CN101149803A - Small false alarm rate test estimation method for point source target detection - Google Patents
Small false alarm rate test estimation method for point source target detection Download PDFInfo
- Publication number
- CN101149803A CN101149803A CNA2007101683227A CN200710168322A CN101149803A CN 101149803 A CN101149803 A CN 101149803A CN A2007101683227 A CNA2007101683227 A CN A2007101683227A CN 200710168322 A CN200710168322 A CN 200710168322A CN 101149803 A CN101149803 A CN 101149803A
- Authority
- CN
- China
- Prior art keywords
- image
- false alarm
- images
- background
- test
- 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.)
- Granted
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
A test estimation method with small false alarm rate of point source target detection, including construct image sample collection, calculate mean value and local statistical standard deviation S of image as characteristic A and B, build the mathematical model of distribution of characteristic A and B, construct sample characteristic space, find the background image prone to occur false alarm, judge whether we can achieve the classification of sample space according to the number of background image found prone to occur false alarm and distribution condition, enter into the step six or seven respectively, expand image classification results to the whole image characteristic space, estimate the detection performance of the entire image space; design new characteristic number of samples according to the distribution of sample collection in feature space, simulate new test sample images. The invention can reduce workload and reduce complexity of the simulated calculation, can be applied in the testing estimation of very low false alarm rate in the task that long-distance infrared cloud moving point target has requirements of high detecting reliability.
Description
Technical Field
The invention belongs to a digital image processing method and a test design, and particularly relates to a small false alarm rate test estimation method for point source target detection, which is mainly applied to performance evaluation of a moving point target detection algorithm in an infrared cloud picture background.
Background
Almost all literature discussing detection performance utilizes detection probability and false alarm probability to evaluate detection performance. Schmieder discusses in detail the relationship between the probability of detection of an object in a visible light image and the signal-to-noise ratio and contrast of the image; see Schmieder D E, weathersby M R, detection Performance in Filter with Variable resolution, IEEE Transactions on Aerospace and Electronic systems, jul 1983 AES-19 (4): 622-630.
Adam discusses the relationship between probability density functions and detection and recognition models, see Adam caromoli, thomas kurien.multitarget Identification in Airborne surveillance. Spie.1989.1098:161-176.
Driggers discusses the relationship between algorithm Performance and the set Infrared image model, see Driggers R G, vollmerhausen R, edwards t. The Target Identification Performance of extracted Imager Models as a Function of Blur and sampling. Spie.1999.3701:26-34.
Mc Williams discusses the relationship between detection probability and false alarm probability and target signal-to-noise ratio for a specific algorithm, see Mc Williams J K, srinath M d. Performance analysis of a target detection system using a simplified image. Ieee Transactions On aqueous and Electronic Systems,1984.20 (1): 38-48.
The filter detection performance of the filter on small and medium targets in a complex background is evaluated, the suppression capability on noise and low-frequency clutter needs to be considered, and the suppression effect on high-frequency clutter needs to be measured. Therefore, it is not appropriate to simply measure the quality of the filter by outputting the signal-to-noise ratio or the signal-to-noise ratio gain, and it is more reasonable to measure the performance of a certain detection technology or algorithm filter by using the false alarm rate Pf and the detection probability Pd.
However, for the measurement of a small false alarm rate, the simple sample sampling simulation test has a large calculation amount and takes a long time, and the false alarm rate test method for deteriorating the signal-to-noise ratio is incomplete in experimental design.
Disclosure of Invention
The invention aims to provide a small false alarm rate test estimation method for point source target detection, which can overcome the difficulty in testing the small false alarm rate of a moving point target, can test the performance of an algorithm more reasonably and effectively, reduces the workload and reduces the complexity of simulation calculation.
The invention provides a small false alarm rate test estimation method for point source target detection, which comprises the following steps:
(1) Randomly extracting images from the cloud layer background population to form an image sample set; then, extracting the characteristics of each sample in the image sample set, and calculating an image mean value m and an image local statistical standard deviation S which are respectively used as a characteristic A and a characteristic B;
(2) Respectively carrying out mathematical modeling on the distribution of the characteristics A and B in the image sample set;
(3) Constructing an image sample feature space by using the features A and B;
(4) Searching a background image which is easy to generate false alarm in an image sample feature space according to the following process:
(4.1) counting the Image of each Image in the Image library by taking the Image mean m and the local statistical standard deviation S as feature statistics k Characteristic value (m) of k ，S k ) Wherein k is more than or equal to 1 and less than or equal to R, R is the number of images in the image library, and k is the serial number of the images in the image library;
(4.2) traversing all images in the Image library, for Image k (m k ，S k ) If the Image is searched in the Image library p (m p ，S p ) Has m p ＞m k Or S p ＞S k If p is not less than 1 and not more than R and p is not equal to k, the Image is abandoned k Otherwise, recording Image k Is an Image c k Continuing to judge the next image until all images are traversed;
(4.3) Image detection by the tested algorithm software to be evaluated c k As a simulation sequence image of the background, if it isThe false alarm or the false alarm is recorded as Image c，k And temporarily rejecting the image from the image library; for all images c k In other words, if no abnormality occurs in the simulation test, the boundary Image is determined c，k And Image c k Then, entering the step (5), otherwise, entering the step (4.2);
(5) Judging whether the classification of the sample space can be realized according to the number and the distribution condition of the searched background images which are easy to generate false alarms, if so, entering the step (6), otherwise, entering the step (7);
(6) If a certain Image is adopted c k (m c k ，S c k ) Simulating the sequence Image and carrying out a target mapping test without false alarm, and then for all images in the Image library q (m q ，S q ) Is provided with
And is provided withQ is not less than 1 and not more than R, q is not equal to k, and Image is presumed to be obtained q Detecting the false alarm rate Pf of a target test for a simulation sequence Image of the background, wherein if a certain critical Image is adopted c k (m c k ，S c k ) Simulating a sequence Image, carrying out target test and having no false alarm, and then determining all the images in the Image library q (m q ，S q ) Is provided with
And is
Q is not less than 1 and not more than R, q is not equal to k, and Image is presumed to be obtained q The false alarm rate Pd of the simulation sequence image detection target test for the background is approximately equal to 100 percent; expanding the image classification result in the step (5) to the whole image feature space, speculating the test simulation test results of all images which are not subjected to simulation test in the image library, and estimating the detection performance of the whole image space by combining the test simulation test results;
(7) And designing a new sample characteristic value according to the distribution of the sample set in the characteristic space, and simulating a new test sample image.
The method utilizes the statistical characteristic of the background of the infrared cloud picture to popularize the characteristic distribution of the sample space into the image characteristic space, so that the false alarm rate test problem is converted into the classification problem of the image characteristic space. The method plays a crucial role in evaluating the performance of high-reliability detection of remote infrared motion points, popularizes the characteristic distribution of a sample space to an image characteristic space by utilizing the statistical characteristics of an infrared cloud picture background according to an infrared cloud picture sample set, converts the problem of false alarm rate test into the problem of classification of the image characteristic space to solve, popularizes the simulation test results of partial samples to the whole image space by simulation test of partial samples and theoretical derivation, predicts the detection performance of a detection algorithm to the whole image space, reduces the workload and reduces the complexity of analog calculation. The method can be used for testing and estimating the extremely low false alarm rate in a task with high reliable detection requirement on a long-distance infrared motion point target in a cloud picture background, such as: the performance evaluation of the satellite platform military target early warning and remote infrared flight target monitoring technology and the like.
Drawings
FIG. 1 is a flow chart of a method of experimental estimation of small false alarm rate for point source target detection in accordance with the present invention;
FIG. 2 is a graph of the analysis of the relationship of detection probability, false alarm rate and background distribution used in the present invention;
FIG. 3 is a mathematical modeling diagram of the medium wave infrared cloud image sample set feature statistics;
FIG. 4 is a schematic representation of a feature space of a medium wave infrared cloud pattern sample;
FIG. 5 is a diagram of a characteristic value distribution of an experimental image tested by a direct search method;
FIG. 6 is a histogram of the feature values of a structuring image.
Detailed Description
The invention is explained in more detail below with reference to the figures and examples.
The invention is used for estimating and testing the small false alarm of the high reliable detection performance of the infrared cloud picture background motion point target, and the flow is shown as the following figure 1:
(1) Firstly, feature extraction is carried out on each sample in an image sample set, an image mean value m is calculated to serve as a feature A, and an image local statistical standard deviation S serves as a feature B. The image sample set used should be drawn from the cloud background population following a random principle to meet the needs of the trial.
The corresponding definition formula is as follows:
image mean
Local statistical standard deviation of image
Wherein, l is the size of the dimensional Gaussian spatial filter template; i (I, j) is the gray level of a pixel point (I, j) in the input image; m is loc (i, j, l) is the average gray value of pixels in a local area with l multiplied by l size by taking the pixel point (i, j) as the center; the input image size is N × M.
(2) The distribution of features a and B in the sample set is mathematically modeled, respectively. Due to the scene in the natural environment, the statistical distribution characteristics of the scene inevitably accord with a certain objective rule.
For example, the feature statistics and mathematical model of the medium-wave infrared daytime cloud chart is shown in fig. 2, fig. 2 (a) is a fitting curve of a sample set local statistical standard deviation histogram and a lognormal distribution, and fig. 2 (b) is a fitting curve of a sample set mean histogram and a lognormal distribution.
The characteristic value of the background image of the medium wave cloud layer for the test can be determined: the local statistical standard deviation S and the mean m can better follow the log-normal distribution. Because the image characteristic mean value and the standard deviation are independent, the two-dimensional joint distribution probability density function
(0＜m＜+∞，0＜S＜+∞)
Probability distribution function
(0＜m＜+∞，0＜S＜+∞)
(3) And constructing an image sample feature space by using the features A and B.
The main reasons for selecting the image mean and the image local statistical standard deviation as the characteristic quantity of the sample characteristic space are as follows:
for a single frame image, based on Neyman-Pearson criterion, the false alarm rate Pf (cpd) under constant false alarm rate and the detection probability Pd (cpf) under constant false alarm rate in the detection process can be well represented, as shown in fig. 3, the shaded part in fig. 3 (a) represents the case that the false alarm rate Pf (cpd) under constant detection probability changes with the translation and stretching of the background, and the shaded part in fig. 3 (b) represents the case that the detection probability Pd (cpf) under constant false alarm rate changes with the translation and stretching of the background. If the user should back upEither scene distribution linear stretching (without mean change) or enhancement (global translation of background) will result in an increase in the false alarm rate Pf (cpd) at constant detection probability and a decrease in the detection probability Pd (cpf) at constant false alarm rate. Background intensity shift will cause the background image mean m to change, and background linear stretching without changing the mean will result in a background global standard deviation S all Local statistical standard deviation S change. Wherein the global standard deviation of the image is defined
Changes due to m increase: pd (cpf) monotonically decreases and Pf (cpd) monotonically increases, as represented by S all Change due to increaseThe method comprises the following steps: s monotonically increases Pd (cpf) monotonically decreases Pf (cpd) monotonically increases. And m and S all Unrelated, S and S all And (6) correlating.
Since m is not related to S, m is related to S all Not related, S all Related to S, for S all And S, selecting one of the two characteristics as a background statistical characteristic. Because the main component causing the false alarm is the background clutter with various characteristics similar to the target, and for the detection of small targets, the false alarm point is necessarily a high-frequency signal, so the image (high-pass filtering) after local mean removal is selected for variation and statistics, that is, the statistical characteristic value is determined as m and S.
(4) And searching a background image which is easy to generate false alarm in the image sample feature space.
The method for searching the background image which is easy to generate false alarm by the direct search method comprises the following specific steps:
(4.1) counting the Image of each Image in the Image library by taking the Image mean m and the local statistical standard deviation S as feature statistics k Characteristic value (m) of k ，S k ) Wherein k is more than or equal to 1 and less than or equal to R, R is the number of images in the image library, and k is the serial number of the images in the image library;
(4.2) traversing all images in the Image library, for Image k (m k ，S k ) If the image Imsge is found in the image library p (m p ，S p ) Has m p ＞m k Or S p ＞S k If p is not less than 1 and not more than R, p is not equal to k, the recording of the Image is abandoned k Otherwise, recording Image k Is an Image c k . Continuing to judge the next image until all images in the image library are traversed;
(4.3) Image detection by the tested algorithm software to be evaluated c k If false alarm or false alarm occurs, recording as Image c，k And temporarily rejecting the image from the image library;for all images c k In other words, if no abnormal condition occurs in the simulation test,determine its boundary Image c，k And Image c k Then, entering the step (5), otherwise, entering the step (4.2);
wherein the simulation and detection steps of the sequence image are as follows
(4.3.1) constructing a camera imaging model, a platform motion model, a random noise model, a target point motion model, a target intensity change model and a background change model.
(4.3.2) using Image c k For background, the sequence images were simulated with the model in (4.3.1) with a certain simulation accuracy.
And (4.3.3) taking the sequence image generated in the step (4.3.2) as an input, testing software to be evaluated, and recording a detection result.
(4.3.4) repeating the steps (4.3.2) - (4.3.3) for ten times, and if at least one simulation detection result shows a false target, judging the Image c k The Image is a background Image which is easy to generate false alarm for the algorithm to be evaluated, otherwise the Image can be considered as the Image c k Is a background image which is not easy to generate false alarm.
(5) And (4) judging whether the classification of the sample space can be realized according to the number and the distribution condition of the searched background images which are easy to generate false alarms, if so, entering the step (6), and otherwise, entering the step (7).
(6) If a certain Image is adopted c k (m c k ，S c k ) Simulating the sequence Image and carrying out a target mapping test without false alarm, and then for all images in the Image library q (m q ，S q ) Is provided with
And is
Q is more than or equal to 1 and less than or equal to R, q is not equal to k, and the Image is q Detecting the false alarm rate Pf of a target test for a background simulation sequence Image, wherein if a certain critical Image is adopted c k (m c k ，S c k ) Simulating sequence images and performing eye mappingIf no false alarm occurs, all images in the Image library are processed q (m q ，S q ) Is provided withAnd is provided withQ is not less than 1 and not more than R, q is not equal to k, and Image can be presumed q The false alarm rate Pd of the target test is about 100 percent for the background simulation sequence image detection. And (5) expanding the image classification result in the step (5) to the whole image feature space, speculating the test simulation test results of all images which are not subjected to simulation test in the image library, and estimating the detection performance of the whole image space by combining the test results.
(7) And designing a new sample characteristic value according to the distribution of the sample set in the characteristic space, and simulating a new test sample image. The construction method comprises the following specific steps:
(7.1) taking the Image mean value m and the local statistical standard deviation S as the horizontal axis and the vertical axis of the coordinate axis of the feature space, and counting the Image of each Image in the Image library k Characteristic value (m) of k ，S k ) Wherein k is more than or equal to 1 and less than or equal to R, R is the number of images in the image library, and k is the serial number of the images in the image library;
(7.2) traversing all images in the Image library, for Image k (m k ，S k ) If the Image is searched in the Image library p (m p ，S p ) Has m p ＞m k Or S p ＞S k If p is not less than 1 and not more than L and p is not equal to k, abandoning the recording of the Image k Otherwise, recording Image k Is an Image c k (ii) a Continuing to judge the next image until all images are traversed;
(7.3) Image c k Changing its characteristic value by conversion, and increasing mean value m by translation k Local statistical standard deviation S of stretch k Has m of new ＝m k +D，S new ＝S k XE causes newly constructed Image new Has a characteristic value of (m) new ，S new ) Wherein the translation coefficient D is more than 0, and the tensile coefficient E is more than or equal to 1;
(7.4) detecting by using algorithm software to be evaluated with Image new If false alarm or false alarm occurs, recording as Image c，k If for Image new If no false alarm or false alarm appears, increasing the translation coefficient D and the stretching coefficient E, and entering the step (7.3) again;
(7.5) repeating the steps (7.3) - (7.4) until the classification of the image feature space can be realized, entering the step (6), and calculating the detection performance of the band evaluation algorithm on the whole image feature space.
Example (c):
FIG. 4 is a distribution diagram of a diurnal medium wave infrared cloud Image sample set in an Image feature space, FIG. 5 is a process diagram of direct search of a background Image prone to false alarm according to the above steps (5.1) - (5.3), and a possible critical Image after each search c k Image with false or false alarm is tested and marked in the feature distribution diagram by circle c，k By stars " * "indicates. Fig. 5 (a) - (e) show the search results of the first to fifth rounds which are prone to generate false alarm images, and the successful classification of the sample set is realized through 5 rounds of search.
FIG. 6 is a process diagram of constructing a background image susceptible to false alarm according to the above steps (8.1) - (8.4), where FIG. 6 (a) shows the distribution of feature values of an edge image with circles, and FIG. 6 (b) shows asterisks " * "identify the feature value distribution of the newly constructed image," FIG. 6 (c) is a diagram for detecting the newly constructed image by simulationThe point of the "o" flag indicates that a false alarm condition has occurred in this feature pair, and the region where no false alarm has occurred (lower left region) and the region where a false alarm has occurred (upper right region) are divided in the sample feature space by a polyline.
Claims (3)
1. A small false alarm rate test estimation method for point source target detection comprises the following steps:
(1) Randomly extracting images from the cloud layer background population to form an image sample set; then, extracting the characteristics of each sample in the image sample set, and calculating an image mean value m and an image local statistical standard deviation S as a characteristic A and a characteristic B respectively;
(2) Respectively carrying out mathematical modeling on the distribution of the characteristics A and B in the image sample set;
(3) Constructing an image sample feature space by using the features A and B;
(4) Finding background images susceptible to false alarms in an image sample feature space according to the following process:
(4.1) taking the Image mean m and the local statistical standard deviation S as feature statistics, and counting the Image of each Image in the Image library k Characteristic value (m) of k ，S k ) Wherein k is more than or equal to 1 and less than or equal to R, R is the number of images in the image library, and k is the serial number of the images in the image library;
(4.2) traversing all images in the Image library, for Image k (m k ，S k ) If the Image is searched in the Image library p (m p ，S p ) Has m p ＞m k Or S p ＞S k If p is not less than 1 and not more than R and p is not equal to k, the Image is abandoned k Otherwise, recording Image k Is an Image c k Continuing to judge the next image until all images are traversed;
(4.3) detecting with Image by using tested algorithm software to be evaluated c k The simulation sequence Image of the background is recorded as Image if false alarm or false alarm occurs c，k And temporarily rejecting the image from the image library; for all images c k In other words, if no abnormality occurs in the simulation test, the boundary Image is determined c，k And Image c k Then entering the step (5), otherwise, turning to the step (4.2);
(5) Judging whether the classification of the sample space can be realized according to the number and the distribution condition of the searched background images which are easy to generate false alarms, if so, entering the step (6), otherwise, entering the step (7);
(6) If a certain Image is adopted c k (m c k ，S c k ) Simulating a sequence Image and carrying out target test without false alarm, and then aiming at all images in the Image library q (m q ，S q ) Is provided with
And is provided with
Q is not less than 1 and not more than R, q is not equal to k, and Image is presumed to be obtained q Background-based simulated sequential image inspectionThe false alarm rate Pf of the target measurement test is approximately equal to 0, if a certain critical Image is adopted c k (m c k ，S c k ) Simulating a sequence Image, carrying out target test and generating no false alarm, and then, for all images in the Image library q (m q ，S q ) Is provided with
And is
Q is not less than 1 and not more than R, q is not equal to k, and Image is presumed to be obtained q The false alarm rate Pd of the simulation sequence image detection target test for the background is approximately equal to 100 percent; expanding the image classification result in the step (5) to the whole image feature space, speculating the test simulation test results of all images which are not subjected to simulation test in the image library, and estimating the detection performance of the whole image space by combining the test simulation test results;
(7) And designing a new sample characteristic value according to the distribution of the sample set in the characteristic space, and simulating a new test sample image.
2. The method of claim 1, wherein: the sequence image simulation and detection steps in the step (4.3) are as follows:
(4.3.1) constructing a camera imaging model, a platform motion model, a random noise model, a target point motion model, a target intensity change model and a background change model;
(4.3.2) using Image c k As background, simulating a sequence image by using the model in the step (4.3.1) under the determined simulation precision;
(4.3.3) taking the sequence image generated in the step (4.3.2) as input, testing algorithm software to be evaluated, and recording a detection result;
(4.3.4) repeating the steps (4.3.2) - (4.3.3) at least ten times, and if at least one simulation detection result has a false target, judging the Image c k Regarding the algorithm to be evaluated as a background Image which is easy to generate false alarm, otherwise, considering the Image as c k Is a background image which is not easy to generate false alarm.
3. The method according to claim 1 or 2, characterized in that: the step (7) comprises the following processes:
(7.1) taking the Image mean value m and the local statistical standard deviation S as the horizontal axis and the vertical axis of the coordinate axes, and counting the Image of each Image in the Image library k The characteristic value of (a);
(7.2) traversing all images in the Image library for Image k (m k ，S k ) If the Image is searched in the Image library p (m p ，S p ) Has m p ＞m k Or S p ＞S k If p is not less than 1 and not more than L and p is not equal to k, abandoning the recording of the Image k Otherwise, recording Image k Is an Image c k (ii) a Continuously judging the next image until all the images are traversed;
(7.3) Image c k Changing its characteristic value by transformation, and increasing mean value m by translation k Local statistical standard deviation S of stretch k Has m of new ＝m k +D，S new ＝S k XE order New Structure Image new Has a characteristic value of (m) new ，S new ) Wherein the translation coefficient D is more than 0, and the tensile coefficient E is more than or equal to 1;
(7.4) detecting with Image by using algorithm software to be evaluated new The simulation sequence Image of the background is recorded as Image if false alarm or false alarm occurs c，k If it is for Image new If no false alarm or false alarm occurs, increasing the translation coefficient D and the stretching coefficient E, and entering the step (7.3) again;
(7.5) repeating the steps (7.3) - (7.4) until the classification of the image feature space is realized, entering the step (6), and calculating the detection performance of the band estimation algorithm on the whole image feature space.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2007101683227A CN100476866C (en) | 2007-11-09 | 2007-11-09 | Small false alarm rate test estimation method for point source target detection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2007101683227A CN100476866C (en) | 2007-11-09 | 2007-11-09 | Small false alarm rate test estimation method for point source target detection |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101149803A true CN101149803A (en) | 2008-03-26 |
| CN100476866C CN100476866C (en) | 2009-04-08 |
Family
ID=39250318
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2007101683227A Expired - Fee Related CN100476866C (en) | 2007-11-09 | 2007-11-09 | Small false alarm rate test estimation method for point source target detection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100476866C (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101551660B (en) * | 2008-03-31 | 2012-09-19 | 鸿富锦精密工业（深圳）有限公司 | Machine platform simulation system and method |
| CN102997902A (en) * | 2012-12-18 | 2013-03-27 | 中国科学院对地观测与数字地球科学中心 | Rapid quality detection method for data of digital camera |
| CN107940888A (en) * | 2017-12-07 | 2018-04-20 | 何旭连 | A kind of intelligent refrigerator system based on cloud service |
| CN111213153A (en) * | 2019-01-30 | 2020-05-29 | 深圳市大疆创新科技有限公司 | Target object motion state detection method, device and storage medium |
| CN112540352A (en) * | 2019-09-20 | 2021-03-23 | 初速度（苏州）科技有限公司 | Method and device for evaluating target detection algorithm based on unmanned vehicle |
| CN112954239A (en) * | 2021-01-29 | 2021-06-11 | 中国科学院长春光学精密机械与物理研究所 | On-board CMOS image dust pollution removal and recovery system and recovery method |
-
2007
- 2007-11-09 CN CNB2007101683227A patent/CN100476866C/en not_active Expired - Fee Related
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101551660B (en) * | 2008-03-31 | 2012-09-19 | 鸿富锦精密工业（深圳）有限公司 | Machine platform simulation system and method |
| CN102997902A (en) * | 2012-12-18 | 2013-03-27 | 中国科学院对地观测与数字地球科学中心 | Rapid quality detection method for data of digital camera |
| CN102997902B (en) * | 2012-12-18 | 2015-08-19 | 中国科学院对地观测与数字地球科学中心 | A kind of rapid quality detection method for data of digital camera |
| CN107940888A (en) * | 2017-12-07 | 2018-04-20 | 何旭连 | A kind of intelligent refrigerator system based on cloud service |
| CN111213153A (en) * | 2019-01-30 | 2020-05-29 | 深圳市大疆创新科技有限公司 | Target object motion state detection method, device and storage medium |
| CN112540352A (en) * | 2019-09-20 | 2021-03-23 | 初速度（苏州）科技有限公司 | Method and device for evaluating target detection algorithm based on unmanned vehicle |
| CN112540352B (en) * | 2019-09-20 | 2022-09-27 | 魔门塔(苏州)科技有限公司 | Method and device for evaluating target detection algorithm based on unmanned vehicle |
| CN112954239A (en) * | 2021-01-29 | 2021-06-11 | 中国科学院长春光学精密机械与物理研究所 | On-board CMOS image dust pollution removal and recovery system and recovery method |
| CN112954239B (en) * | 2021-01-29 | 2022-07-19 | 中国科学院长春光学精密机械与物理研究所 | On-board CMOS image dust pollution removal and recovery system and recovery method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN100476866C (en) | 2009-04-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106778595B (en) | Method for detecting abnormal behaviors in crowd based on Gaussian mixture model | |
| CN104899866B (en) | A kind of intelligentized infrared small target detection method | |
| CN101149803A (en) | Small false alarm rate test estimation method for point source target detection | |
| CN110598613B (en) | Expressway agglomerate fog monitoring method | |
| CN102231844B (en) | Video image fusion performance evaluation method based on structure similarity and human vision | |
| Lian et al. | A novel method on moving-objects detection based on background subtraction and three frames differencing | |
| CN104834915A (en) | Small infrared object detection method in complex cloud sky background | |
| CN112084826A (en) | Image processing method, image processing apparatus, and monitoring system | |
| CN108305265B (en) | Real-time processing method and system for weak and small target image | |
| CN111611918A (en) | Traffic flow data set acquisition and construction method based on aerial photography data and deep learning | |
| CN115272871A (en) | Method for detecting dim small target under space-based background | |
| CN116402852A (en) | Dynamic high-speed target tracking method and device based on event camera | |
| CN114782859A (en) | Method for establishing space-time perception positioning model of target behaviors and application | |
| CN106778822B (en) | Image straight line detection method based on funnel transformation | |
| KR101394473B1 (en) | Method for detecting moving object and surveillance system thereof | |
| CN116228712A (en) | Multi-scale slope disaster monitoring method, system and device | |
| CN114429593A (en) | Infrared small target detection method based on rapid guided filtering and application thereof | |
| Zhao et al. | Point target detection in space-based infrared imaging system based on multi-direction filtering fusion | |
| CN117314783B (en) | Satellite lightning imaging false alarm signal filtering method based on multistage denoising network | |
| He et al. | Feature aggregation convolution network for haze removal | |
| CN112102365A (en) | Target tracking method based on unmanned aerial vehicle pod and related device | |
| CN111784743A (en) | Infrared weak and small target detection method | |
| Jia et al. | An unsupervised change detection based on automatic relationship analysis | |
| Li et al. | Real Time Dust Detection with Image Pyramid and Improved ResNeSt | |
| CN114842012B (en) | Medical image small target detection method and device based on position awareness U-shaped network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20090408 Termination date: 20171109 |