CN113343758B - Long-distance unmanned aerial vehicle small target detection method based on infrared image - Google Patents
Long-distance unmanned aerial vehicle small target detection method based on infrared image Download PDFInfo
- Publication number
- CN113343758B CN113343758B CN202110454766.7A CN202110454766A CN113343758B CN 113343758 B CN113343758 B CN 113343758B CN 202110454766 A CN202110454766 A CN 202110454766A CN 113343758 B CN113343758 B CN 113343758B
- Authority
- CN
- China
- Prior art keywords
- response
- target
- scale
- infrared image
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Abstract
The invention provides a remote unmanned aerial vehicle small target detection method and system based on infrared images, which comprises the steps of firstly constructing cumulative direction derivative convolution kernels of a plurality of scales in all directions of the infrared images; and then optimizing the output of the average absolute gray difference algorithm, calculating the weighting coefficient of the optimized average absolute gray difference algorithm for each scale by adopting an accumulative direction derivative convolution kernel to obtain the weighting coefficient of each direction on each scale, taking the response with the minimum weighting coefficient of each direction on each scale as a response graph of the scale, taking the response graph as the weighting coefficient of the optimized average absolute gray difference algorithm on each scale to obtain response graphs on different scales, taking the maximum response graphs on different scales as a final response graph, and performing contour extraction on the final response graph to obtain a suspected target on the infrared image. The method effectively inhibits background clutter, enhances the target area and realizes accurate detection of the target.
Description
Technical Field
The invention relates to the field of unmanned aerial vehicle detection, in particular to a method for detecting a small target of a remote unmanned aerial vehicle based on an infrared image.
Background
An infrared target searching and tracking system (IRST) has the characteristics of hidden operation, electronic interference resistance and the like, and the detection of a long-distance small target in an infrared image is still difficult to realize due to low signal-to-noise ratio and complex background environment interference. The approximate background is then subtracted from the original image. Finally, the target is extracted by adopting a proper threshold segmentation strategy. Although this method is simple and has low computational complexity, it has a high false alarm rate for complex backgrounds. And the camera can only be opposite to a fixed area, the learning background cannot be scanned, and the detection range is limited.
The second method is to directly enhance the target area and suppress the background area in the infrared small target detection process, and to consider the gaussian model of the small target spatial distribution, and to solve the target size change in the continuous infrared image by using normalized laplacian gaussian (LoG) in the scale space. Local Contrast Method (LCM) target detection [1] showed good target enhancement capability. However, considering the regional maximum response value as a criterion (single pixel noise of the maximum response value) makes these algorithms very sensitive to salt-and-pepper noise. Meanwhile, the algorithms are implemented by using local background windows, so that maneuvering targets near high complex background clutter can be weakened, the average absolute gray difference is a small infrared target detection algorithm which is relatively strong and robust, and although the average absolute gray difference can effectively inhibit background noise (due to local average characteristics) and enhance weak and small targets, the algorithm is sensitive to high-intensity edges. In addition, as with other local contrast based algorithms, the algorithm utilizes a local background window, thereby increasing the miss rate when the target approaches a high intensity edge.
However, since the average absolute gray difference algorithm has no difference between positive and negative contrasts, the region where the target region of the local contrast and the void region of the local contrast are both negative is strengthened, as shown in fig. 2, the region corresponding to the hole region is falsely strengthened, which in turn increases the false alarm rate, and meanwhile, when the background of the infrared image contains a sharp edge with high intensity, the detection performance of the AAGD algorithm is sharply reduced, as shown in fig. 3.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a method for detecting a small target of a remote unmanned aerial vehicle based on an infrared image, which is used for realizing accurate detection of the small target.
The invention is realized by the following technical scheme:
a remote unmanned aerial vehicle small target detection method based on infrared images comprises the following steps:
step 2, optimizing the output of the average absolute gray difference algorithm, introducing a step function, eliminating negative contrast and only keeping positive contrast;
step 4, taking the response with the minimum weighting coefficient at the same position in each direction on each scale as the response graph D of the scale2；
Step 5, response graph D is formed on each scale2Obtaining response maps AAD on different scales as weighting coefficients of optimized average absolute gray difference algorithmi；
Step 6, forming a maximum response graph AAD by the maximum response values of the same position on different scalesiAnd as a final response image, carrying out contour extraction on the final response image to obtain a suspected target on the infrared image.
Preferably, the directions in step 1 are four directions, namely east-west-south-north;
the multiple scales in step 1 are four scales, 3 × 3,5 × 5,7 × 7 and 9 × 9 respectively.
Preferably, in step 2, the expression of the optimized mean absolute gray difference algorithm is as follows:
D1＝|μΦ-μΩ|2×H(μΦ-μΩ)
wherein, muΦIs the total number of pixels in the target area and muΩThe total number of pixels in the area is background, and H is a step function and has a positive value.
Preferably, the expression of the step function H is as follows:
preferably, the expression of the weighting coefficient of each direction in each scale in step 3 is as follows:
CDDn＝I*kn×H(I*kn)
CDDe＝I*ke×H(I*ke)
CDDs＝I*ks×H(I*ks)
CDDw＝I*kw×H(I*kw)
wherein I is an original infrared image, ke,ks,kw,knPerforming convolution operation on the image and the convolution kernel by four-direction convolution kernel, H is a step function, I x k is the image, and CDDn,CDDe,CDDs,CDDwRespectively, the weighting coefficients of four directions.
Preferably, the expression of the final response graph in step 6 is as follows:
AADCDD＝max{AADi,AADi+2,AADi+5,AADi+7}
preferably, the method further comprises the following steps:
step 8, determining the credibility of the suspected target by adopting the peak signal-to-noise ratio;
preferably, the method for determining the reliability of the suspected target by using the peak signal-to-noise ratio is as follows:
PSNR＝(MAXI-bmean)/sqrtf(MSE)
wherein bmean is the target background mean, MAXIThe maximum value of the pixel in the suspected area.
Preferably, step 9 is executed after step 8, the peak signal-to-noise ratio is compared with a preset sensitivity threshold, and a final target is input according to a comparison result.
A system of a remote unmanned aerial vehicle small target detection method based on infrared images comprises,
the system comprises an accumulation direction derivative convolution kernel module, a calculation module and a calculation module, wherein the accumulation direction derivative convolution kernel module is used for constructing an accumulation direction derivative convolution kernel of a plurality of scales in each direction of the infrared image;
the average absolute gray difference algorithm optimization module is used for optimizing the output of the average absolute gray difference algorithm, introducing a step function, eliminating negative contrast and only keeping positive contrast;
the weighting coefficient module is used for calculating the weighting coefficient of the optimized average absolute gray difference algorithm by adopting an accumulative direction derivative convolution kernel for each scale;
the scale response graph module is used for taking the response with the smallest weighting coefficient at the same position in each direction on each scale as the response graph of the scale;
the maximum response image module is used for taking the response image as a weighting coefficient of an optimized average absolute gray difference algorithm on each scale to obtain response images on different scales;
and the suspected target extraction module is used for taking the maximum response values of the same positions on different scales as a final response image, and performing contour extraction on the final response image to obtain a suspected target on the infrared image.
Compared with the prior art, the invention has the following beneficial technical effects:
according to the method for detecting the small target of the remote unmanned aerial vehicle based on the infrared image, provided by the invention, for the detection problem of the moving target in the complex background environment, an effective cumulative direction derivative convolution kernel is constructed and used for distinguishing between a target area and a non-target area, and the convolution operation is carried out on the complex background image containing the small target in multiple directions by adopting the method for constructing the convolution kernel, so that the small target close to a complex cloud layer can be effectively enhanced, and the problem of insufficient detection performance of the moving target close to the complex background environment is solved by introducing the cumulative direction derivative convolution kernel. Constructing the cumulative directional derivative convolution kernel solves the problem of the AAGD algorithm that addresses the high-intensity response of abrupt edges. Secondly, the main reason that the existing algorithm is sensitive to the abrupt edge is that due to two-dimensional filtering, directional information is ignored in the filtering process executed by a two-dimensional kernel, and the directional method for eliminating the sharp edge and the complex background is based on the cumulative derivative convolution kernel.
Drawings
FIG. 1 is two locally nested windows per pixel for a prior AAGD algorithm;
FIG. 2 is a comparison of an original infrared image with an AAGD algorithm;
wherein, fig. 2a is an original infrared picture, and fig. 2b is a response graph of the conventional AAGD algorithm.
FIG. 3 is a graph comparing an original infrared image with a response graph of an AAGD algorithm when the background of the original image contains a sharp edge of high intensity;
wherein, 3a is an original picture, 3b is an AAGD algorithm response graph, and 3c is an enlarged view of a box area in the 3a graph;
FIG. 4 is a graph of the decay of the response of the AAGD to the target zone when the target is on a uniform local background;
wherein 4a is an original picture, 4b is a state diagram of the target far away from the thick cloud layer, and 4c is a state diagram of the target close to the cloud layer; 4d is an AAGD algorithm response graph;
FIG. 5 is a response graph of the optimized AAGD algorithm of the present invention versus the existing AAGD algorithm;
wherein, 5a is the AADG response diagram of FIG. 2a, and 5b is the response diagram of the optimized AAGD algorithm for eliminating holes;
FIG. 6a is a spatial distribution model of a small infrared target with a two-dimensional Gaussian distribution in the prior art, and FIG. 6b is a 5X5 target window surrounded by a single pixel;
FIG. 7 is a schematic diagram of an accumulate direction convolution kernel of the detection method of the present invention;
FIG. 8 is a comparison graph of the effect of the object detection method of the present invention;
FIG. 8a is a raw infrared image and contains objects near the edge of a high intensity cloud; FIG. 8b is a response graph of the modified AAGD algorithm; FIG. 8c is a graph of the response of 8a after the convolution kernel of FIG. 7 using the accumulation direction;
FIG. 9 is a matrix diagram of the cumulative direction convolution kernel of the present invention;
FIG. 10 is a comparison of the sharp edges of the original picture and the response of the AAGD algorithm after optimization;
wherein, FIG. 10a is an original infrared image and contains sharp edges; FIG. 10b is a response graph of the modified AAGD algorithm; FIG. 10c is a graph of the response after applying the cumulative direction convolution kernel of FIG. 7;
FIG. 11 is a flow chart of the detection method of the present invention.
FIG. 12 is a diagram showing the effect of the detection method of the present invention.
Detailed Description
The present invention will now be described in further detail with reference to the attached drawings, which are illustrative, but not limiting, of the present invention.
Referring to fig. 1-12, a method for detecting a small target of a remote unmanned aerial vehicle based on infrared images comprises the following steps:
Specifically, four scales are adopted to construct the cumulative direction derivative convolution kernel k in four directions of east, west, south and northe,ks,kw,knThe four dimensions are 3 × 3,5 × 5,7 × 7 and 9 × 9, respectively.
Step 2, outputting average absolute gray difference algorithm (AAGD)Line optimization, step function introduction, negative contrast elimination, positive contrast preservation, and average absolute gray difference algorithm D after optimization1The output value of (a) is expressed as follows:
average absolute gray difference algorithm expression before optimization
Optimized average absolute gray difference algorithm expression
D1＝|μΦ-μΩ|2×H(μΦ-μΩ) (2)
Wherein, I (x, y), Nφ，NΩRespectively representing the pixel value at the (x, y) position, the total number of pixels contained in the phi set and the total number of pixels contained in the omega set, wherein H is a step function and has a positive value, and the expression is as follows:
the output value of the original AAGD algorithm utilizes the square of the absolute value of the difference between the average values of the target window and the background window, so that positive and negative contrasts are enhanced, the temperature of a small infrared object is higher than that of the surrounding environment, so that only positive local contrast exists, the optimized AAGD algorithm D1 is provided based on the small infrared object, the negative contrast is eliminated, only positive contrast is reserved, and the optimization can greatly reduce false alarm of images and enhance the reliability of detection targets.
The conventional AAGD has many false responses in the filtered image (fig. 5 a). These spurious responses are eliminated by the thresholding represented in equation (2), as shown in fig. 5 b. Therefore, false alarm of some cloud layers, especially thick cloud layers, can be reduced greatly, and the problem of false alarm of hole-like regions is solved by introducing a step function into an AAGD algorithm.
CDDn＝I*kn×H(I*kn)
CDDe＝I*ke×H(I*ke)
CDDs＝I*ks×H(I*ks)
CDDw＝I*kw×H(I*kw) (4)
wherein I is an original infrared image, ke,ks,kw,knPerforming convolution operation on the image and the convolution kernel by four-direction convolution kernel, H is a step function, I x k is the image, and CDDn,CDDe,CDDs,CDDwThe weighting coefficients of the AAGD algorithm are optimized for four directions, respectively, thus obtaining the weighting coefficients of each direction on 4 scales.
Step 4, taking the response with the smallest weighting coefficient at the same position in each direction on each scale as a response graph D of the scale2Will respond to the graph D2As an output result, to eliminate high intensity edges and structural background.
D2＝min{CDDn,CDDe,CDDs,CDDw} (5)
Wherein D is2The response map is a response map of a certain scale, and the same position is the same coordinate.
Step 5, response graph D is formed on each scale2AAGD algorithm D as optimization1To obtain response maps AAD on different scalesi。
AADi＝D1×D2 (6)
And 6, selecting the maximum response value at the same position on different scales as a final response graph.
AADCDD＝max{AAD3,AAD5,AAD7,AAD9} (7)
And 7, carrying out contour extraction on the final response image to obtain a suspected target area.
The suspected target area comprises position information and size information, and the probability that the suspected target area is a target is higher when the response value of the suspected target area is higher.
And 8, determining the reliability of the suspected target by adopting the peak signal-to-noise ratio.
There are typically multiple suspect regions in a map. For further quantitative analysis, the confidence of the suspected target is calculated using the peak signal-to-noise ratio (PSNR), which is defined as:
given a suspected target area of size m × n, the Mean Square Error (MSE) is defined as:
wherein bmean is the target background mean, MAXIThe maximum value of the pixel in the suspected area. In order to ensure the accuracy of psnr, calculation is carried out on 16-bit (short type) originally acquired image data, so that conversion errors caused by conversion of the data into 8-bit (uchar type) calculation are avoided, reliable information around a suspected target is fully utilized, and the reliability is improved.
The peak signal-to-noise ratio (PSNR) is introduced, the peak signal-to-noise ratio (PSNR) is combined with a suspected target area and a background area, surrounding information is fully utilized, the reliability of a target is well represented, and the specific calculation process is that calculation is carried out on collected original 16-bit data instead of converting 16 bits into 8-bit images, so that errors caused by data conversion are reduced.
And 9, comparing the peak signal-to-noise ratio with a preset sensitivity threshold, and inputting a final target according to a comparison result.
The sensitivity of the detection system is set. The algorithm can perform reconnaissance on three different levels, high, medium and low, according to different sensitivity thresholds, and the sensitivity is usually set to medium.
Simulation verification
To study the performance of the algorithm, a real infrared image containing a small target was simulated, and an ideal target detection algorithm could map the entire input image (except the target region) to zero (black), enhancing only the target region. The algorithm effectively inhibits background clutter and enhances target areas, and the images comprise complex scenes such as thick clouds, fog, backlight and the like.
Through verification, the low-altitude slow-speed small target detection method based on the infrared image, such as the remote unmanned aerial vehicle, detects various complex environments, such as thick clouds, rainy days, cloudy days, sunny days, cloudy days, backlight and the like, in all weather (day and night), no dead angle is formed around the equipment by 360 degrees, the detection distance can reach 6km at the maximum, and 4 pixels are the minimum target pixel 2 x 2.
The most efficient existing infrared small target detection algorithm Average Absolute Gray Difference (AAGD), which directly benefits from the local contrast of the target region, is based on the natural contrast of the infrared target region with its surrounding local background. Consider two locally nested windows per pixel (fig. 1), in which the square of the absolute difference between the average of the pixels belonging to the target window and the average of the pixels belonging to the background window is calculated as the output value. Since AAGD benefits from a local averaging operator, it can effectively suppress background noise. Furthermore, since the algorithm can be implemented using local averaging, real-time operation is easily achieved. Although the AAGD algorithm has excellent performance in real and noisy scenes, there are three major limitations to the algorithm, specifically as follows:
first, there is no difference in positive and negative contrast in the AAGD algorithm. Thus, regions where both local contrast (target region) and local contrast (void region) are negative will be enhanced. As shown in fig. 2, the area corresponding to the hole-like area (the upper rectangle of fig. 2 a) is erroneously enhanced, which in turn increases the false alarm rate.
Secondly, when the background of the infrared image contains a sharp edge of high intensity, the detection performance of the AAGD algorithm is drastically reduced. To better illustrate this problem, the AAGD algorithm in FIG. 3 becomes more responsive to edges, which returns a strong response to non-target regions. It has been found through extensive research that the response is proportional to the square of the edge sharpness.
Finally, the algorithm cannot enhance target maneuvers close to complex backgrounds. When the flying small target approaches the cloud layer edge and other complex scenes, most background windows are occupied by clutter, which in turn is reflectedTo increase the mean value μ of the background windowΩ. Therefore, the AAGD target enhancement ability sharply decreases. As shown in fig. 4. The AAGD algorithm can effectively enhance the target region when the target is on a uniform local background (fig. 4 b). As flying small targets get closer and closer into the high intensity cloud edge (fig. 4c), the AAGD response to the target zone becomes more attenuated (fig. 4 d).
The invention provides a method for detecting a small target of a remote unmanned aerial vehicle based on an infrared image, which is used for solving the problem that a moving target is detected in a complex background environment, wherein the problem is not limited to an AAGD algorithm, but all detection algorithms (such as PCM and LCM) utilizing a background window have the problem. To address this limitation, the detection capability of the algorithm is enhanced, the spatial distribution of the target pixels is considered, and an effective cumulative direction derivative convolution kernel is constructed for discrimination between the target region and the non-target region. The two-dimensional gaussian distribution (fig. 6a) is the most common model for modeling the spatial distribution of small infrared targets. The spatial distribution of the small infrared targets is simulated by considering two-dimensional Gaussian distribution, and the convolution operation is carried out on the complex background image containing the small targets in four directions by adopting a method for constructing a convolution kernel, so that the small targets close to the complex cloud layer can be effectively enhanced. FIG. 7 is a convolution kernel constructed by us, and an accumulative direction derivative convolution kernel is introduced to solve the problem that the detection performance of a maneuvering target is insufficient when the maneuvering target is close to a complex background environment. Fig. 8 shows the result of applying the convolution kernel to the infrared image containing the small target close to the high-intensity background clutter, and the proposed kernel can effectively enhance the small target close to the high-intensity background clutter, reduce false negative, and achieve accurate detection of the target.
Secondly, the problem of high-intensity response of abrupt edges is solved by constructing an accumulative directional derivative convolution kernel by an AAGD algorithm. This problem is not limited to only the AAGD algorithm, and almost all other detection algorithms suffer from this limitation. The main reason that existing algorithms are sensitive to abrupt edges is due to two-dimensional filtering. When the image is a two-dimensional signal, performing a filtering process by a two-dimensional kernel ignores directional information, which is important for distinguishing target and non-target areas in the field of small target detection. For example, sharp edges have only large intensity variations in some directions, while small infrared targets have relative contrast in all directions. Based on this fact, we propose a two-dimensional kernel based on cumulative derivatives (fig. 7) which is divided into four different directions to eliminate the sharp edges and the complex background orientation method, four-direction cumulative derivative kernels as shown in fig. 9, which use cumulative direction derivative convolution kernels to solve the high intensity sharp edge problem, and fig. 10 shows the edge removal feature of the proposed weighting coefficients.
The invention also provides a system of the remote unmanned aerial vehicle small target detection method based on the infrared image, which comprises an accumulated direction derivative convolution kernel module, an average absolute gray difference algorithm optimization module and a weighting coefficient module
The system comprises an accumulation direction derivative convolution kernel module, a calculation module and a calculation module, wherein the accumulation direction derivative convolution kernel module is used for constructing an accumulation direction derivative convolution kernel of a plurality of scales in each direction of the infrared image;
the average absolute gray difference algorithm optimization module is used for optimizing the output of the average absolute gray difference algorithm, introducing a step function, eliminating negative contrast and only keeping positive contrast;
the weighting coefficient module is used for calculating the weighting coefficient of the optimized average absolute gray difference algorithm by adopting an accumulative direction derivative convolution kernel for each scale;
the scale response graph module is used for taking the response with the minimum directional weighting coefficient on each scale as the response graph of the scale;
the maximum response image module is used for taking the response image as a weighting coefficient of an optimized average absolute gray difference algorithm on each scale to obtain response images on different scales;
and the suspected target extraction module is used for taking the maximum response values on different scales as a final response image, and performing contour extraction on the final response image to obtain a suspected target on the infrared image.
In an exemplary embodiment, a computer readable storage medium is also provided, which stores a computer program that, when executed by a processor, implements the steps of the infrared image based long range drone small target detection method. The computer storage medium may be any available medium or data storage device that can be accessed by a computer, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO), etc.), optical memory (e.g., CD, DVD, BD, HVD, etc.), and semiconductor memory (e.g., ROM, EPROM, EEPROM, nonvolatile memory (NANDFLASH), Solid State Disk (SSD)), etc.
In an exemplary embodiment, there is also provided a terminal correction system, which includes a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the infrared image-based remote unmanned aerial vehicle small target detection method when executing the computer program. The processor may be a Central Processing Unit (CPU), other general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, etc.
The above-mentioned contents are only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited thereby, and any modification made on the basis of the technical idea of the present invention falls within the protection scope of the claims of the present invention.
Claims (10)
1. A remote unmanned aerial vehicle small target detection method based on infrared images is characterized by comprising the following steps:
step 1, constructing cumulative direction derivative convolution kernels of multiple scales in all directions of an infrared image;
step 2, optimizing the output of the average absolute gray difference algorithm, introducing a step function, eliminating negative contrast and only keeping positive contrast;
step 3, calculating the weighting coefficient of the optimized average absolute gray difference algorithm by adopting an accumulative direction derivative convolution kernel for each scale to obtain the weighting coefficient of each direction on each scale;
step 4, taking the response with the minimum weighting coefficient at the same position in each direction on each scale as the response graph D of the scale2；
Step 5, response graph D is formed on each scale2Obtaining response maps AAD on different scales as weighting coefficients of optimized average absolute gray difference algorithmi；
Step 6, forming a maximum response graph AAD by the maximum response values of the same position on different scalesiAs a final response graph;
and 7, extracting the outline of the final response image to obtain a suspected target on the infrared image.
2. The method for detecting the small target of the remote unmanned aerial vehicle based on the infrared image as claimed in claim 1, wherein the directions in step 1 are four directions of east, west, south and north;
the multiple scales in step 1 are four scales, 3 × 3,5 × 5,7 × 7 and 9 × 9 respectively.
3. The method for detecting the small target of the unmanned aerial vehicle at a long distance based on the infrared image as claimed in claim 1, wherein in the step 2, the expression of the optimized mean absolute gray difference algorithm is as follows:
D1＝|μΦ-μΩ|2×H(μΦ-μΩ)
wherein, muΦIs the total number of pixels in the target area and muΩThe total number of pixels in the area is background, and H is a step function and has a positive value.
4. The method for detecting the small target of the unmanned remote control based on the infrared image as claimed in claim 3, wherein the expression of the step function H is as follows:
5. the method for detecting the small target of the unmanned remote control based on the infrared image as claimed in claim 1, wherein the expression of the weighting coefficient of each direction in each scale in step 3 is as follows:
CDDn＝I*kn×H(I*kn)
CDDe＝I*ke×H(I*ke)
CDDs＝I*ks×H(I*ks)
CDDw＝I*kw×H(I*kw)
wherein I is an original infrared image, ke,ks,kw,knPerforming convolution operation on the image and the convolution kernel by four-direction convolution kernel, H is a step function, I x k is the image, and CDDn,CDDe,CDDs,CDDwRespectively, the weighting coefficients of four directions.
6. The method for detecting the small target of the unmanned aerial vehicle at a long distance based on the infrared image as claimed in claim 1, wherein the expression of the final response graph in step 6 is as follows:
AADCDD＝max{AADi,AADi+2,AADi+5,AADi+7}。
7. the method for detecting the small target of the unmanned aerial vehicle at a long distance based on the infrared image as claimed in claim 1, further comprising the following steps:
and 8, determining the reliability of the suspected target by adopting the peak signal-to-noise ratio.
8. The method for detecting the small target of the remote unmanned aerial vehicle based on the infrared image as claimed in claim 7, wherein the method for determining the credibility of the suspected target by using the peak signal-to-noise ratio is as follows:
PSNR＝(MAXI-bmean)/sqrtf(MSE)
wherein bmean isTarget background mean, MAXIThe MSE is the mean square error of the maximum value of the pixels in the suspected area.
9. The method for detecting the small target of the unmanned aerial vehicle at a long distance based on the infrared image as claimed in claim 7, wherein step 8 is followed by step 9 of comparing the peak signal-to-noise ratio with a preset sensitivity threshold and inputting a final target according to the comparison result.
10. A system of the infrared image based remote unmanned aerial vehicle small target detection method of any one of claims 1 to 9, comprising,
the system comprises an accumulation direction derivative convolution kernel module, a calculation module and a calculation module, wherein the accumulation direction derivative convolution kernel module is used for constructing an accumulation direction derivative convolution kernel of a plurality of scales in each direction of the infrared image;
the average absolute gray difference algorithm optimization module is used for optimizing the output of the average absolute gray difference algorithm, introducing a step function, eliminating negative contrast and only keeping positive contrast;
the weighting coefficient module is used for calculating the weighting coefficient of the optimized average absolute gray difference algorithm by adopting an accumulative direction derivative convolution kernel for each scale;
the scale response graph module is used for taking the response with the smallest weighting coefficient at the same position in each direction on each scale as the response graph of the scale;
the maximum response image module is used for taking the response image as a weighting coefficient of an optimized average absolute gray difference algorithm on each scale to obtain response images on different scales;
and the suspected target extraction module is used for taking the maximum response values of the same positions on different scales as a final response image, and performing contour extraction on the final response image to obtain a suspected target on the infrared image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110454766.7A CN113343758B (en) | 2021-04-26 | 2021-04-26 | Long-distance unmanned aerial vehicle small target detection method based on infrared image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110454766.7A CN113343758B (en) | 2021-04-26 | 2021-04-26 | Long-distance unmanned aerial vehicle small target detection method based on infrared image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113343758A CN113343758A (en) | 2021-09-03 |
| CN113343758B true CN113343758B (en) | 2022-03-15 |
Family
ID=77468675
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110454766.7A Active CN113343758B (en) | 2021-04-26 | 2021-04-26 | Long-distance unmanned aerial vehicle small target detection method based on infrared image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113343758B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115861309B (en) * | 2023-02-22 | 2023-06-16 | 和普威视光电股份有限公司 | MPCM acceleration method, device, terminal and medium for infrared weak and small target detection |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107194355A (en) * | 2017-05-24 | 2017-09-22 | 北京航空航天大学 | A kind of utilization orientation derivative constructs the method for detecting infrared puniness target of entropy contrast |
| CN108038856A (en) * | 2017-12-22 | 2018-05-15 | 杭州电子科技大学 | Based on the infrared small target detection method for improving Multi-scale Fractal enhancing |
| CN109523575A (en) * | 2018-11-12 | 2019-03-26 | 南通理工学院 | Method for detecting infrared puniness target |
| CN111861968A (en) * | 2019-04-23 | 2020-10-30 | 中国科学院长春光学精密机械与物理研究所 | Infrared weak and small target detection method and detection system |
| CN112395944A (en) * | 2020-10-19 | 2021-02-23 | 周口师范学院 | Multi-scale ratio difference combined contrast infrared small target detection method based on weighting |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108182690B (en) * | 2017-12-29 | 2019-02-15 | 中国人民解放军63861部队 | A kind of infrared Weak target detecting method based on prospect weighting local contrast |
-
2021
- 2021-04-26 CN CN202110454766.7A patent/CN113343758B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107194355A (en) * | 2017-05-24 | 2017-09-22 | 北京航空航天大学 | A kind of utilization orientation derivative constructs the method for detecting infrared puniness target of entropy contrast |
| CN108038856A (en) * | 2017-12-22 | 2018-05-15 | 杭州电子科技大学 | Based on the infrared small target detection method for improving Multi-scale Fractal enhancing |
| CN109523575A (en) * | 2018-11-12 | 2019-03-26 | 南通理工学院 | Method for detecting infrared puniness target |
| CN111861968A (en) * | 2019-04-23 | 2020-10-30 | 中国科学院长春光学精密机械与物理研究所 | Infrared weak and small target detection method and detection system |
| CN112395944A (en) * | 2020-10-19 | 2021-02-23 | 周口师范学院 | Multi-scale ratio difference combined contrast infrared small target detection method based on weighting |
Non-Patent Citations (6)
| Title |
|---|
| A small dim infrared maritime targetdetection algorithm based on local peak detection and pipeline-filtering;WANG B;《Proceedings of the SPIE》;20161231(第2期);全文 * |
| Small infrared target detection utilizingLocal Region Similarity Difference map;QI H;《Infrared Physics and Technology》;20171231;全文 * |
| 基于全卷积递归网络的弱小目标检测方法;杨其利;《光学学报》;20200731;第40卷(第13期);全文 * |
| 基于梯度方向一致性和特征分解的红外小目标检测算法;范明明;《红外与激光工程》;20200131;第49卷(第1期);全文 * |
| 空中红外小目标检测跟踪方法研究;左健宏;《中国优秀博硕士学位论文全文数据库（硕士） 信息科技辑》;20201215(第12期);全文 * |
| 红外与可见光图像融合的无人机目标检测方法研究;赵烨;《中国优秀博硕士学位论文全文数据库（硕士） 工程科技II辑》;20200215(第2期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113343758A (en) | 2021-09-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107680054B (en) | Multi-source image fusion method in haze environment | |
| US7430303B2 (en) | Target detection method and system | |
| CN107767400B (en) | Remote sensing image sequence moving target detection method based on hierarchical significance analysis | |
| Wang et al. | Target detection via Bayesian-morphological saliency in high-resolution SAR images | |
| WO2023193401A1 (en) | Point cloud detection model training method and apparatus, electronic device, and storage medium | |
| Peng et al. | Model and context‐driven building extraction in dense urban aerial images | |
| CN113743385A (en) | Unmanned ship water surface target detection method and device and unmanned ship | |
| Guo et al. | Haze and thin cloud removal using elliptical boundary prior for remote sensing image | |
| CN113343758B (en) | Long-distance unmanned aerial vehicle small target detection method based on infrared image | |
| Venugopal | Sample selection based change detection with dilated network learning in remote sensing images | |
| CN115493612A (en) | Vehicle positioning method and device based on visual SLAM | |
| CN112329677B (en) | Remote sensing image river channel target detection method and device based on feature fusion | |
| CN106778822B (en) | Image straight line detection method based on funnel transformation | |
| Kusetogullari et al. | Unsupervised change detection in landsat images with atmospheric artifacts: a fuzzy multiobjective approach | |
| CN112734745B (en) | Unmanned aerial vehicle thermal infrared image heating pipeline leakage detection method fusing GIS data | |
| CN115272482A (en) | Camera external reference calibration method and storage medium | |
| CN114511879A (en) | Multisource fusion human body target detection method based on VIS-IR image | |
| CN114724023A (en) | Twin network-based water body change detection method | |
| CN113962900A (en) | Method, device, equipment and medium for detecting infrared dim target under complex background | |
| CN112800932A (en) | Method for detecting obvious ship target in marine background and electronic equipment | |
| Bragdon et al. | Environmentally informed buried object recognition | |
| Du et al. | A Hierarchical Saliency Based Target Detection Method for High-Resolution Sar Images | |
| US20240070832A1 (en) | Device and method for detecting wildfire | |
| Qin et al. | A Novel Approach to Object Detection in Remote-Sensing Images Based on YOLOv3 | |
| Yan et al. | Multiscale saliency detection method for ship targets in synthetic aperture radar images |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20220805 Address after: Room 1920, 19th Floor, Military-civilian Integration Innovation Port, No. 25 Development Avenue, High-tech Zone, Xi'an City, Shaanxi Province 710075 Patentee after: Xi'an Excellent Video Technology Co.,Ltd. Patentee after: Beijing Huacheng Zhiyun Software Co.,Ltd. Address before: Room 2501, 25 / F, building B1, Jinye times, No. 32, Jinye Road, Zhangba street, high tech Zone, Xi'an, Shaanxi 710075 Patentee before: Xi'an Excellent Video Technology Co.,Ltd. |