CN113222821A - Image super-resolution processing method for annular target detection - Google Patents
Image super-resolution processing method for annular target detection Download PDFInfo
- Publication number
- CN113222821A CN113222821A CN202110563493.XA CN202110563493A CN113222821A CN 113222821 A CN113222821 A CN 113222821A CN 202110563493 A CN202110563493 A CN 202110563493A CN 113222821 A CN113222821 A CN 113222821A
- Authority
- CN
- China
- Prior art keywords
- image
- resolution
- annular target
- super
- target image
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 14
- 238000003672 processing method Methods 0.000 title claims abstract description 10
- 239000013598 vector Substances 0.000 claims abstract description 30
- 230000000007 visual effect Effects 0.000 claims abstract description 23
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 238000012549 training Methods 0.000 claims description 16
- 238000000034 method Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 9
- 238000005516 engineering process Methods 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 3
- 230000006870 function Effects 0.000 description 2
- 230000001976 improved effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4046—Scaling of whole images or parts thereof, e.g. expanding or contracting using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an image super-resolution processing method facing to annular target detection, which comprises the following steps of 1, collecting an annular target image through a camera; step 2, inputting the annular target image into an image hyper-resolution model to extract image abstract features, and obtaining visual feature vectors of the image; step 3, preprocessing the visual feature vector in the step 2; step 4, upsampling the visual characteristic vector processed in the step 3; and 5, perfecting the up-sampled image through the convolutional layer to obtain the super-resolution annular target image. The invention can improve the sawtooth distortion of the edge of the annular target in the image, expand the shooting range of the camera when the annular target is detected, reduce the requirement of the resolution of the camera when the detection precision of the annular target is the same, and reduce the hardware cost for purchasing and using the high-resolution camera.
Description
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to an image super-resolution processing method for annular target detection.
Background
In the field of vision-based target tracking and measurement, the annular target has obvious visual characteristics, and is often placed on the surface of an object to be measured or other suitable areas for tracking or positioning the object to be measured.
The identification and positioning accuracy of the annular target is reduced along with the increase of the measured distance and the measured angle, and the jagged distortion of the image edge of the annular target is caused due to the insufficient sampling rate of the visual equipment.
When the annular target is identified and positioned, the target edge needs to be extracted, then the center of the target is fitted, and whether the target edge extraction is accurate or not is directly influenced by the positioning precision. With the development of computer technology, many practical circular target recognition algorithms are proposed.
Disclosure of Invention
The invention aims to solve the technical problem of the prior art, and provides an image super-resolution processing method facing to annular target detection, which comprises the steps of collecting high-resolution and low-resolution images of an annular target under the conditions of different distances, different inclination angles and the like, extracting low-resolution image features by using an image super-resolution model, improving the resolution by using an up-sampling module, finally obtaining an annular target image with clear edges, improving the identification precision of the annular target from the angle of recovering the image quality, and solving the problem of edge distortion of the annular target image under the conditions of long distance and large inclination angle.
In order to achieve the technical purpose, the technical scheme adopted by the invention is as follows:
an image super-resolution processing method facing to annular target detection comprises the following steps:
step 1, collecting an annular target image through a camera;
step 2, inputting the annular target image into an image hyper-resolution model to extract image abstract features, and obtaining visual feature vectors of the image;
step 3, preprocessing the visual feature vector in the step 2;
step 4, upsampling the visual characteristic vector processed in the step 3;
and 5, perfecting the up-sampled image through the convolutional layer to obtain the super-resolution annular target image.
In order to optimize the technical scheme, the specific measures adopted further comprise:
the step 3 is specifically: and rearranging the visual feature vectors to form the visual feature vectors with larger length and width.
The step 4 includes:
step 4-1: calculating the distribution of each characteristic image on the original annular target image based on the visual characteristic vectors processed in the step 3;
step 4-2: extracting feature vectors corresponding to elements in the original annular target image according to the distribution of the feature maps on the original annular target image;
step 4-3: and adding the feature vectors of all elements in the attention area aiming at the attention area in the original annular target image to obtain the feature vector corresponding to the attention area.
The method further comprises the following steps:
adopting a data set consisting of high-resolution images and low-resolution images of the annular target under the conditions of different distances and different inclination angles, and dividing the data set into a training set, an evaluation set and a test set;
setting training parameters, setting a loss function and setting training times;
and training, evaluating and testing the image hyper-segmentation model according to the training set, the evaluation set and the test set, and judging whether the image hyper-segmentation model is converged.
After the annular target image is input into the image hyper-resolution model, the annular target image is transmitted into a first convolution layer block through a first convolution layer operation, the numerical value is reduced after a plurality of convolution operations, and then the numerical value is added with the result of the first convolution layer to form a residual error structure, and then the residual error structure is transmitted into a second convolution block;
and (4) after the three-time residual error structure, enabling the obtained result to enter an upper sampling layer to change the resolution, and obtaining and outputting a final result after convolution operation.
Using TensoAnd the rRT technology is used for setting an image super-resolution model to realize real-time super-resolution of the image.
The invention has the following beneficial effects:
1. compared with the prior art, the method is based on the image super-resolution technology, only needs to use a low-resolution camera to shoot the annular target image, and does not need to use an expensive high-resolution camera. And secondly, when the resolution of the camera is lower, the frame rate can be improved, so that the annular target has better dynamic property during tracking. And when a low-resolution camera is used, the number of cameras can be increased and the shooting range can be expanded when the data transmission bandwidth is unchanged.
2. The method uses the deep learning technology, can obtain the highlight-resistant and dim-resistant image hyper-resolution model in the training process, and does not need to meet the environmental requirements of good illumination conditions and the like compared with the prior art.
3. The invention uses TensoThe rRT technology sets an image super-resolution model, can achieve the calculation speed of 12 frames per second, and realizes the real-time super-resolution of images.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a schematic diagram of an image hyper-resolution model structure according to the present invention;
FIG. 3 shows the improved effect of the hyper-segmentation model on the annular target image.
Detailed Description
Embodiments of the present invention are described in further detail below with reference to the accompanying drawings.
Referring to fig. 1, the image super-resolution processing method for annular target detection of the present invention includes:
step 1, collecting an annular target image through a camera;
step 2, inputting the annular target image into an image hyper-resolution model to extract image abstract features, and obtaining visual feature vectors of the image;
step 3, preprocessing the visual feature vector in the step 2;
step 4, upsampling the visual characteristic vector processed in the step 3;
and 5, perfecting the up-sampled image through the convolutional layer to obtain the super-resolution annular target image.
In an embodiment, the step 3 specifically includes: and rearranging the visual feature vectors to form the visual feature vectors with larger length and width.
In an embodiment, step 4 comprises:
step 4-1: calculating the distribution of each characteristic image on the original annular target image based on the visual characteristic vectors processed in the step 3;
step 4-2: extracting feature vectors corresponding to elements in the original annular target image according to the distribution of the feature maps on the original annular target image;
step 4-3: and adding the feature vectors of all elements in the attention area aiming at the attention area in the original annular target image to obtain the feature vector corresponding to the attention area.
In an embodiment, the method further comprises:
adopting a data set consisting of high-resolution images and low-resolution images of the annular target under the conditions of different distances and different inclination angles, and dividing the data set into a training set, an evaluation set and a test set;
setting training parameters, setting a loss function and setting training times;
and training, evaluating and testing the image hyper-segmentation model according to the training set, the evaluation set and the test set, and judging whether the image hyper-segmentation model is converged.
Referring to fig. 2, in the embodiment, after the annular target image is input into the image hyper-resolution model, the annular target image is transmitted into a first convolution layer block through a first convolution layer operation, a numerical value is reduced after a plurality of convolution operations, and then the numerical value is added with a result of the first convolution layer to form a residual error structure, and then the residual error structure is transmitted into a second convolution block;
and (4) after the three-time residual error structure, enabling the obtained result to enter an upper sampling layer to change the resolution, and obtaining and outputting a final result after convolution operation.
An image is input, and a high-resolution image with the length and the width being doubled is obtained after the image is processed by the model, so that the sawtooth distortion phenomenon of the edge of the annular target image is slowed down or removed, and the target image with clear edge is realized. The improvement effect of the model on the annular target image is shown in fig. 3, wherein the images of the annular target at a distance of 3 meters and 4 meters from the camera are respectively from top to bottom, and the images of the annular target processed by the model are respectively the original image, the local enlarged image of the annular target and the annular target image from left to right.
The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above-mentioned embodiments, and all technical solutions belonging to the idea of the present invention belong to the protection scope of the present invention. It should be noted that modifications and embellishments within the scope of the invention may be made by those skilled in the art without departing from the principle of the invention.
Claims (6)
1. An image super-resolution processing method for annular target detection is characterized by comprising the following steps:
step 1, collecting an annular target image through a camera;
step 2, inputting the annular target image into an image hyper-resolution model to extract image abstract features, and obtaining visual feature vectors of the image;
step 3, preprocessing the visual feature vector in the step 2;
step 4, upsampling the visual characteristic vector processed in the step 3;
and 5, perfecting the up-sampled image through the convolutional layer to obtain the super-resolution annular target image.
2. The method for processing super-resolution of images for circular target detection according to claim 1, wherein the step 3 specifically comprises: and rearranging the visual feature vectors to form the visual feature vectors with larger length and width.
3. The method for processing the image super-resolution facing the annular target detection according to claim 1, wherein the step 4 comprises:
step 4-1: calculating the distribution of each characteristic image on the original annular target image based on the visual characteristic vectors processed in the step 3;
step 4-2: extracting feature vectors corresponding to elements in the original annular target image according to the distribution of the feature maps on the original annular target image;
step 4-3: and adding the feature vectors of all elements in the attention area aiming at the attention area in the original annular target image to obtain the feature vector corresponding to the attention area.
4. The method for processing super-resolution of images for circular target detection according to claim 1, further comprising:
adopting a data set consisting of high-resolution images and low-resolution images of the annular target under the conditions of different distances and different inclination angles, and dividing the data set into a training set, an evaluation set and a test set;
setting training parameters, setting a loss function and setting training times;
and training, evaluating and testing the image hyper-segmentation model according to the training set, the evaluation set and the test set, and judging whether the image hyper-segmentation model is converged.
5. The image super-resolution processing method for annular target detection as claimed in claim 1, wherein after the annular target image is input into the image super-resolution model, the annular target image is introduced into a first convolution layer block through a first convolution layer operation, the numerical value is reduced after a plurality of convolution operations, and then the numerical value is added with the result of the first convolution layer to form a residual structure, and then the residual structure is introduced into a second convolution block;
and (4) after the three-time residual error structure, enabling the obtained result to enter an upper sampling layer to change the resolution, and obtaining and outputting a final result after convolution operation.
6. The image super-resolution processing method for annular target detection according to claim 1, wherein a TensorRT technology is used to set an image super-resolution model to realize real-time image super-resolution.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110563493.XA CN113222821A (en) | 2021-05-24 | 2021-05-24 | Image super-resolution processing method for annular target detection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110563493.XA CN113222821A (en) | 2021-05-24 | 2021-05-24 | Image super-resolution processing method for annular target detection |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113222821A true CN113222821A (en) | 2021-08-06 |
Family
ID=77099323
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110563493.XA Pending CN113222821A (en) | 2021-05-24 | 2021-05-24 | Image super-resolution processing method for annular target detection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113222821A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111192200A (en) * | 2020-01-02 | 2020-05-22 | 南京邮电大学 | Image super-resolution reconstruction method based on fusion attention mechanism residual error network |
| CN111402128A (en) * | 2020-02-21 | 2020-07-10 | 华南理工大学 | Image super-resolution reconstruction method based on multi-scale pyramid network |
| CN112001843A (en) * | 2020-07-28 | 2020-11-27 | 南京理工大学 | Infrared image super-resolution reconstruction method based on deep learning |
-
2021
- 2021-05-24 CN CN202110563493.XA patent/CN113222821A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111192200A (en) * | 2020-01-02 | 2020-05-22 | 南京邮电大学 | Image super-resolution reconstruction method based on fusion attention mechanism residual error network |
| CN111402128A (en) * | 2020-02-21 | 2020-07-10 | 华南理工大学 | Image super-resolution reconstruction method based on multi-scale pyramid network |
| CN112001843A (en) * | 2020-07-28 | 2020-11-27 | 南京理工大学 | Infrared image super-resolution reconstruction method based on deep learning |
Non-Patent Citations (4)
| Title |
|---|
| 刘祥龙等: "《飞桨PaddlePaddle深度学习实践》", 31 August 2020, 机械工业出版社, pages: 341 * |
| 戴亚平等: "《多传感器数据融合理论及应用》", 29 February 2004, 北京理工大学出版社, pages: 58 * |
| 李志朋等: ""多尺度融合超分辨率算法在无人机探测中的应用"", 《应用光学》, vol. 42, no. 3, 15 May 2021 (2021-05-15), pages 462 - 472 * |
| 郭琳等: ""基于通道注意递归残差网络的图像超分辨率重建"", 《计算机科学》, vol. 41, no. 8, 21 April 2021 (2021-04-21), pages 139 - 144 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111062915B (en) | Real-time steel pipe defect detection method based on improved YOLOv3 model | |
| CN111950330B (en) | Pointer instrument indication detection method based on target detection | |
| CN109708658B (en) | Visual odometer method based on convolutional neural network | |
| CN109376740A (en) | A kind of water gauge reading detection method based on video | |
| CN105547602A (en) | Subway tunnel segment leakage water remote measurement method | |
| CN111354047B (en) | Computer vision-based camera module positioning method and system | |
| CN110619328A (en) | Intelligent ship water gauge reading identification method based on image processing and deep learning | |
| CN109584259B (en) | Quartz crucible bubble layered counting device and method | |
| CN103699876B (en) | Method and device for identifying vehicle number based on linear array CCD (Charge Coupled Device) images | |
| CN110728269B (en) | High-speed rail contact net support pole number plate identification method based on C2 detection data | |
| CN116844147A (en) | Pointer instrument identification and abnormal alarm method based on deep learning | |
| CN115775236A (en) | Surface tiny defect visual detection method and system based on multi-scale feature fusion | |
| CN110909772B (en) | High-precision real-time multi-scale dial pointer detection method and system | |
| CN117291913B (en) | Apparent crack measuring method for hydraulic concrete structure | |
| CN113705564A (en) | Pointer type instrument identification reading method | |
| CN112801141B (en) | Heterogeneous image matching method based on template matching and twin neural network optimization | |
| CN115830514B (en) | Whole river reach surface flow velocity calculation method and system suitable for curved river channel | |
| CN111161264A (en) | Method for segmenting TFT circuit image with defects | |
| CN113222821A (en) | Image super-resolution processing method for annular target detection | |
| CN111860332B (en) | Dual-channel electrokinetic diagram part detection method based on multi-threshold cascade detector | |
| CN111260625B (en) | Automatic extraction method for offset printing large image detection area | |
| CN110111371B (en) | Speckle image registration method based on convolutional neural network | |
| CN112819823A (en) | Furniture board-oriented circular hole detection method, system and device | |
| CN111626299A (en) | Outline-based digital character recognition method | |
| CN113808049B (en) | Train line image distortion correction method and electronic equipment |
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 |