CN111161227B - Target positioning method and system based on deep neural network - Google Patents
Target positioning method and system based on deep neural network Download PDFInfo
- Publication number
- CN111161227B CN111161227B CN201911326585.5A CN201911326585A CN111161227B CN 111161227 B CN111161227 B CN 111161227B CN 201911326585 A CN201911326585 A CN 201911326585A CN 111161227 B CN111161227 B CN 111161227B
- Authority
- CN
- China
- Prior art keywords
- feature vector
- target
- image data
- image
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Geometry (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a target positioning method and a system based on a deep neural network, which comprises the following steps: establishing a simulation model matched with a preset object to obtain simulation image data; acquiring real image data after preprocessing; carrying out data enhancement processing on the simulation image data and the real image data to obtain a data set; training a deep neural network; inputting the data set into the trained deep neural network to obtain target type judgment and target area judgment results; screening target type judgment and target area judgment results to obtain a round hole target; extracting the image coordinates of each pixel point in the region corresponding to the round hole target; and calculating the average value of horizontal and vertical coordinates of the pixel points in the circular hole target area to obtain the image position coordinates of the target center of the preset object. The invention solves the problems that the low-definition image is difficult to position and the image shot at an inclined angle is difficult to identify based on the traditional image processing technical method at present.
Description
Technical Field
The invention relates to the field of computer vision application, in particular to a target center positioning method and system based on a deep neural network.
Background
In an actual production process, there are a large number of targets that cannot be measured in a contact-type manner. Particularly in a severe industrial environment, some moving objects cannot be measured by a conventional sensor, so that measuring the moving objects in real time in a non-contact manner is one of the hot spots of current research. In recent years, CCD image sensors have been widely applied in the fields of aerospace, satellite reconnaissance, remote sensing and remote measuring, optical image processing and the like. The target identification based on the DIM hardware equipment foundation is one of the problems to be solved in the current image identification field.
The current identification method is difficult to accurately detect the edge of a target by using an edge detection algorithm on an image with low definition and a complex background, and further difficult to locate a target. Secondly, the shape of the target changes due to the difference of the shooting angles, and the target is difficult to judge by a single rule. Based on the above limitations, it is difficult to achieve full-automatic target center identification at present, manual processing is required, efficiency is low, and consistency of reconnaissance results is poor.
With the rapid development of deep learning technology, mature application cases are already available in the fields of voice recognition, image recognition and the like. One of the key technologies for real-time digital photogrammetry is the real-time processing of two-dimensional images. Therefore, automatic image target detection and recognition can be realized by utilizing the deep learning technology.
Disclosure of Invention
The invention provides a method and a system for positioning a target based on a deep neural network aiming at an image under a real and complex condition, and aims to solve the problems that the low-definition image is difficult to position and the image shot at an inclined angle is difficult to identify based on the traditional image processing technical method.
Because the data acquisition difficulty is higher in a real scene, the image quality is lower, and the target positioning effect by directly using the neural network model is poorer. The invention increases the training sample size of the neural network model by using the mode of creating the simulation image data, thereby achieving the purpose of improving the performance of the neural network model and the generalization capability of the model.
In order to achieve the above object, in one aspect, the present invention provides a method for locating a bulls-eye based on a deep neural network, including:
step 1, acquiring simulation image data:
step 1.1, a 3D cylindrical target model is created, the upper bottom surface and the lower bottom surface of the cylindrical target model are both oval, round holes are formed in the upper bottom surface and the lower bottom surface, and a column body is a slender column;
step 1.2, setting a light source to irradiate towards the cylindrical target model, wherein the light source type uses a point light source or a natural light source; the purpose of using the light source is to simulate the shape and characteristics of an observed target in a dark scene, and the illumination direction is emitted to the target;
step 1.3, setting the material of the cylindrical target model to make the cylindrical target model present the texture of metal material;
step 1.4, setting different viewpoint observation cylindrical target models, and intercepting images to obtain two-dimensional images at different angles;
and step 1.5, Gaussian noise, Poisson noise, salt and pepper noise and multiplicative noise are respectively added to the image obtained in step 1.4.
Step 2, data preprocessing and data enhancement:
step 2.1, carrying out histogram equalization on the marked real image data, and achieving the effect of enhancing contrast by homogenizing the gray value distribution of the image;
2.2, performing data enhancement on the real image data in a mode of turning the real image data up and down, turning the real image data left and right, adjusting the brightness of the image and the like, and expanding a data set;
and 2.3, performing data enhancement on the simulation image data in modes of turning up and down, turning left and right, adjusting the brightness of the image and the like, and expanding a data set.
Step 3, learning of a network model:
step 3.1, extracting the features of the image by using a ResNext network, and respectively extracting a first feature vector, a second feature vector, a third feature vector and a fourth feature vector;
step 3.2, respectively carrying out up-sampling and convolution operation on the first feature vector, the second feature vector, the third feature vector and the fourth feature vector to obtain a fifth feature vector, a sixth feature vector, a seventh feature vector and an eighth feature vector;
step 3.3, performing pooling operation on the eighth feature vector to obtain a ninth feature vector;
step 3.4, inputting the first feature vector, the second feature vector, the third feature vector and the fourth feature vector into a Region generation network (RPN) to obtain Region proposals (Region suggestions);
step 3.5, matching the candidate area with the real marked area to obtain a positive sample and a negative sample; the candidate region is the screened region proposal.
And 3.6, inputting the positive sample, the negative sample, the fifth feature vector, the sixth feature vector, the seventh feature vector, the eighth feature vector and the ninth feature vector into a classification network and a frame regression network to perform target category judgment and target area judgment.
Step 4, target center positioning:
4.1, screening the result output in the step 3.6 to obtain a round hole target; removing non-circular hole targets from the circular hole-shaped recessed area on the top surface or the bottom surface of the circular hole target, namely the top surface or the bottom surface of the cylindrical model, according to the recognition result of the deep neural network;
step 4.2, extracting the image coordinates of each pixel point in the region of the round hole target, and recording the image coordinates as (x) i ,y i ) Wherein i is 1, 2, 3 … M; m is the number of pixel points in the target area;
step 4.3, calculating the average value of the horizontal and vertical coordinates of the pixel points of the 'round hole' target area to obtain the image position coordinates (X, Y) of the target center, wherein the formula is as follows:
on the other hand, corresponding to the method, the invention also provides a target positioning system based on the deep neural network, and the system comprises:
the simulation image data acquisition unit is used for establishing a simulation model matched with a preset object, and acquiring an image of the simulation model to obtain simulation image data;
the real image data acquisition unit is used for acquiring image data of a preset object, labeling the image data of the preset object, and preprocessing the labeled data to obtain preprocessed real image data;
the data enhancement unit is used for carrying out data enhancement processing on the simulation image data and the real image data to obtain a data set;
the deep neural network training unit is used for obtaining a training set based on the data set, training the deep neural network by using the training set, inputting the data set into the trained deep neural network, and obtaining target type judgment and target area judgment results;
the round hole target screening unit is used for screening target type judgment and target area judgment results to obtain a round hole target; removing non-circular hole targets from the circular hole-shaped recessed area on the top surface or the bottom surface of the circular hole target, namely the top surface or the bottom surface of the cylindrical model, according to the recognition result of the deep neural network;
an image coordinate extraction unit for extracting each region corresponding to the round hole targetThe image coordinates of each pixel point are recorded as (x) i ,y i ) Wherein i is 1, 2, 3 … M; m is the number of pixel points in the corresponding area of the round hole target;
and the bull's-eye image position coordinate calculating unit is used for calculating the average value of horizontal and vertical coordinates of pixel points in the circular hole target area to obtain the image position coordinates (X, Y) of the bull's-eye of the preset object.
One or more technical schemes provided by the invention at least have the following technical effects or advantages:
the method and the system have better identification accuracy rate for the images acquired under various complex conditions and have certain noise resistance. More than 99% of identification accuracy and recall rate in 8 pixel points can be achieved for ideal simulation image data. The real scene image data under the complex condition can reach more than 95% of recognition accuracy and recall rate in 10 pixel points. The method and the system use the deep neural network model to carry out target positioning on the cylindrical model under the complex background condition, thereby obtaining remarkable effect.
Drawings
The accompanying drawings, which are included to provide a further understanding of the embodiments of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention;
FIG. 1 is a schematic flow chart of a method for locating a target based on a deep neural network according to the present invention;
fig. 2 is a schematic diagram of a target location system based on a deep neural network according to the present invention.
Detailed Description
In order that the above objects, features and advantages of the present invention can be more clearly understood, a more particular description of the invention will be rendered by reference to the appended drawings. It should be noted that the embodiments and features of the embodiments of the present invention may be combined with each other without conflicting with each other.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, however, the present invention may be practiced in other ways than those specifically described and thus the scope of the present invention is not limited by the specific embodiments disclosed below.
Examples
Fig. 1 is a flow chart of a bulls-eye positioning method according to an embodiment of the invention. As shown in fig. 1, the data sources of the bulls-eye localization method include simulation image data and real image data. The method comprises the following specific steps:
step 1, acquiring simulation image data, namely establishing a simulation model matched with a preset object, acquiring an image of the simulation model, and obtaining simulation image data':
step 1.1 imitates the cylindrical model in the real image data, and uses OpenGL to create a 3D cylindrical target model. The upper bottom surface and the lower bottom surface of the cylindrical target model are both oval, circular holes are formed in the upper bottom surface and the lower bottom surface, and the cylinder body is a slender cylinder. In a preferred embodiment of the present invention, the ratio of the radius of the circular hole to the radius of the cylinder above and below the ground is, for example, 1: 2;
step 1.2, setting point light sources with different brightness to irradiate the cylindrical model, wherein the light sources are point light sources or natural light sources;
step 1.3, setting a cylindrical model material to present a metal texture;
step 1.4, setting different viewpoint observation cylindrical models, and intercepting images to obtain two-dimensional images at different angles; in a preferred embodiment of the invention, the included angle between the cylindrical model in the image and the vertical direction is kept between 0 and 60 degrees;
and 1.5, marking the round hole of the cylindrical model to obtain a binary mask image, and preparing for subsequent model training.
Step 1.6 is to add gaussian noise, poisson noise, salt and pepper noise, multiplicative noise to the two-dimensional image generated in step 1.4 respectively to simulate real image data.
Step 2, acquiring real image data, namely acquiring image data of a preset object, labeling the image data of the preset object, and preprocessing the labeled data to obtain preprocessed real image data:
step 2.1, collecting image data collected by a CCD under a real scene;
step 2.2, aiming at the image collected in the step 2.1, manually marking a round hole of the cylindrical model, and processing to obtain a binary mask image to prepare for subsequent model training;
and 2.3, carrying out histogram equalization on the image acquired in the step 2.1, and achieving the effect of enhancing the contrast by homogenizing the gray value distribution of the image.
Step 3, data enhancement, namely, data enhancement processing is carried out on the simulation image data and the real image data to obtain a data set:
3.1, carrying out up-down turning, left-right turning and picture brightness adjustment on the real scene image data to expand a data set;
and 3.2, carrying out up-down overturning, left-right overturning and picture brightness adjustment on the simulation image data to expand the data set.
Step 4, training the deep neural network, namely, training the deep neural network based on the data set, and inputting the data set into the trained deep neural network by using the training set to obtain a target type judgment result and a target area judgment result':
step 4.1, extracting the features of the image by using a ResNext network, and respectively extracting a first feature vector, a second feature vector, a third feature vector and a fourth feature vector;
step 4.2, performing up-sampling and convolution operations on the first feature vector, the second feature vector, the third feature vector and the fourth feature vector to obtain a fifth feature vector, a sixth feature vector, a seventh feature vector and an eighth feature vector;
step 4.3, performing pooling operation on the eighth feature vector to obtain a ninth feature vector;
step 4.4, inputting the first feature vector, the second feature vector, the third feature vector and the fourth feature vector into a region to generate a network to obtain a region proposal, and obtaining a candidate region based on the region proposal;
step 4.5, matching the candidate area with the real marked area to obtain a positive sample and a negative sample; the candidate region is the screened region proposal.
And 4.6, inputting the positive sample, the negative sample, the fifth feature vector, the sixth feature vector, the seventh feature vector, the eighth feature vector and the ninth feature vector into a classification network and a frame regression network to perform target category judgment and target area judgment.
Step 5, area division, namely 'obtaining a round hole target by screening target type judgment and target area judgment results':
step 5.1, calling the deep neural network model trained in the step 4;
and 5.2, inputting the real image data and the simulation image data in the data set into a deep neural network model to perform image region detection.
Step 6, calculating a target center, namely extracting image coordinates of each pixel point in an area corresponding to the round hole target; calculating the average value of horizontal and vertical coordinates of pixel points in the circular hole target area to obtain the image position coordinates of the target center of the preset object':
step 6.1, extracting a 'round hole' target area in the detection result of the step 5.2;
step 6.2, counting the image coordinates of each pixel point in the round hole target area, and recording as (x) i ,y i ) Wherein i is 1, 2, 3 … M; m is the number of pixel points in the corresponding area of the round hole target;
step 6.3, calculating the average value of the horizontal and vertical coordinates of the pixel points of the 'round hole' target area to obtain the image set coordinates (X, Y) of the target center, wherein the formula is as follows:
referring to fig. 2, an embodiment of the present invention further provides a target positioning system based on a deep neural network, the system including:
the simulation image data acquisition unit is used for establishing a simulation model matched with a preset object and acquiring an image of the simulation model to obtain simulation image data;
the real image data acquisition unit is used for acquiring image data of a preset object, labeling the image data of the preset object, and preprocessing the labeled data to obtain preprocessed real image data;
the data enhancement unit is used for carrying out data enhancement processing on the simulation image data and the real image data to obtain a data set;
the deep neural network training unit is used for obtaining a training set based on the data set, training the deep neural network by using the training set, inputting the data set into the trained deep neural network, and obtaining target category judgment and target area judgment results;
wherein the deep neural network training unit comprises:
the characteristic vector extraction module is used for extracting the characteristics of the simulation image and the real image by using a ResNext network, and respectively extracting a first characteristic vector, a second characteristic vector, a third characteristic vector and a fourth characteristic vector;
the sampling convolution module is used for respectively carrying out up-sampling and convolution operation on the first feature vector, the second feature vector, the third feature vector and the fourth feature vector to obtain a fifth feature vector, a sixth feature vector, a seventh feature vector and an eighth feature vector;
the pooling module is used for performing pooling operation on the eighth feature vector to obtain a ninth feature vector;
the proposing module is used for inputting the first feature vector, the second feature vector, the third feature vector and the fourth feature vector into the region to generate a network to obtain a region proposition, and obtaining a candidate region based on the region proposition;
the sample screening module is used for matching the candidate area with the real labeling area to obtain a positive sample and a negative sample;
and the judging module is used for inputting the positive sample, the negative sample, the fifth feature vector, the sixth feature vector, the seventh feature vector, the eighth feature vector and the ninth feature vector into the classification network and the frame regression network to judge the target category and the target area so as to obtain a target category judgment result and a target area judgment result.
The round hole target screening unit is used for screening target type judgment and target area judgment results to obtain a round hole target;
an image coordinate extraction unit for extracting the image coordinate of each pixel point in the region corresponding to the round hole target and recording as (x) i ,y i ) Wherein i is 1, 2, 3 … M; m is the number of pixel points in the corresponding area of the round hole target;
and the bull's-eye image position coordinate calculating unit is used for calculating the average value of horizontal and vertical coordinates of pixel points in the circular hole target area to obtain the image position coordinates (X, Y) of the bull's-eye of the preset object.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.
Claims (9)
1. A target center positioning method based on a deep neural network is characterized by comprising the following steps:
establishing a simulation model matched with a preset object, and acquiring an image of the simulation model to obtain simulation image data;
acquiring image data of a preset object, labeling the image data of the preset object, and preprocessing the labeled data to obtain preprocessed real image data;
carrying out data enhancement processing on the simulation image data and the real image data to obtain a data set;
obtaining a training set based on the data set, training the deep neural network by using the training set, and inputting the data set into the trained deep neural network to obtain target category judgment and target area judgment results;
screening target type judgment and target area judgment results to obtain a round hole target;
extracting the image coordinates of each pixel point in the area corresponding to the round hole target, and recording the image coordinates as (x) i ,y i ) Wherein i is 1, 2, 3 … M; m is the number of pixel points in the corresponding area of the round hole target;
calculating the average value of horizontal and vertical coordinates of pixel points in the circular hole target area to obtain image position coordinates (X, Y) of the target center of the preset object;
the method includes the steps of obtaining a training set based on a data set, training a deep neural network by using the training set, inputting the data set into the trained deep neural network, and obtaining target type judgment and target area judgment results, and specifically includes the following steps:
extracting the characteristics of the simulated image and the real image by using a ResNext network, and respectively extracting a first characteristic vector, a second characteristic vector, a third characteristic vector and a fourth characteristic vector;
respectively performing up-sampling and convolution operation on the first feature vector, the second feature vector, the third feature vector and the fourth feature vector to obtain a fifth feature vector, a sixth feature vector, a seventh feature vector and an eighth feature vector;
performing pooling operation on the eighth feature vector to obtain a ninth feature vector;
inputting the first feature vector, the second feature vector, the third feature vector and the fourth feature vector into a region to generate a network to obtain a region proposal, and obtaining a candidate region based on the region proposal;
matching the candidate area with the real labeling area to obtain a positive sample and a negative sample;
and inputting the positive sample, the negative sample, the fifth feature vector, the sixth feature vector, the seventh feature vector, the eighth feature vector and the ninth feature vector into a classification network and a frame regression network to perform target type judgment and target area judgment so as to obtain target type judgment and target area judgment results.
2. The method for locating the bulls-eye based on the deep neural network as claimed in claim 1, wherein the steps of: establishing a simulation model matched with a preset object, and acquiring an image of the simulation model to obtain simulation image data, wherein the simulation image data specifically comprises the following steps:
creating a simulation model;
setting a light source illumination direction simulation model;
setting different viewpoint observation simulation models, and intercepting images to obtain two-dimensional images at different angles;
and adding noise to the obtained two-dimensional image to obtain simulated image data.
3. The method as claimed in claim 2, wherein the simulation model is a 3D cylindrical target model, the upper and lower bottom surfaces of the cylindrical target model are elliptical, and the upper and lower bottom surfaces are provided with circular holes, and the cylindrical target model is made of metal.
4. The method for locating the bulls-eye based on the deep neural network as claimed in claim 2, wherein the light source is set to illuminate the cylindrical target model to simulate the shape and characteristics of the target observed in a dark scene, and the type of the light source is a point light source or a natural light source.
5. The method as claimed in claim 2, wherein Gaussian noise, Poisson noise, salt and pepper noise or multiplicative noise is added to the obtained two-dimensional image to obtain simulated image data.
6. The method of claim 1, wherein the pre-processing of the labeled data is performed by histogram equalization on the labeled image data.
7. The method of claim 1, wherein the artificial image data and the real image data are subjected to data enhancement by flipping the artificial image data and the real image data and adjusting the brightness of the image to expand the data set.
9. a target positioning system based on a deep neural network, the system comprising:
the simulation image data acquisition unit is used for establishing a simulation model matched with a preset object, and acquiring an image of the simulation model to obtain simulation image data;
the real image data acquisition unit is used for acquiring image data of a preset object, labeling the image data of the preset object, and preprocessing the labeled data to obtain preprocessed real image data;
the data enhancement unit is used for carrying out data enhancement processing on the simulation image data and the real image data to obtain a data set;
the deep neural network training unit is used for obtaining a training set based on the data set, training the deep neural network by using the training set, inputting the data set into the trained deep neural network, and obtaining target category judgment and target area judgment results;
the round hole target screening unit is used for screening target type judgment and target area judgment results to obtain a round hole target;
an image coordinate extraction unit for extracting the image coordinate of each pixel point in the region corresponding to the round hole target and recording as (x) i ,y i ) Wherein i is 1, 2, 3 … M; m is the number of pixel points in the corresponding area of the round hole target;
the bull's eye image position coordinate calculating unit is used for calculating the average value of horizontal and vertical coordinates of pixel points in the circular hole target area to obtain image position coordinates (X, Y) of the bull's eye of the preset object;
wherein the deep neural network training unit further comprises
The characteristic vector extraction module is used for extracting the characteristics of the simulation image and the real image by using a ResNext network, and respectively extracting a first characteristic vector, a second characteristic vector, a third characteristic vector and a fourth characteristic vector;
the sampling convolution module is used for respectively carrying out up-sampling and convolution operations on the first feature vector, the second feature vector, the third feature vector and the fourth feature vector to obtain a fifth feature vector, a sixth feature vector, a seventh feature vector and an eighth feature vector;
the pooling module is used for performing pooling operation on the eighth feature vector to obtain a ninth feature vector;
the proposing module is used for inputting the first feature vector, the second feature vector, the third feature vector and the fourth feature vector into the area to generate a network to obtain an area proposal and obtaining a candidate area based on the area proposal;
the sample screening module is used for matching the candidate area with the real labeling area to obtain a positive sample and a negative sample;
and the judging module is used for inputting the positive sample, the negative sample, the fifth feature vector, the sixth feature vector, the seventh feature vector, the eighth feature vector and the ninth feature vector into the classification network and the frame regression network to perform target category judgment and target area judgment so as to obtain a target category judgment and target area judgment result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911326585.5A CN111161227B (en) | 2019-12-20 | 2019-12-20 | Target positioning method and system based on deep neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911326585.5A CN111161227B (en) | 2019-12-20 | 2019-12-20 | Target positioning method and system based on deep neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111161227A CN111161227A (en) | 2020-05-15 |
CN111161227B true CN111161227B (en) | 2022-09-06 |
Family
ID=70557493
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911326585.5A Active CN111161227B (en) | 2019-12-20 | 2019-12-20 | Target positioning method and system based on deep neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111161227B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111797832B (en) * | 2020-07-14 | 2024-02-02 | 成都数之联科技股份有限公司 | Automatic generation method and system for image region of interest and image processing method |
CN112052946A (en) * | 2020-07-21 | 2020-12-08 | 重庆邮电大学移通学院 | Neural network training method and related product |
Family Cites Families (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170337682A1 (en) * | 2016-05-18 | 2017-11-23 | Siemens Healthcare Gmbh | Method and System for Image Registration Using an Intelligent Artificial Agent |
CN106597425B (en) * | 2016-11-18 | 2019-02-12 | 中国空间技术研究院 | A kind of radar target localization method based on machine learning |
CN108229252B (en) * | 2016-12-15 | 2020-12-15 | 腾讯科技(深圳)有限公司 | Pupil positioning method and system |
JP2019139346A (en) * | 2018-02-07 | 2019-08-22 | シャープ株式会社 | Image recognition device, image recognition system and program |
JP6962263B2 (en) * | 2018-04-17 | 2021-11-05 | 日本電信電話株式会社 | 3D point cloud label learning device, 3D point cloud label estimation device, 3D point cloud label learning method, 3D point cloud label estimation method, and program |
CN108960134A (en) * | 2018-07-02 | 2018-12-07 | 广东容祺智能科技有限公司 | A kind of patrol UAV image mark and intelligent identification Method |
CN109190625B (en) * | 2018-07-06 | 2021-09-03 | 同济大学 | Large-angle perspective deformation container number identification method |
CN109446973B (en) * | 2018-10-24 | 2021-01-22 | 中车株洲电力机车研究所有限公司 | Vehicle positioning method based on deep neural network image recognition |
CN109685066B (en) * | 2018-12-24 | 2021-03-09 | 中国矿业大学(北京) | Mine target detection and identification method based on deep convolutional neural network |
CN110059676B (en) * | 2019-04-03 | 2021-05-07 | 北京航空航天大学 | Aviation plug hole site identification method based on deep learning and multi-target distribution sequencing |
CN110321750A (en) * | 2019-04-23 | 2019-10-11 | 成都数之联科技有限公司 | Two-dimensional code identification method and system in a kind of picture |
CN110223352B (en) * | 2019-06-14 | 2021-07-02 | 浙江明峰智能医疗科技有限公司 | Medical image scanning automatic positioning method based on deep learning |
-
2019
- 2019-12-20 CN CN201911326585.5A patent/CN111161227B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN111161227A (en) | 2020-05-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113450307B (en) | Product edge defect detection method | |
CN106875381B (en) | Mobile phone shell defect detection method based on deep learning | |
CN105260699B (en) | A kind of processing method and processing device of lane line data | |
CN111640157B (en) | Checkerboard corner detection method based on neural network and application thereof | |
CN104778721B (en) | The distance measurement method of conspicuousness target in a kind of binocular image | |
CN106548462B (en) | Non-linear SAR image geometric correction method based on thin-plate spline interpolation | |
CN107392929B (en) | Intelligent target detection and size measurement method based on human eye vision model | |
CN113160192A (en) | Visual sense-based snow pressing vehicle appearance defect detection method and device under complex background | |
CN110232389A (en) | A kind of stereoscopic vision air navigation aid based on green crop feature extraction invariance | |
CN108596880A (en) | Weld defect feature extraction based on image procossing and welding quality analysis method | |
CN111507976B (en) | Defect detection method and system based on multi-angle imaging | |
CN112801074B (en) | Depth map estimation method based on traffic camera | |
CN109118528A (en) | Singular value decomposition image matching algorithm based on area dividing | |
CN107230203A (en) | Casting defect recognition methods based on human eye vision attention mechanism | |
CN108985337A (en) | A kind of product surface scratch detection method based on picture depth study | |
CN111127417B (en) | Printing defect detection method based on SIFT feature matching and SSD algorithm improvement | |
CN111161227B (en) | Target positioning method and system based on deep neural network | |
CN113920081A (en) | Cutter wear degree detection method | |
CN113313116B (en) | Underwater artificial target accurate detection and positioning method based on vision | |
CN112067233A (en) | Six-degree-of-freedom motion capture method for wind tunnel model | |
CN110334727B (en) | Intelligent matching detection method for tunnel cracks | |
CN114998308A (en) | Defect detection method and system based on photometric stereo | |
CN114972646B (en) | Method and system for extracting and modifying independent ground objects of live-action three-dimensional model | |
CN113205099A (en) | Visual detection system and method for material-removing characters on surface of nickel-plated metal | |
CN116977960A (en) | Rice seedling row detection method based on example 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 | ||
CB02 | Change of applicant information |
Address after: 610000 No. 270, floor 2, No. 8, Jinxiu street, Wuhou District, Chengdu, Sichuan Applicant after: Chengdu shuzhilian Technology Co.,Ltd. Address before: 610000 No.2, 4th floor, building 1, Jule Road intersection, West 1st section of 1st ring road, Wuhou District, Chengdu City, Sichuan Province Applicant before: CHENGDU SHUZHILIAN TECHNOLOGY Co.,Ltd. |
|
CB02 | Change of applicant information | ||
GR01 | Patent grant | ||
GR01 | Patent grant |