CN113011440B - Coal-bed gas well site monitoring and re-identification technology - Google Patents
Coal-bed gas well site monitoring and re-identification technology Download PDFInfo
- Publication number
- CN113011440B CN113011440B CN202110296521.6A CN202110296521A CN113011440B CN 113011440 B CN113011440 B CN 113011440B CN 202110296521 A CN202110296521 A CN 202110296521A CN 113011440 B CN113011440 B CN 113011440B
- Authority
- CN
- China
- Prior art keywords
- distance
- sample
- samples
- cluster
- feature
- 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
- 238000012544 monitoring process Methods 0.000 title claims abstract description 20
- 238000005516 engineering process Methods 0.000 title claims abstract description 11
- 238000010168 coupling process Methods 0.000 claims abstract description 21
- 230000008878 coupling Effects 0.000 claims abstract description 20
- 238000005859 coupling reaction Methods 0.000 claims abstract description 20
- 238000004364 calculation method Methods 0.000 claims abstract description 14
- 238000011176 pooling Methods 0.000 claims abstract description 12
- 238000013507 mapping Methods 0.000 claims abstract description 4
- 239000000523 sample Substances 0.000 claims description 51
- 238000000034 method Methods 0.000 claims description 23
- 239000011159 matrix material Substances 0.000 claims description 11
- 239000013074 reference sample Substances 0.000 claims description 10
- 230000000694 effects Effects 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 7
- 239000013598 vector Substances 0.000 claims description 6
- 230000007246 mechanism Effects 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000000007 visual effect Effects 0.000 claims description 2
- 238000000354 decomposition reaction Methods 0.000 claims 1
- 238000000605 extraction Methods 0.000 abstract description 6
- 238000001514 detection method Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 239000003245 coal Substances 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000011840 criminal investigation Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005553 drilling Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 239000003208 petroleum Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
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
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Probability & Statistics with Applications (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a coal-bed gas well site monitoring and re-identification technology. Firstly, constructing a feature extraction network, replacing average pooling with weighted pooling, and performing classified calculation as inner product calculation between feature attention heat maps to extract the significant features of the target; secondly, carrying out corresponding feature coupling clustering on the features, randomly selecting high-confidence ID samples to form a clustering center, establishing a relative distance threshold of heterogeneous samples, calculating the distance between the features and the clustering center, and distinguishing target features from background interference and heterogeneous features to obtain feature coupling clustering loss; and finally, mapping the reference, similar and heterogeneous samples into feature spaces by using a three-configuration network, and comparing the similarity. The invention can better re-identify the target crossing the camera, can well improve the characteristic identification capability, and improves the accuracy and the robustness of re-identification.
Description
Technical Field
The invention relates to a coal-bed gas well site monitoring and re-identification method technology, and belongs to the technical field of computer vision and image processing.
Background
With the rapid development of computer software and hardware, high-performance computers and camera terminals are very popular, and are widely applied to urban intelligent monitoring. With the development and progress of society, the automobile popularity is higher and higher, and for urban safety, a need for an image analysis technology under monitoring and a target recognition technology are the focus of attention. Pedestrian re-identification has general application in video monitoring, intelligent transportation and city computing, and can be rapidly found, positioned and tracked in large-scale monitoring videos. Unlike pedestrian detection, tracking or classification, pedestrian re-recognition may be found as an example-level object search problem. The pedestrian re-recognition problem is a search problem of judging whether a pedestrian image photographed in a non-overlapping area belongs to the same person or not in a traffic monitoring scene in a specific range. At present, monitoring cameras exist in relatively developed traffic areas (crossroads, accident-prone areas, road sections with large traffic flow and the like), and the monitoring cameras are very valuable in researching how to better utilize the cameras for traffic supervision, criminal investigation and the like. At present, related researches are mainly carried out by three methods: the first method is mainly used for re-identifying pedestrians by combining with a sensor, the method is greatly influenced by weather interference and distance factors, and the corresponding hardware cost is also very high, so that the real requirements cannot be met. The second method mainly uses a manual feature extraction mode, and the accuracy cannot be improved because some manual features are extracted by using some extraction algorithms. The third method based on deep learning is generated, the feature extraction accuracy is higher, and the recognition accuracy is improved. The three-configuration network is an improvement and promotion of the twin network, has remarkable promotion on discrimination capability, and has good effect on the identification detection field for the extraction of the salient features under the condition of complex background interference.
Disclosure of Invention
The patent aims to overcome the defects of the prior art introduced above, aims at identifying the same target of a three-configuration network crossing a camera under a complex environment aiming at the corresponding pedestrian characteristic distinguishing capability, and provides a coal-bed gas well site monitoring re-identification method technology.
The technical scheme adopted by the invention is as follows: and constructing a three-configuration network based on feature coupling clustering to carry out target re-identification. For image data, the manual feature extraction is to extract the semantic features from the lower layer at present, focus on the feature information such as pedestrian contours, textures, colors and the like to identify, and can extract the semantic information of the higher layer including spatial characteristics, time and position information through a neural network. The high-level semantic information is very important for the feature expression of the target pedestrian, and is well complementary with the extracted bottom-layer semantic information, so that corresponding re-identification is better realized. The invention can fully utilize the spatial characteristics to acquire the spatial semantic information of the corresponding target pedestrian, and the advantage of complementation with the low-level semantic information is utilized to improve the discrimination capability of the method for the same target crossing the camera in the complex environment, thereby improving the reliability and the robustness of re-identification.
A coal-bed gas well site monitoring and re-identification method technology comprises the following steps:
firstly, building an embedded network structure, adding a weighted attention pooling layer to replace an average pooling layer in the embedded network structure, and regarding classification calculation as inner product calculation between the output pedestrian characteristic attention heat maps to extract the saliency characteristics of a target pedestrian;
and secondly, randomly selecting a plurality of different ID samples with higher confidence coefficient as a clustering center by the characteristic coupling clustering method, establishing a relative distance threshold of the heterogeneous samples, calculating the space Markov distance between the characteristic points and the clustering center, correspondingly clustering the characteristic points of the target pedestrians, the background interference points and different pedestrians to obtain corresponding characteristic coupling clustering loss, and calculating the similarity by a comparator.
And thirdly, mapping the reference sample, the similar sample and the heterogeneous sample images into a feature space by integrating three embedded networks into a three-configuration network framework.
The invention has the advantages that the invention provides a coalbed methane well site monitoring re-identification method technology based on characteristic coupling clustering, fully utilizes the advantages of similar characteristic coupling clustering, simultaneously utilizes a weighted attention pool to replace an average pooling layer to extract corresponding obvious characteristic information of a target pedestrian and construct a characteristic heat map, activates the characteristic of positive correlation of a specific target through a weighted attention pool module, and utilizes the spatial characteristic of the obvious characteristic to endow a corresponding channel with better weight deviation so as to realize better pedestrian re-identification.
The invention can better identify the same pedestrian crossing the camera, can well improve the characteristic identification capability, and improves the accuracy and the robustness of pedestrian re-identification.
Drawings
Fig. 1 is a schematic diagram of a three-configuration network pedestrian re-identification structure based on feature coupling clustering.
FIG. 2 is a graph of reorder versus reorder accuracy on a MarKet1501 dataset.
FIG. 3 is a graph of the present invention versus other classical algorithms for accurate curve comparison on a MarKet1501 dataset.
FIG. 4 is a graph of confidence accuracy versus reordering on a MarKet1501 dataset.
FIG. 5 is a graph comparing confidence accuracy of the present invention with that of the classical algorithm on the MarKet1501 dataset.
FIG. 6 is a graph of the characteristic coupling clustering effect in the present invention.
FIG. 7 is a diagram of the weighted attention pool structure in the present invention.
FIG. 8 the pedestrian recognition effect of the invention in a coal bed gas well recognition mine FIG. 1.
FIG. 9 the pedestrian recognition effect of the invention in a coal bed gas well recognition mine FIG. 2.
Detailed Description
The following describes the embodiments of the present invention further with reference to the drawings and technical schemes.
A coal-bed gas well site monitoring and re-identification method technology comprises the following steps:
first, an embedded network is constructed, sample characteristics are extracted, the function of each sub-network is used for extracting the characteristics of related images, a convolution network is used for extracting the salient characteristics of reference pedestrians, similar samples and heterogeneous samples, and the salient characteristics are projected to a characteristic space.
For the first step of embedded network architecture we replace the average pooling layer with a weighted attention pool, the architecture is shown in figure 1. The invention consists of two network structures, namely an embedded network and a three-configuration network. Matrix representation of the original averaged pooling layer pooling using weighted attention pools and representation using matrices of second order pools, followed by a downgrading of the weighted matrices, and then a bottom-up and top-down attention interpretation using the downgraded results. The specific implementation mode is as follows:
wherein X ε R n*f ,W∈R f*f X represents the feature matrix, n represents the number of samples, and f represents the number of channels
Then the lowest rank approximation W is performed k Weight, W k ∈R f*f :
Wherein a, b E R f*1 Where a, b represents the rank of the weight matrix and k represents the number of categories.
We choose to simulate the visual salience of class-specific "top-down" attention mechanisms from bottom to top, independent of class. One of the attention parameter vectors is usually forced to be class independent, so we will b k =b, so our last low rank attention pool mechanism module is expressed as:
Grade attention (X,a,b)＝(Xa k ) T (Xb k ) (equation 3)
Finally, continuing to simplify as:
wherein T is k ＝Xa k ,h＝Xb。
This is equivalent to a top-down (special class) T k And bottom-up based saliency h-attention mapping. Our previous average or maximum pool was to reduce the feature matrix to R f*1 Then the classification number is generated by FC (fully connected layer) weight vector W processing. We consider the attention heat map as Xa ε R n*1 B is considered as a classifier that notices pool features, so we finally Grade attention Can be seen as the inner product between two attention heat maps defined over all n spatial locations.
In our experiments we predicted a single channel bottom-up saliency map identical to the last feature map and used a linear classifier on top, as well as we generated n 1 ×n 2 The x num_class dimension is a top-down attention map Xa, where num_class is the number of categories. We multiply the two attention attempts by a spatial average, generating an output prediction of the num_class dimension (Xa k ) T (Xb k ) These operations correspond to first combining features with salient features X T (Xb)) and then passed through a classifier (a (X) T (Xb))。
Second, the reference sample, homogeneous sample and heterogeneous sample images are mapped into the feature space by merging the three embedded networks into a three-configuration network framework. According to the characteristic coupling clustering method, different ID samples with higher confidence coefficient are randomly selected to serve as clustering centers, the characteristic points of the target pedestrians, background interference points and different pedestrians can be correspondingly clustered and distinguished by establishing a relative distance threshold value of the heterogeneous samples and calculating the space mahalanobis distance between the characteristic points and the clustering centers, corresponding characteristic coupling clustering loss is obtained, and then similarity is calculated through a comparator.
In the space feature set, the similar and heterogeneous are divided by using feature coupling clustering, so that the later re-identification is facilitated, and the implementation process is as follows:
randomly selecting different ID samples with higher confidence as Cluster centers Cluster center However, each selection of the cluster center can only select one of all IDs of the sample as the cluster center, and for each cluster center, the spatial Euclidean distance between the surrounding samples and the cluster center is calculated, and a negative sample relative distance threshold d is defined n Negative sample d n Defined as the shortest distance between different cluster centers, when we calculate the cluster center distance and other samples S i At the distance between them, if dist (cluster center ,F i )>d n At the time, where F i Representing the bottom i feature samples toThe clustering center is taken as the center, and d is taken as n The distance is a radius, a cluster is divided, candidate feature points are clustered, and min (cluster center ,F i ))>d n Continuing the clustering in the same manner to attach a pseudo tag to all samples that are similar and dissimilar, wherein we use a feature-coupled cluster penalty to calculate the corresponding distance, unlike K-means clustering, which automatically marks part of the data as similar high confidence, which is to divide all data into clusters, which marks similar samples as the same ID. During the clustering process, there may be many candidate feature points satisfying this constraint, from which we choose the most differentiated point as the cluster center sample point.
min(Dist M (f i ,f j ))>d n
Wherein f i ,f j All belong to feature images in the to-be-searched and image library, cluster center Is a cluster center, d n Is a threshold.
Spatial feature points are calculated using mahalanobis distance:
the mahalanobis distance between the feature points is calculated as follows:
if the covariance matrix is a unit vector, the dimensions are independently and uniformly distributed, so that the mahalanobis distance is the euclidean distance, the calculated euclidean distance is not necessarily similar due to the fact that the euclidean distance has larger errors in the flow pattern space under different dimensions, and later, people want to adopt a normalized euclidean distance to solve the problem of inconsistent dimension scale, but as the sample size increases, the difference of the sample distribution is larger, and then the normalized euclidean distance can not be well removed, so that we use the euclidean distance to calculate the space distance; the mahalanobis distance is specifically derived as follows:
the feature points are rotated to a Principal Component (PCA) first, so that the dimensions are independent of each other and become linear, and new coordinates are generated at the moment.
The mahalanobis distance is the euclidean distance after rotation change, and the calculation of the mahalanobis distance formula 6 is as follows:
when Dist Cluster< d n When the feature points are clustered, d, taking the clustering center as the center n Distance, conversely, we push distance far d n Is a distance of (3). For TripletNet (three configuration network), since three pairs of samples, i.e. sample x, are input, positive sample x + Negative sample x - . We select positive and negative samples x from the positive and negative sample sets + ,x - Thus, by the feature coupled clustering described above, we estimate the average of the cluster centers for each positive sample:
wherein N is + Is the positive sample number
The resulting relative distance calculation for positive and negative samples can be reflected by the following equation:
where i belongs to the positive sample set and j belongs to the negative sample set, they satisfy the relationship:
Dist postive +θ≤Dist negtive (equation 10)
Wherein θ is a distance deviation, θ < d n
A coupled cluster loss can be obtained:
wherein x is i Is a positive sample set, x j Is a negative sample set, N + Is the positive number of samples.
If it isIf the value is less than or equal to 0, the partial derivative of the positive and negative samples is 0, otherwise, the partial derivative of the positive and negative samples is
The constraint of this Loss function is the same as the Triplet Loss, so that the intra-class distance should be much smaller than the inter-class distance. However, this loss can be a distance change of the sample for multiple samples, and the distance between the cluster center and the sample is calculated and is not random.
Thirdly, after the three-configuration network structure is adopted, after the characteristic information is obtained through calculation, a reference sample is calculated through the following calculation formula, and the distance comparison function is solved by the positive sample and the negative sample in pairs of 2 norms:
the triplenet () represents a characteristic distance function obtained after a three-configuration network structure, and if the result of the obtained cmp comparator is greater than 0, we can know that the positive and negative samples have higher similarity or are similar to the reference sample, otherwise, if the difference between the positive and negative samples is greater than or equal to 0, we can know that the positive and negative samples have good recognition and classification effects, and compared with the reference sample x, the positive sample x + Pull-up, negative sample x - Is pushed away.
The invention utilizes the MarKet1501 data set to carry out training and experimental data comparison on the network structure, and the invention can verify average accuracy and confidence accuracy in complex environments by drawing an average accuracy curve comparison graph of the invention and a classical algorithm on the MarKet1501 data set, a characteristic coupling process graph, a network structure graph and a detection effect graph of the invention in a related coal-bed gas well.
In the MarKet1501 dataset, we use the average accuracy and confidence accuracy to conduct the re-recognition evaluation. The average accuracy is an evaluation index commonly used in object detection and label image classification, and because more than one label is used in the multi-classification task, the classification standard of a single common single label image, namely mAP, namely average accuracy AP in the multi-classification task is not used for summation and then average. Average accuracy mean vs. graph for various identification methods are shown in fig. 2 and 3. Compared with other classical algorithms, the coal-bed gas well site monitoring re-identification method has better accuracy. Confidence accuracy refers to the probability of having the correct result with the top confidence in the search results, i.e., the accuracy of being able to find image tags similar to the target pedestrian under different cameras. Compared with other algorithms, the coal-bed gas well site monitoring re-identification method has good confidence accuracy, as shown in fig. 4 and 5, so that whether the same pedestrian is identified under the cross-camera can be accurately identified.
The characteristic clustering coupling algorithm and the weighted attention pool used in the invention are shown in the process diagrams shown in fig. 6 and 7, which can show that the characteristic clustering coupling algorithm and the weighted attention pool in the invention can well improve the recognition accuracy and can better promote the re-recognition. The invention is applied to practical projects, pedestrian identification in coal-bed gas wells and petroleum drilling, can effectively help the improvement of yield and capacity monitoring, intelligent security protection, security accident monitoring and the like, and has the effect shown in figures 8 and 9, and the good performance and accuracy of the invention in re-identification tasks are shown.
Claims (1)
1. The coal-bed gas well site monitoring and re-identifying technology is characterized by comprising the following steps of:
firstly, extracting sample characteristics by constructing an embedded network structure, adding a weighted attention pooling layer to replace an average pooling layer in the embedded network structure, and regarding classification calculation as inner product calculation between the attention heat maps of the pedestrian characteristics output by classification calculation to extract the salient characteristics of the target pedestrian;
the network structure consists of an embedded network and a three-configuration network; matrix representation of the original average pooling layer pooling using weighted attention pools and representation using matrices of second order pools, then a downgrading decomposition of the weighted matrices, and then a bottom-up and top-down attention interpretation using the downgraded results;
the specific implementation mode is as follows:
wherein X ε R n*f ,W∈R f*f X represents a feature matrix, n represents the number of samples, and f represents the number of channels;
for the above W k The weight matrix performs the lowest rank approximation, W k ∈R f*f :
Wherein a, b E R f*1 Wherein a, b represents the row and column of the weight matrix, and k represents the number of categories;
selecting a class-specific "top-down" attention mechanism that simulates the visual salience from bottom to top, independent of the class; forcing one of the attention parameter vectors to be class independent, thus letting b k =b, so the last low rank attention pool mechanism module is expressed as:
Grade attention (X,a,b)＝(Xa k ) T (Xb k ) (equation 3)
Continuing to reduce to:
wherein T is k ＝Xa k ,h＝Xb；
Equation 4 corresponds to top-down T k And solving an inner product between the saliency h-based attention maps from bottom to top; average or maximum pool is to reduce the feature matrix to R f*1 Then generating a classification number through the weight vector W processing of the FC; the method regards the attention heat map as Xa epsilon R n*1 B is considered as a classifier that notices pool features, so the final Grade attention What is considered to be the inner product between the two attention heat maps defined over all n spatial locations;
by predicting the first single channel bottom-up saliency map, which is identical to the last feature map, and using a linear classifier on top, n is also generated 1 ×n 2 A top-down attention map Xa of the x num_class dimension, where num_class is the number of categories; multiplying the two attention attempts by a spatial average generates an output prediction in the num_class dimension (Xa k ) T (Xb k ) These operations correspond to first combining features with salient features X T (Xb) and then passed through a classifier a (X T (Xb))；
Secondly, mapping the reference sample, the similar sample and the heterogeneous sample images into a feature space by integrating three embedded networks into a three-configuration network framework; randomly selecting a plurality of different ID samples with higher confidence coefficient as a clustering center by a feature coupling clustering method, establishing a relative distance threshold of heterogeneous samples, calculating the space Markov distance between a feature point and the clustering center, correspondingly clustering the feature point of a target pedestrian, a background interference point and different pedestrians to distinguish the feature points, obtaining corresponding feature coupling clustering loss, and calculating similarity by a comparator; the specific implementation is as follows:
randomly selecting different ID samples with higher confidence as Cluster centers Cluster center However, each time a cluster center is selected, only one of all IDs of the sample can be selected as the cluster center, for each cluster center, the spatial Euclidean distance between the surrounding samples and the cluster center is calculated, and a negative sample relative distance threshold d is defined n Negative sample d n Defined as the shortest distance between different cluster centers, when the cluster center distance is calculated to be the same as other samples S i At the distance between them, if dist (Cluster center ,F i )>d n At that time, the cluster center will be centered, where F i Represents the ith feature set, d n The distance is a radius, a Cluster is divided, candidate feature points are clustered, and min (dist (Cluster) center ,F i ))>d n Continuing the division of the cluster in the same manner by taking the furthest point of the cluster as a cluster center, so as to attach a corresponding similar and dissimilar pseudo tag to all the samples, wherein a characteristic coupling cluster loss is used for calculating the corresponding distance, and the difference is that the data coupling cluster automatically marks partial data as similar high confidence, the K-means cluster divides all the data into a plurality of clusters, and the cluster center marks similar samples as the same ID; in the clustering process, a plurality of candidate feature points probably meet the constraint, and the most differentiated point is selected as the clustering pointA heart sample point;
min(Dist M (f i ,f j ))＞d n
wherein f i ,f j All belong to feature images in the to-be-searched and image library, cluster center Is a cluster center, d n Is a threshold;
spatial feature points are calculated using mahalanobis distance:
the mahalanobis distance between the feature points is calculated as follows:
if the covariance matrix is a unit vector, the respective dimensions are independently and uniformly distributed, so that the mahalanobis distance is the euclidean distance, the calculated euclidean distance is not necessarily similar due to the fact that the euclidean distance has larger errors in the flow pattern space under different dimensions, the problem of inconsistent dimension scale is solved by adopting a normalized euclidean distance later, but the normalized euclidean distance cannot well eliminate outlier after the variability of sample distribution is larger and larger along with the increase of the sample quantity, so that the space distance is calculated by adopting the mahalanobis distance; the mahalanobis distance is specifically as follows:
the feature points are rotated to the main component, so that the dimensions are independent of each other and independent of each other, and new coordinates are generated at the moment;
the mahalanobis distance is the euclidean distance after rotation change, and the mahalanobis distance formula 8 becomes:
when Dist Cluster< d n When the feature points are clustered, d, taking the clustering center as the center n Distance, otherwise, distance is pushed far d n Is a distance of (2); for a three configuration network triplenet, since three pairs of samples, reference sample x, positive sample x, are input + Negative sample x - The method comprises the steps of carrying out a first treatment on the surface of the Selecting positive sample and negative sample x from the positive sample set and the negative sample set + ,x - Thus, the average value of the clustering center of each positive sample is estimated through the characteristic coupling clustering:
wherein N is + Is the positive number of samples;
the resulting relative distance calculation for positive and negative samples is reflected by the following equation:
where i belongs to the positive sample set and j belongs to the negative sample set, they satisfy the relationship:
Dist postive +θ≤Dist negtive (equation 10)
Wherein θ is a distance deviation, θ < d n ；
Obtaining a coupling clustering loss:
wherein x is i Is a positive sample set, x j Is a negative sample set, N + Is a positive sampleThe number of the samples;
if it isIf the value is less than or equal to 0, the partial derivative of the positive and negative samples is 0, otherwise, the partial derivative of the positive and negative samples is
The constraint of this Loss function is the same as the Triplet Loss, so that the intra-class distance should be much smaller than the inter-class distance;
thirdly, after the three-configuration network structure is adopted, after the characteristic information is obtained through calculation, a reference sample is calculated through the following calculation formula, and the distance comparison function is solved by the positive sample and the negative sample in pairs of 2 norms:
the tripletNet (x) represents a characteristic distance function obtained after a three-configuration network structure, and the obtained cmp comparator has higher similarity of positive and negative samples or is similar to a reference sample if the result of the comparator is more than 0, otherwise, if the difference between the positive and negative samples is more than or equal to 0, the positive and negative samples are recognized to have good classification effect, and compared with the reference sample, the positive sample x + Pull-up, negative sample x - Is pushed away.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110296521.6A CN113011440B (en) | 2021-03-19 | 2021-03-19 | Coal-bed gas well site monitoring and re-identification technology |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110296521.6A CN113011440B (en) | 2021-03-19 | 2021-03-19 | Coal-bed gas well site monitoring and re-identification technology |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113011440A CN113011440A (en) | 2021-06-22 |
| CN113011440B true CN113011440B (en) | 2023-11-28 |
Family
ID=76403360
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110296521.6A Active CN113011440B (en) | 2021-03-19 | 2021-03-19 | Coal-bed gas well site monitoring and re-identification technology |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113011440B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114333039B (en) * | 2022-03-03 | 2022-07-08 | 济南博观智能科技有限公司 | Method, device and medium for clustering human images |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109800794A (en) * | 2018-12-27 | 2019-05-24 | 上海交通大学 | A kind of appearance similar purpose identifies fusion method and system across camera again |
| CN110110642A (en) * | 2019-04-29 | 2019-08-09 | 华南理工大学 | A kind of pedestrian's recognition methods again based on multichannel attention feature |
| CN111476168A (en) * | 2020-04-08 | 2020-07-31 | 山东师范大学 | Cross-domain pedestrian re-identification method and system based on three stages |
| CN112101150A (en) * | 2020-09-01 | 2020-12-18 | 北京航空航天大学 | Multi-feature fusion pedestrian re-identification method based on orientation constraint |
| CN112270355A (en) * | 2020-10-28 | 2021-01-26 | 长沙理工大学 | Active safety prediction method based on big data technology and SAE-GRU |
| WO2021029881A1 (en) * | 2019-08-14 | 2021-02-18 | Google Llc | Systems and methods using person recognizability across a network of devices |
| CN112507941A (en) * | 2020-12-17 | 2021-03-16 | 中国矿业大学 | Cross-vision field pedestrian re-identification method and device for mine AI video analysis |
| CN113158901A (en) * | 2021-04-22 | 2021-07-23 | 天津大学 | Domain-adaptive pedestrian re-identification method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220147735A1 (en) * | 2020-11-10 | 2022-05-12 | Nec Laboratories America, Inc. | Face-aware person re-identification system |
-
2021
- 2021-03-19 CN CN202110296521.6A patent/CN113011440B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109800794A (en) * | 2018-12-27 | 2019-05-24 | 上海交通大学 | A kind of appearance similar purpose identifies fusion method and system across camera again |
| CN110110642A (en) * | 2019-04-29 | 2019-08-09 | 华南理工大学 | A kind of pedestrian's recognition methods again based on multichannel attention feature |
| WO2021029881A1 (en) * | 2019-08-14 | 2021-02-18 | Google Llc | Systems and methods using person recognizability across a network of devices |
| CN111476168A (en) * | 2020-04-08 | 2020-07-31 | 山东师范大学 | Cross-domain pedestrian re-identification method and system based on three stages |
| CN112101150A (en) * | 2020-09-01 | 2020-12-18 | 北京航空航天大学 | Multi-feature fusion pedestrian re-identification method based on orientation constraint |
| CN112270355A (en) * | 2020-10-28 | 2021-01-26 | 长沙理工大学 | Active safety prediction method based on big data technology and SAE-GRU |
| CN112507941A (en) * | 2020-12-17 | 2021-03-16 | 中国矿业大学 | Cross-vision field pedestrian re-identification method and device for mine AI video analysis |
| CN113158901A (en) * | 2021-04-22 | 2021-07-23 | 天津大学 | Domain-adaptive pedestrian re-identification method |
Non-Patent Citations (7)
| Title |
|---|
| A Cross-Camera Multi-Face Tracking System Based on Double Triplet Networks;Guoyin Ren，等;《IEEE Access》;第9卷;第43759-43774页 * |
| Adaptive energy-efficient clustering path planning routing protocols for heterogeneous wireless sensor networks;Aslam, Muhammad，等;《SUSTAINABLE COMPUTING-INFORMATICS & SYSTEMS》;第12卷;第57-71页 * |
| Clustering-Guided Pairwise Metric Triplet Loss for Person Reidentification;Weiyu Zeng，等;《IEEE Internet of Things Journal》;第9卷(第16期);第15150-15160页 * |
| Distribution Context Aware Loss for Person Re-identification;Zhigang Chang，等;《2019 IEEE Visual Communications and Image Processing (VCIP)》;第1-4页 * |
| 基于质量增强和注意力机制的井下行人重识别算法;丁嘉婕;《中国优秀硕士学位论文全文数据库 信息科技辑》(第2期);I138-1902 * |
| 密集场景下基于轮廓运动的人群行为分析;唐志鸿，等;《江苏通信》;第37卷(第4期);第102-107页 * |
| 融合随机擦除和残差注意力网络的行人重识别;厍向阳，等;《计算机工程与应用》;第58卷(第3期);第215-221页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113011440A (en) | 2021-06-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wu et al. | Deep learning-based methods for person re-identification: A comprehensive review | |
| Sheng et al. | Crowd counting via weighted VLAD on a dense attribute feature map | |
| CN112101150B (en) | Multi-feature fusion pedestrian re-identification method based on orientation constraint | |
| CN111666843B (en) | Pedestrian re-recognition method based on global feature and local feature splicing | |
| CN110942025A (en) | Unsupervised cross-domain pedestrian re-identification method based on clustering | |
| Bąk et al. | Boosted human re-identification using Riemannian manifolds | |
| US9275269B1 (en) | System, method and apparatus for facial recognition | |
| CN111709311B (en) | Pedestrian re-identification method based on multi-scale convolution feature fusion | |
| CN110717411A (en) | Pedestrian re-identification method based on deep layer feature fusion | |
| CN111507217A (en) | Pedestrian re-identification method based on local resolution feature fusion | |
| US20150317511A1 (en) | System, method and apparatus for performing facial recognition | |
| Zhang et al. | Hierarchical building recognition | |
| Everingham et al. | Pascal visual object classes challenge results | |
| Chen et al. | Vehicle re-identification using distance-based global and partial multi-regional feature learning | |
| CN113361464B (en) | Vehicle weight recognition method based on multi-granularity feature segmentation | |
| WO2006073647A2 (en) | Method and apparatus for unsupervised learning of discriminative edge measures for vehicle matching between non-overlapping cameras | |
| Tang et al. | Multi-modal metric learning for vehicle re-identification in traffic surveillance environment | |
| CN108537137B (en) | Multi-modal biological characteristic fusion recognition method based on label identification correlation analysis | |
| Wang et al. | Compressed holistic convnet representations for detecting loop closures in dynamic environments | |
| Zheng et al. | Aware progressive clustering for unsupervised vehicle re-identification | |
| Khan et al. | Multi-person tracking based on faster R-CNN and deep appearance features | |
| Kobyshev et al. | Matching features correctly through semantic understanding | |
| Liu et al. | Urban area vehicle re-identification with self-attention stair feature fusion and temporal Bayesian re-ranking | |
| CN113011440B (en) | Coal-bed gas well site monitoring and re-identification technology | |
| Cai et al. | Patch-NetVLAD+: Learned patch descriptor and weighted matching strategy for place recognition |
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 |