by comprising
Data processing module carries out image preprocessing, generates training sample for establishing face database；
Light weight convolution grade neural network module, the sample based on input carry out feature using lightweight convolutional neural networks
It extracts；The lightweight convolutional neural networks include the convolutional layer that convolution kernel size is respectively 3x3 and 1x1；
Fusion Features module merges for the different characteristic layer to core network, and exports Analysis On Multi-scale Features figure；
Anchor point frame chooses module, for the ratio of width to height of anchor point frame to be arranged.
7. the face detection system according to claim 6 based on lightweight convolutional neural networks, which is characterized in that described
Lightweight convolutional neural networks include: sequentially connected 1 A block and 5 B block, wherein convolution in A block
Port number is set as 16, and it is 16,32,64,128,128 that the port number of each B block, which is set gradually,.
8. the face detection system according to claim 7 based on lightweight convolutional neural networks, which is characterized in that described
A block by convolution kernel size be 3 × 3, step-length be 2 convolutional layer realize it is down-sampled, thereafter add convolution kernel size be 1 ×
1, the convolutional layer that step-length is 1 achievees the purpose that input feature vector dimensionality reduction, is sequentially connected behind each convolutional layer for improving network
Nonlinear characteristic Batch Normalization and nonlinear function ReLU.
9. the face detection system according to claim 8 based on lightweight convolutional neural networks, which is characterized in that described
B block is down-sampled by the maximum pond layer realization that convolution kernel size is 2 × 2, and step-length 2 has then been sequentially connected volume
The convolutional layer that product core size is 3 × 3,1 × 1,3 × 3, each convolutional layer step-length are 1, and are sequentially connected for improving net below
The nonlinear characteristic Batch Normalization and nonlinear function ReLU of network.
10. the face detection system according to claim 6 based on lightweight convolutional neural networks, which is characterized in that institute
Stating Fusion Features module includes the convolutional layer that convolution kernel size is respectively 3 × 3 and 1 × 1, and step-length is 1, uses deconvolution
It is up-sampled, and is fused together low-level feature figure and transformed high-level characteristic figure by the way of element summation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910567225.8A CN110427821A (en) | 2019-06-27 | 2019-06-27 | A kind of method for detecting human face and system based on lightweight convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910567225.8A CN110427821A (en) | 2019-06-27 | 2019-06-27 | A kind of method for detecting human face and system based on lightweight convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110427821A true CN110427821A (en) | 2019-11-08 |
Family
ID=68409751
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910567225.8A Pending CN110427821A (en) | 2019-06-27 | 2019-06-27 | A kind of method for detecting human face and system based on lightweight convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110427821A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111126173A (en) * | 2019-12-04 | 2020-05-08 | 玉林师范学院 | High-precision face detection method |
| CN111291637A (en) * | 2020-01-19 | 2020-06-16 | 中国科学院上海微系统与信息技术研究所 | Face detection method, device and equipment based on convolutional neural network |
| CN111311634A (en) * | 2020-01-23 | 2020-06-19 | 支付宝实验室(新加坡)有限公司 | Face image detection method, device and equipment |
| CN111310509A (en) * | 2020-03-12 | 2020-06-19 | 北京大学 | Real-time bar code detection system and method based on logistics waybill |
| CN111428606A (en) * | 2020-03-19 | 2020-07-17 | 华南师范大学 | Lightweight face comparison verification method facing edge calculation |
| CN112560825A (en) * | 2021-02-23 | 2021-03-26 | 北京澎思科技有限公司 | Face detection method and device, electronic equipment and readable storage medium |
| CN112699826A (en) * | 2021-01-05 | 2021-04-23 | 风变科技（深圳）有限公司 | Face detection method and device, computer equipment and storage medium |
| CN113221985A (en) * | 2021-04-29 | 2021-08-06 | 大连海事大学 | Method for extracting basic features based on fusion network of pyramid model |
| CN113408327A (en) * | 2020-06-28 | 2021-09-17 | 河海大学 | Dam crack detection model and method based on improved Faster-RCNN |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108520219A (en) * | 2018-03-30 | 2018-09-11 | 台州智必安科技有限责任公司 | A kind of multiple dimensioned fast face detecting method of convolutional neural networks Fusion Features |
| CN108875618A (en) * | 2018-06-08 | 2018-11-23 | 高新兴科技集团股份有限公司 | A kind of human face in-vivo detection method, system and device |
| CN109145983A (en) * | 2018-08-21 | 2019-01-04 | 电子科技大学 | A kind of real-time scene image, semantic dividing method based on lightweight network |
| CN109214336A (en) * | 2018-09-05 | 2019-01-15 | 高新兴科技集团股份有限公司 | A kind of vehicle window marker detection method and device |
| CN109685017A (en) * | 2018-12-26 | 2019-04-26 | 中山大学 | A kind of ultrahigh speed real-time target detection system and detection method based on light weight neural network |
| CN109919013A (en) * | 2019-01-28 | 2019-06-21 | 浙江英索人工智能科技有限公司 | Method for detecting human face and device in video image based on deep learning |
-
2019
- 2019-06-27 CN CN201910567225.8A patent/CN110427821A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108520219A (en) * | 2018-03-30 | 2018-09-11 | 台州智必安科技有限责任公司 | A kind of multiple dimensioned fast face detecting method of convolutional neural networks Fusion Features |
| CN108875618A (en) * | 2018-06-08 | 2018-11-23 | 高新兴科技集团股份有限公司 | A kind of human face in-vivo detection method, system and device |
| CN109145983A (en) * | 2018-08-21 | 2019-01-04 | 电子科技大学 | A kind of real-time scene image, semantic dividing method based on lightweight network |
| CN109214336A (en) * | 2018-09-05 | 2019-01-15 | 高新兴科技集团股份有限公司 | A kind of vehicle window marker detection method and device |
| CN109685017A (en) * | 2018-12-26 | 2019-04-26 | 中山大学 | A kind of ultrahigh speed real-time target detection system and detection method based on light weight neural network |
| CN109919013A (en) * | 2019-01-28 | 2019-06-21 | 浙江英索人工智能科技有限公司 | Method for detecting human face and device in video image based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 张传磊等: "《基于图像分析的植物及其病虫害识别方法研究》", 31 October 2018, 北京：中国经济出版社 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111126173A (en) * | 2019-12-04 | 2020-05-08 | 玉林师范学院 | High-precision face detection method |
| CN111126173B (en) * | 2019-12-04 | 2023-05-26 | 玉林师范学院 | High-precision face detection method |
| CN111291637A (en) * | 2020-01-19 | 2020-06-16 | 中国科学院上海微系统与信息技术研究所 | Face detection method, device and equipment based on convolutional neural network |
| CN111311634A (en) * | 2020-01-23 | 2020-06-19 | 支付宝实验室(新加坡)有限公司 | Face image detection method, device and equipment |
| CN111311634B (en) * | 2020-01-23 | 2024-02-27 | 支付宝实验室(新加坡)有限公司 | Face image detection method, device and equipment |
| CN111310509A (en) * | 2020-03-12 | 2020-06-19 | 北京大学 | Real-time bar code detection system and method based on logistics waybill |
| CN111428606A (en) * | 2020-03-19 | 2020-07-17 | 华南师范大学 | Lightweight face comparison verification method facing edge calculation |
| CN111428606B (en) * | 2020-03-19 | 2023-03-31 | 华南师范大学 | Lightweight face comparison verification method facing edge calculation |
| CN113408327B (en) * | 2020-06-28 | 2022-08-26 | 河海大学 | Dam crack detection model and method based on improved Faster-RCNN |
| CN113408327A (en) * | 2020-06-28 | 2021-09-17 | 河海大学 | Dam crack detection model and method based on improved Faster-RCNN |
| CN112699826A (en) * | 2021-01-05 | 2021-04-23 | 风变科技（深圳）有限公司 | Face detection method and device, computer equipment and storage medium |
| CN112699826B (en) * | 2021-01-05 | 2024-05-28 | 风变科技（深圳）有限公司 | Face detection method, device, computer equipment and storage medium |
| CN112560825A (en) * | 2021-02-23 | 2021-03-26 | 北京澎思科技有限公司 | Face detection method and device, electronic equipment and readable storage medium |
| CN113221985A (en) * | 2021-04-29 | 2021-08-06 | 大连海事大学 | Method for extracting basic features based on fusion network of pyramid model |
| CN113221985B (en) * | 2021-04-29 | 2024-04-05 | 大连海事大学 | Method for extracting image basic features based on pyramid model fusion network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110427821A (en) | A kind of method for detecting human face and system based on lightweight convolutional neural networks | |
| CN106650693B (en) | Multi-feature fusion recognition algorithm for face comparison | |
| CN105069746B (en) | Video real-time face replacement method and its system based on local affine invariant and color transfer technology | |
| CN109034210A (en) | Object detection method based on super Fusion Features Yu multi-Scale Pyramid network | |
| CN108764063A (en) | A kind of pyramidal remote sensing image time critical target identifying system of feature based and method | |
| CN106354735A (en) | Image target searching method and device | |
| Luo et al. | Learning high-level features by fusing multi-view representation of MLS point clouds for 3D object recognition in road environments | |
| CN104361313B (en) | A kind of gesture identification method merged based on Multiple Kernel Learning heterogeneous characteristic | |
| CN107239793A (en) | Many quantisation depth binary feature learning methods and device | |
| CN111178208A (en) | Pedestrian detection method, device and medium based on deep learning | |
| CN108830237B (en) | Facial expression recognition method | |
| CN111310773A (en) | Efficient license plate positioning method of convolutional neural network | |
| CN110674685B (en) | Human body analysis segmentation model and method based on edge information enhancement | |
| CN109726619A (en) | A kind of convolutional neural networks face identification method and system based on parameter sharing | |
| CN109711416A (en) | Target identification method, device, computer equipment and storage medium | |
| Li et al. | MSAt-GAN: a generative adversarial network based on multi-scale and deep attention mechanism for infrared and visible light image fusion | |
| CN109993803A (en) | The intellectual analysis and evaluation method of city tone | |
| CN106250909A (en) | A kind of based on the image classification method improving visual word bag model | |
| CN110222636A (en) | The pedestrian's attribute recognition approach inhibited based on background | |
| CN108230330A (en) | A kind of quick express highway pavement segmentation and the method for Camera Positioning | |
| Du | High-precision portrait classification based on mtcnn and its application on similarity judgement | |
| CN104463210B (en) | Classification of Polarimetric SAR Image method based on object-oriented and spectral clustering | |
| CN103942572A (en) | Method and device for extracting facial expression features based on bidirectional compressed data space dimension reduction | |
| Wang et al. | CDFF: a fast and highly accurate method for recognizing traffic signs | |
| Weng et al. | Traffic scene perception based on joint object detection and semantic segmentation |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191108 |