CN114898117A - Sequence infrared image target detection method based on equalized structure texture representation - Google Patents
Sequence infrared image target detection method based on equalized structure texture representation Download PDFInfo
- Publication number
- CN114898117A CN114898117A CN202210574392.7A CN202210574392A CN114898117A CN 114898117 A CN114898117 A CN 114898117A CN 202210574392 A CN202210574392 A CN 202210574392A CN 114898117 A CN114898117 A CN 114898117A
- Authority
- CN
- China
- Prior art keywords
- image
- infrared image
- gradient
- tensor
- target
- 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 20
- 238000000034 method Methods 0.000 claims abstract description 27
- 238000012512 characterization method Methods 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000013459 approach Methods 0.000 claims description 2
- 238000007781 pre-processing Methods 0.000 claims 2
- 238000004422 calculation algorithm Methods 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 3
- 238000005381 potential energy Methods 0.000 description 3
- 230000001629 suppression Effects 0.000 description 3
- 238000005457 optimization Methods 0.000 description 2
- 230000000717 retained effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/54—Extraction of image or video features relating to texture
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Transforming Light Signals Into Electric Signals (AREA)
Abstract
A sequence infrared image target detection method based on equalized structure texture representation includes the steps that firstly, a local intensity information graph is constructed by means of the fact that small targets in an infrared image are brighter than a background, background clutter of which the central pixel value is smaller than a local average pixel is fully suppressed, and possible image blocks of the targets are locked. Secondly, after image blocks possibly having small targets are obtained, a polar coordinate system is established at the center points of the image blocks, the small targets are detected by utilizing local gradient attributes, and the small targets are detected by testing given threshold values. And finally, combining the local strength and the local gradient to form an equalized structure texture representation, and adding a balance factor to adjust the proportion of the strength and the gradient in the equalized structure texture representation. Compared with other algorithms, the method can detect the position of the small target more efficiently and accurately.
Description
Technical Field
The invention belongs to the field of infrared small target detection, and particularly relates to a sequence infrared image target detection method based on equalization structure texture representation.
Background
With the rapid development of modern military technology, various precisely guided missiles play an important role in modern wars, and have the capability of precisely striking various aircraft carrier operation platforms such as military targets, airplanes, ships and aircrafts at any time. Therefore, in order to ensure the safety of the operating platform, the hit target needs to be effectively detected and tracked.
So far, methods for detecting infrared small targets at home and abroad can be divided into two types: a sequential detection method and a single frame detection method. Although the existing method works well, it is very necessary to consider time information because a single frame image lacks enough information to detect a small object. The self-regularized weighted (SRWS) model (Zhang T, Peng Z, Wu H, et al. Infrared small target detection part self-regulated weighted sparse model [ J ]. neuro-compression, 2021,420:124-148.) proposed by Zhang T, Peng Z, Wu H, et al, utilizes the overlapping edge information of the detectable background result information to constrain the sparse term, improves the detection precision, but needs a certain screening for the detected target. A norm constraint-based non-convex optimization (NOLC) model (Zhang T, Wu H, Liu Y, et al. Infrared small target detection based on non-concurrent optimization with Lp-norm constraint [ J ] removal Sensing,2019,11(5):559.) proposed by Zhang et al strengthens sparse term constraint and simultaneously properly scales low-rank term constraint, and the method has poor detection effect under a strong noise environment. Zhang et al propose an edge and Corner perception Based spatio-Temporal Tensor (BTR) Model (Zhang P, Zhang L, Wang X, et al edge and Corner aware-Based Spatial-Temporal Tensor Model for extracted Small-Target Detection [ J ]. IEEE Transactions on Geoscience and Remote Sensing,2020,59(12):10708-10724.) for Infrared Small Target Detection, which has problems in Target enhancement, the edge information of the Target cannot be effectively retained, causing the Target to be over-shrunk. At present, there are many methods based on a space-time tensor model, but in the space-time tensor model, how to rapidly detect the position of a small target is still a difficult problem.
Disclosure of Invention
The invention aims to provide a sequence infrared image target detection method based on equalized structure texture representation, so that the position of a target can be accurately detected in an infrared image lacking texture information and structure information. The method starts from the potential space in the infrared image, and utilizes the local intensity information and the local gradient information of the original infrared image to construct the balanced structure texture representation so as to more accurately detect the target. In addition, in order to improve the compatibility of the method under more scenes, a variable balance factor is additionally added in the model.
The technical scheme of the invention mainly comprises the following three contents: one is the value of using local intensity information for the detection of small targets. Since background intensities are typically small or nearly the same, local intensity information can be used to suppress a relatively dark or relatively uniform background. By calculating the average of the image blocks, background clutter, where the central pixel value is less than the local average pixel, can be suppressed. Secondly, most of the gradient vectors of the target point to the center of the target, and the local gradient attribute means that the gradient directions of the background with strong edges are consistent, so that the small target can be detected by using the local gradient attribute. Thirdly, a new equalized structure texture characterization is proposed by combining the local intensity graph and the local gradient graph to reduce the false alarm rate of the model. A new importance factor is added to the texture characterization of the equalization structure to measure the relative importance of the intensity map and the gradient map.
Compared with the prior art, the invention has the following beneficial effects:
(1) the method is based on an infrared image space-time model, and a tensor model of an original image is obtained through space-time.
(2) The method used by the invention effectively inhibits the background clutter, overcomes the influence of the background clutter on the detected target, constructs a balanced structure texture representation by combining the strength and the gradient, and can quickly position the target in the original image.
(3) The invention effectively balances the relative weights of the intensity graph and the gradient graph in different scenes by adding the adjustable balance factor in the texture characterization of the equalization structure, so that the method has stronger adaptability and robustness.
Drawings
FIG. 1 is a general flow diagram of the method of the present invention.
Fig. 2 is an infrared image data set original according to an embodiment of the present invention.
Fig. 3 is a diagram of a target result after detection in the embodiment of the present invention.
Fig. 4 is a 3D potential energy diagram of a detection result diagram in an embodiment of the present invention.
Fig. 5 is a comparative demonstration of the experimental results in the example of the present invention.
Detailed Description
The technical scheme of the invention is further explained in detail by combining the drawings in the specification.
As shown in fig. 1, the present invention provides an infrared small target detection method based on texture characterization of an equalization structure, which specifically includes the following steps:
step 1: the original infrared image is preprocessed, namely the RGB image is converted into a gray image, and the sequence picture is constructed into a space-time tensor. For the input image, if the input image is an RGB image, the image can be converted into a gray image through a matlab function RGB2gray, and the information of the image is effectively retained. For m pieces of infrared images of L multiplied by H size sequence, wherein L and H are respectively the length and the width of the infrared images, the method for constructing the space-time tensor is to construct a three-dimensional L multiplied by H multiplied by m tensor.
In the embodiment of the present invention, there are 5 data sets used, and as shown in fig. 2, partial infrared images are respectively shown in fig. (a): the size of the pictures of the airplane in cloudy weather is 128 multiplied by 128, and the number of the pictures is 84; FIG. (b): the target under the background of cloudy days, because the target is too small, there is no shape information available, the picture size is 256 x 200, totally 30; FIG. (c): the size of the picture of the ship under the sea-sky background is 320 multiplied by 240, and the number of the pictures is 120; FIG. (d): 160 vehicles in the field with the picture size of 384 by 288; FIG. (e): the size of the pictures of the airplane under the cloudy background is 256 multiplied by 200, and 70 pictures are obtained in total;
step 2: and (4) dividing the infrared picture into pieces according to the size of the sliding window n multiplied by n, and calculating the local intensity value I of the infrared image.
For an n × n image block of a given size, the average of its surrounding areas is:
wherein N is m The number of pixels in an image block is indicated. v. of ij A pixel specific value representing the position of ith row and jth column in the image block, where v 0 Representing the value of the central pixel. By the formula (1), the local intensity value of the image block can be obtained:
and step 3: and establishing a polar coordinate by taking the target image block as a center, and calculating the gradient value G of the target image. And combining the intensity value I with the gradient value G to form the balanced structure texture representation.
First, polar coordinates are established in the center of an image block, and the image block is divided into 4 parts, each of which can be expressed as:
wherein p is i Each part of the image is represented (i ═ 1,2,3, 4). Since the gradient of a small target is not strictly directed to the center, a more relaxed constraint is used to formulate the gradient towards the center:
wherein
Indicates satisfaction of region p i The set of gradients of the constraint is,
is one gradient element in the set and m and a are its magnitude and direction, respectively. Then, in each zone, the meterCalculating out
Mean square of (d):
wherein N is j (j ═ 1,2, …, N) is a set
In
The number of the cells. After the gradient values of the four parts are calculated, the local gradient value of the image block can be obtained:
wherein k is a constant, G max And G min Respectively representing the mean square of the maximum gradient and the mean square of the minimum gradient.
Analysis of the ir images combined found two notable results: (1) in a place with high intensity, a target exists and a strong edge in a background clutter exists at the same time; (2) the gradient values for small and irregular objects are generally low. Therefore, a new equalized structure texture characterization is proposed to reduce the false alarm rate of the model by combining the local intensity map and the local gradient map.
Wherein, IG ij ，I ij ，G ij Respectively representing pixel values in the texture characterization, intensity map and gradient map of the equalized structure. The weights of the intensity map and the gradient map may be adapted to various scenarios by adjusting the balance factor α. Mathematically, when α is 1, the gradient map and intensity map have equal influence; when alpha is>At 1, the gradient map takes up higher weight;when α is<The intensity map is more important at 1. In the method, the value of alpha is adjusted according to the actual situation of the infrared image so as to flexibly deal with various complex infrared scenes.
And 4, step 4: and (3) inputting the image obtained in the step (1) and the balanced structure texture representation obtained in the step (3) into a truncated nuclear norm tensor model, and iterating by an alternative direction multiplier method to obtain small target information detected in the infrared image. For a given sequence infrared image, a balanced structure texture representation of the image is obtained through calculation, pixels in a target area can be almost completely highlighted, meanwhile, the low rank and sparsity of an infrared image tensor are utilized, the tensor is regarded as the combination of a sparse tensor and a low rank tensor, a truncated nuclear norm tensor model is adopted to quickly approach the rank of the low rank tensor in the model, the value of the sparse tensor is enhanced through the balanced texture structure representation, the low rank tensor is finally restored to the background, and the sparse tensor is restored to the target.
The purpose of detecting the infrared small target is finally achieved through the method, the experimental result is shown in fig. 3, and the method can effectively detect the small target in the image. The 3D potential energy diagram of the target image is shown in fig. 4, and from the potential energy diagram, the patent can intuitively feel that the infrared image has remarkable effects on background suppression and target highlighting.
As shown in FIG. 5, in order to verify the advancement of the present patent, some advanced algorithms were selected for comparison, mainly IPI model, BTR model, TIP model, NOLC model, SRWS model. Fig. 5 shows a graph of the results of the detection of each algorithm model on the same data. At the same time, parameter indexes are respectively selected
(background suppression factor) and
(signal to noise ratio gain) to more intuitively judge the effect of the algorithm, as shown in tables 1 and 2, respectively. By contrast, the method is in a leading position, whether on background suppression or target highlighting.
Table 1: each method of
Value of
| IPI | BTR | TIP | NOLC | SRWS | OURS | |
| a | 0.31 | 2.30 | 12.75 | 21.08 | 7.60 | 29.04 |
| b | 7.12 | 27.04 | 15.41 | 31.34 | 20.20 | 25.38 |
| c | INF | 10.82 | 9.34 | 44.98 | 13.81 | 48.09 |
| d | 4.14 | 22.89 | 16.62 | 27.09 | 21.23 | 24.16 |
| e | 1.33 | 4.10 | 9.16 | 4.14 | 2.07 | 6.94 |
Table 2: each method of
Value of
| IPI | BTR | TIP | NOLC | SRWS | OURS | |
| a | 0.31 | 9.02 | 4.24 | 21.45 | 7.73 | 41.76 |
| b | 7.12 | 27.04 | 6.65 | 31.34 | 20.20 | 25.38 |
| c | NAN | 12.53 | 5.28 | 52.11 | 16.00 | 125.14 |
| d | 0.34 | 11.08 | 7.42 | 10.56 | 9.30 | 14.29 |
| e | 1.33 | 4.10 | 2.43 | 4.14 | 2.13 | 5.08 |
The above description is only a preferred embodiment of the present invention, and the scope of the present invention is not limited to the above embodiment, but equivalent modifications or changes made by those skilled in the art according to the present disclosure should be included in the scope of the present invention as set forth in the appended claims.
Claims (7)
1. A sequence infrared image target detection method based on equalized structure texture representation is characterized by comprising the following steps: the method comprises the following steps:
step 1: preprocessing an original infrared image, including converting an RGB image into a gray image, and constructing a sequence picture into a space-time tensor;
step 2: dividing the infrared image into pieces according to the size of the sliding window n multiplied by n, and calculating a local intensity value I of the infrared image;
and step 3: establishing a polar coordinate by taking the target image block as a center, and calculating a gradient value G of the target image; combining the intensity value I with the gradient value G to form balanced structure texture representation;
and 4, step 4: and (3) inputting the image obtained in the step (1) and the balanced structure texture representation obtained in the step (3) into a truncated nuclear norm tensor model, and iterating by an alternative direction multiplier method to obtain small target information detected in the infrared image.
2. The method for detecting the sequence infrared image target based on the equalization structure texture characterization according to claim 1, characterized in that: the image preprocessing in the step 1 is realized by the following steps: for an input RGB image, converting the image into a gray image through a function RGB2gray of matlab, and keeping the information of the image; for m pieces of L × H size sequence infrared images, wherein L, H are the length and width of the infrared image respectively, and m represents the frame number of the infrared image, a three-dimensional L × H × m space-time tensor is constructed.
3. The method for detecting the sequence infrared image target based on the equalization structure texture characterization according to claim 1, characterized in that: the specific implementation steps of calculating the local intensity value I of the infrared image in the step 2 are as follows:
for an n × n image block of a given size, the average of its surrounding areas is:
wherein N is m Indicating the number of pixels in an image block, v ij A pixel specific value representing the position of ith row and jth column in the image block, where v 0 A value representing the center pixel; the local intensity value I of the image block is obtained by equation (1):
4. the method for detecting the sequence infrared image target based on the equalization structure texture characterization according to claim 1, characterized in that: the specific implementation steps of calculating the gradient value G of the infrared image in the step 3 are as follows:
firstly, a polar coordinate system (r, theta) is established at the center of an image block i ) The image block is divided into 4 parts,each part is represented as:
wherein p is i Representing each part of the image, i ═ 1,2,3, 4; the gradient towards the center is formulated using a constraint:
wherein
Indicates satisfaction of region p i The set of gradients of the constraint is,
is one gradient element in the set, and m and a are its magnitude and direction, respectively; then, in each region, calculation is performed
Mean square of G i ：
Wherein N is j (j ═ 1,2, …, N) is a set
In
The number of (2); after the gradient values of the four parts are calculated, obtaining the local gradient value of the image block:
wherein k is a constant, G max And G min Respectively representing the mean square of the maximum gradient and the mean square of the minimum gradient.
5. The method for detecting the sequence infrared image target based on the equalization structure texture characterization according to claim 1, characterized in that: the specific implementation steps for calculating the texture characterization of the equalization structure of the infrared image in the step 3 are as follows:
comprehensively analyzing the infrared image, and providing balanced structure texture representation by combining the local intensity graph and the local gradient graph to reduce the false alarm rate of the model:
wherein, IG ij ，I ij ，G ij Respectively representing pixel values in the texture characterization, the intensity map and the gradient map of the equalized structure; the weights of the intensity map and the gradient map are adapted to various scenes by adjusting a balance factor alpha; when α is 1, the gradient map and the intensity map have equal influence; when alpha is>At 1, the gradient map takes up higher weight; when α is<The intensity map is more important at 1.
6. The method for detecting the sequence infrared image target based on the equalization structure texture characterization according to claim 1, characterized in that: in the step 4, a sequence infrared image is given, the texture representation of the equalization structure of the image is obtained through calculation so as to highlight the pixels of the target area, meanwhile, the tensor is regarded as the combination of the sparse tensor and the low-rank tensor by utilizing the low-rank property and the sparsity of the infrared image tensor, the low-rank tensor is restored to the background by cutting off the nuclear norm tensor model, and the sparse tensor is restored to the target.
7. The method for detecting the sequence infrared image target based on the equalization structure texture characterization according to claim 6, characterized in that: in the truncation nuclear norm tensor model, the truncation nuclear norm is adopted in the model to quickly approach the rank of the low-rank tensor, meanwhile, the value of the sparse tensor is enhanced by using the representation of the equalized texture structure, finally, the low-rank tensor is restored to the background, and the sparse tensor is restored to the target.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210574392.7A CN114898117A (en) | 2022-05-25 | 2022-05-25 | Sequence infrared image target detection method based on equalized structure texture representation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210574392.7A CN114898117A (en) | 2022-05-25 | 2022-05-25 | Sequence infrared image target detection method based on equalized structure texture representation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114898117A true CN114898117A (en) | 2022-08-12 |
Family
ID=82725295
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210574392.7A Pending CN114898117A (en) | 2022-05-25 | 2022-05-25 | Sequence infrared image target detection method based on equalized structure texture representation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114898117A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070230788A1 (en) * | 2006-03-31 | 2007-10-04 | Sony Deutschland Gmbh | Adaptive histogram equalization for images with strong local contrast |
| CN109934815A (en) * | 2019-03-18 | 2019-06-25 | 电子科技大学 | A kind of tensor recovery method for detecting infrared puniness target of combination ATV constraint |
| CN110807742A (en) * | 2019-11-21 | 2020-02-18 | 西安工业大学 | Low-light-level image enhancement method based on integrated network |
-
2022
- 2022-05-25 CN CN202210574392.7A patent/CN114898117A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070230788A1 (en) * | 2006-03-31 | 2007-10-04 | Sony Deutschland Gmbh | Adaptive histogram equalization for images with strong local contrast |
| CN109934815A (en) * | 2019-03-18 | 2019-06-25 | 电子科技大学 | A kind of tensor recovery method for detecting infrared puniness target of combination ATV constraint |
| CN110807742A (en) * | 2019-11-21 | 2020-02-18 | 西安工业大学 | Low-light-level image enhancement method based on integrated network |
Non-Patent Citations (1)
| Title |
|---|
| 刘忠林等: "多尺度红外超像素图像模型的小目标检测", 中国图象图形学报, no. 12, 16 December 2019 (2019-12-16) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Xu et al. | Review of video and image defogging algorithms and related studies on image restoration and enhancement | |
| CN106023257B (en) | A kind of method for tracking target based on rotor wing unmanned aerial vehicle platform | |
| CN112434745B (en) | Occlusion target detection and identification method based on multi-source cognitive fusion | |
| CN108182690B (en) | A kind of infrared Weak target detecting method based on prospect weighting local contrast | |
| CN103440502A (en) | Infrared small-target detection method based on mixing Gauss and sparse representation | |
| CN105160648A (en) | Radar target and shadow segmentation method based on wavelet and constant false alarm rate | |
| Yeriomina et al. | Formation of reference images and decision function in radiometric correlation-extremal navigation systems | |
| Guo et al. | Aircraft detection in high-resolution SAR images using scattering feature information | |
| Chen et al. | Space-based infrared aerial target detection method via interframe registration and spatial local contrast | |
| CN105046706B (en) | SAR image ship detection method based on rational polynominal Function Fitting sea clutter | |
| CN105139432B (en) | Infrared DIM-small Target Image emulation mode based on Gauss model | |
| Li et al. | Multi-scale rock detection on Mars | |
| CN113989631A (en) | Infrared image target detection network compression method based on convolutional neural network | |
| CN110276371B (en) | Container corner fitting identification method based on deep learning | |
| CN114898117A (en) | Sequence infrared image target detection method based on equalized structure texture representation | |
| Su et al. | A Space Target Detection Method Based on Spatial–Temporal Local Registration in Complicated Backgrounds | |
| CN111931558A (en) | Ship category identification method and system | |
| Xi et al. | Multitarget detection algorithms for multitemporal remote sensing data | |
| Xie et al. | UAV cluster detection algorithm based on weighted tri-layer window local contrast | |
| CN109410142A (en) | A kind of defogging method optimal based on image quality evaluation | |
| CN115187856A (en) | SAR image ship detection method based on human eye vision attention mechanism | |
| CN115205683A (en) | Infrared small target detection method | |
| Que et al. | Low altitude, slow speed and small size object detection improvement in noise conditions based on mixed training | |
| Lin et al. | Coarse to fine aircraft detection from front-looking infrared images | |
| Malarvel et al. | An Improved Kernel Density Estimation with adaptive bandwidth selection for Edge detection |
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 | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Deng Lizhen Inventor after: Xi Chunmei Inventor after: Zhu Hu Inventor before: Xi Chunmei Inventor before: Deng Lizhen Inventor before: Zhu Hu |