CN105354273A - Method for fast retrieving high-similarity image of highway fee evasion vehicle - Google Patents
Method for fast retrieving high-similarity image of highway fee evasion vehicle Download PDFInfo
- Publication number
- CN105354273A CN105354273A CN201510714633.3A CN201510714633A CN105354273A CN 105354273 A CN105354273 A CN 105354273A CN 201510714633 A CN201510714633 A CN 201510714633A CN 105354273 A CN105354273 A CN 105354273A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- feature
- image
- quick
- high similarity
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/757—Matching configurations of points or features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/584—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of vehicle lights or traffic lights
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/59—Context or environment of the image inside of a vehicle, e.g. relating to seat occupancy, driver state or inner lighting conditions
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Library & Information Science (AREA)
- Databases & Information Systems (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention discloses a method for fast retrieving a high-similarity image of a highway fee evasion vehicle. The method comprises: performing feature extraction and recognition on a collected sample image of an illegal vehicle by using a convolutional neural network model of a computer, constructing a corresponding K-d tree for the recognized feature, extracting a feature for a collected unknown image by using a fast high-similarity nearest neighbor search algorithm, and matching the feature of the unknown image with the feature of the sample image, to implement fast retrieval of a high-similarity image. According to the method provided by the present invention, better image feature expression can be obtained for retrieving a similar vehicle. Not only a retrieval result is high in similarity, but also retrieval is faster. Therefore, a function of image retrieval for cracking down on fee evasion, gaining evidence, cross-examination and a legal evidence chain can be sufficiently utilized, and behaviors of toll evasion can be fought and prohibited effectively.
Description
Technical field
The present invention relates to a kind of method of quick-searching highway fee evasion vehicle high similarity graph picture, belong to the technical field of traffic pattern identification.
Background technology
Under economic profit incentive, expense phenomenon evaded by national highway (except Hainan Province) ubiquity payment vehicle, various expense means of evading emerge in an endless stream, the hidden violent resistance against law phenomenon of the form of expression is rampant, rushes in the process escaped cause charge station managerial personnel to disable the severe event of even death at punching card.Restriction with some condition limited due to freeway management side, it is comparatively difficult that the illegal evidence obtaining of expense evaded by vehicle, weakens and hit the dynamics that illegal driver clique evades expense behavior, have influence on the regular fee operation activity of freeway management person.
Sum up investigation strike in the past and evade toll behavior, be mainly main by setting up the blacklist in fee evasion vehicle management system with experience, transfer corresponding fee evasion vehicle video image and capture image and analyse and compare, but this method exchange card is not changed trains, licence plate is existing also comparatively applicable.For both changing card, the detection inspection of changing again the fee evasion vehicle of licence plate just seemed that time-consuming difficulty of taking a lot of work was huge.Therefore must make full use of ITS Information technology and there is the excavation of large data depth, the data of magnanimity are analyzed to the feature of fast processing by intelligent Modeling, the image that application bayonet socket high-definition camera is captured is to license plate, vehicle color, vehicle, the vehicle attribute such as vehicle facial characteristics, analyze vehicle similar features through computing machine degree of depth learning training to mate, realize carrying out fast processing and retrieval to high similarity violation vehicle image, to car plate travel track and the vehicle travel track of evading expense vehicle, the time of carrying out, space, feature three dimension location, for public security system sternly hits the evidence that fee evasion behavior provides strong.
Realize the image retrieval of high similarity vehicle, just need that individual features extraction is carried out to picture material and realize classification and identify.What traditional image characteristics extraction great majority adopted is manually extract feature, such as object identification adopts scale invariant feature conversion (ScaleInvariantFeatureTransform, SIFT), recognition of face adopts local binary patterns (LocalBinaryPatterns, LBP), pedestrian detection adopts histograms of oriented gradients (HistogramofOrientedGradient, HOG) feature etc.But comprising the progressively increase of information along with image, a kind of effectively desirable features of engineer is expressed obviously more difficult, and need expend the longer time cycle.
Summary of the invention
The invention provides a kind of quick-searching highway fee evasion vehicle high similarity image method.Method of the present invention can obtain better image feature representation, for the retrieval carrying out similar vehicle, result for retrieval not only similarity is high, and retrieval is more quick, and then the effect of image retrieval in strike fee evasion evidence obtaining cross-examination legal argument chain can be given full play to, effectively toll behavior is evaded in strike and suppression.
Technical scheme of the present invention: a kind of method of quick-searching highway fee evasion vehicle high similarity graph picture, be characterized in: the convolutional neural networks model utilizing computing machine, Feature extraction and recognition is carried out to the sample image of the violation vehicle collected, and corresponding K-d tree structure is carried out to the feature identified, then utilize convolutional neural networks to carry out feature extraction to the unknown images collected and utilize quick high similarity nearest neighbor search algorithm to mate sample characteristics, to realize the quick-searching of high similarity graph picture.
The method of above-mentioned quick-searching highway fee evasion vehicle high similarity graph picture specifically comprises the following steps:
1. collect vehicle sample image in violation of rules and regulations, set up vehicle management database.
2. sample image is normalized, unified vehicle image form;
3. utilize convolutional neural networks model, Feature extraction and recognition is carried out to vehicle, obtains sample characteristics;
4. corresponding K-d is carried out to sample characteristics and set structure;
5. feature is extracted to the unknown images collected, set on basis at the K-d built, quick for these characteristic use high similarity nearest neighbor search algorithm is gone to mate with sample characteristics, to realize scheming to search figure effect;
6. according to the matching degree size that 5. step obtains, export similarity graph picture (arranging according to similarity size, time, path) according to time sequencing, finally carry out manual confirmation.
In the method for aforesaid quick-searching highway fee evasion vehicle high similarity graph picture, because current fee evasion vehicle fraudulent means is varied and by force disguised, the image retrieval of single features may miss similar vehicle, and vehicle image not high for similarity retrieval is collected into, lose practical value from the sample making to retrieve, therefore need to carry out drawing to the various features of vehicle carefully to determine to improve similarity comparison accuracy of detection.So described step 3. in extract feature comprise:
(1) characters on license plate feature.Car license recognition is an obvious and important feature in vehicle retrieval, serves larger effect for the high similarity probability of identification.
(2) vehicle feature.According to the current statistical conditions to highway fee evasion vehicle, same logistics company car of the same type accounts for major part in fee evasion vehicle, therefore needs to carry out corresponding classification to vehicle.
(3) car mark, body color feature.For fake-licensed car, now other features such as car mark, body color changed by car plate just becomes the main recognition feature of this vehicle, and therefore the good representation of car mark, body color feature is also most important.
(4) vehicle facial characteristics.Vehicle annual test, automobile interior, suspension member feature.For violation vehicle and fake-licensed car, now car plate, vehicle, car color, Che Biao etc. are almost identical, therefore can only be differentiated by vehicle facial characteristics, now vehicle annual test, and the thin feature such as automobile interior, suspension member just seems particularly important.
(5) external form, the facial characteristics of driver and conductor.By in violation of rules and regulations and the external form of fake license plate vehicle driver and conductor, facial characteristics obtain, by contrasting the inquiry of historical data, can substantially fix the driver and conductor of this car, then with current view data comparison, the driver and conductor implementing this car can be differentiated further.
In the method for aforesaid quick-searching highway fee evasion vehicle high similarity graph picture, described step 5. in quick high similarity nearest neighbor search algorithm be carry out by backtracking (Backtracking) operation the method improved when utilizing K-d to set to carry out neighbor searching.
In the method for aforesaid quick-searching highway fee evasion vehicle high similarity graph picture, described step 3. in the convolutional neural networks model that uses be obtain after carrying out convolutional neural networks training as sample set using the vector of the formation of input vector and desirable output vector, and before training, use little random number (ensureing that network can not enter state of saturation because weights are excessive) to carry out initialization to every weights of convolutional neural networks.
The concrete grammar of wherein training comprises the following steps:
The first step, gets a sample input convolutional neural networks from sample set;
Second step, calculates corresponding actual output.
3rd step, calculates the difference that actual output exports with corresponding ideal;
4th step, by the method backpropagation adjustment weight matrix of minimization error.
Compared with prior art, method of the present invention uses convolutional neural networks model image to be carried out to the extraction of feature, avoid explicit feature extraction, can implicitly learn from training data, because the neuron weights on same Feature Mapping face are identical, the convolutional neural networks that the present invention is used can collateral learning, reduce the complicacy of network, the sub sampling structure in employing time or space, displacement to a certain degree, yardstick, deformation robustness can be obtained, and input information and network topology structure can well be coincide.Method of the present invention can effectively prevent artificial extract feature complicacy and the problem such as time consuming nature.
And the present invention adopts quick approximate KNN searching algorithm, carry out k-d tree to the vehicle characteristics of convolutional neural networks model extraction to build, and by after improving nearest neighbor search algorithm, can solve and reduce higher-dimension k-d tree shortcoming consuming time in trace-back process, thus accelerate search time and ensure the search precision of being correlated with simultaneously.
Accompanying drawing explanation
Fig. 1 is basic step process flow diagram of the present invention;
Fig. 2 is the concept exemplary view of convolutional neural networks;
Fig. 3 is the algorithm flow that the embodiment of the present invention uses convolutional neural networks model extraction feature.
Embodiment
Below in conjunction with embodiment, the present invention is further illustrated, but not as the foundation limited the present invention.
Embodiment.A method for quick-searching highway fee evasion vehicle high similarity graph picture, its basic procedure as shown in Figure 1, comprises the following steps:
1. collect vehicle sample image in violation of rules and regulations, set up vehicle management database.
2. sample image is normalized, unified vehicle image form;
3. utilize convolutional neural networks model, Feature extraction and recognition is carried out to vehicle, obtains sample characteristics;
4. corresponding K-d is carried out to sample characteristics and set structure;
5. feature is extracted to obtain to the unknown images collected, set on basis at the K-d built, quick for these characteristic use high similarity nearest neighbor search algorithm is gone to mate with sample characteristics, to realize scheming to search figure effect;
6. according to the matching degree size that 5. step obtains, export similarity graph picture according to time sequencing, finally carry out manual confirmation.
Above 6 formation highway fee evasion vehicle feature recognition, coupling, retrieval, investigate, review, process the system flow that required chain of evidence obtains.
Wherein the machine learning training method of convolutional neural networks of the present invention (CNN) model is as follows.
The invention provides a kind of convolutional neural networks (CNN) model that adopts and carry out machine learning training, after computing machine obtains training sample, degree of depth study and uninterrupted training will be carried out by data operation platform.By all kinds of validity feature in the vehicle image that extracts, thus provide reliable matched rule for follow-up image retrieval.
CNN structure reduces the complexity of network model, and the quantity convolutional neural networks decreasing weights is the one of artificial neural network, has become the study hotspot of current speech analysis and field of image recognition.Its weights shared network structure makes it more to be similar to biological neural network, reduces the complexity of network model, decreases the quantity of weights.It is more obvious that this advantage shows when the input of network is multidimensional image, makes image directly as the input of network, can avoid feature extraction complicated in tional identification algorithm and data reconstruction processes.Convolutional network is a multilayer perceptron for identifying two-dimensional shapes and particular design, and the distortion of this network structure to translation, proportional zoom, inclination or his form altogether has height unchangeability.To be networks learn having under monitor mode these good performances, and the structure of network mainly contains partially connected and weights share two features, comprises the constraint of following form:
(1) feature extraction.Each neuron obtains the defeated people of cynapse from the local acceptance domain of last layer, thus forces it to extract local feature.Once a feature is extracted, as long as it is remained approx relative to the position of other features, its exact position just becomes so unimportant.
(2) Feature Mapping.Each computation layer of network is made up of multiple Feature Mapping, and each Feature Mapping is plane form.Neuron independent in plane shares identical synaptic weight collection under the constraints, and this version has following beneficial effect, the reduction (shared by weights and realize) of translation invariance and free parameter quantity.
(3) son sampling.And then a computation layer realizing local average and sample with son, thus the resolution reduction of Feature Mapping after each convolutional layer.This operation has the effect that the output of Feature Mapping is declined to the susceptibility of translation and other forms of distortion.
Convolutional neural networks is the neural network of a multilayer, and every layer is made up of multiple two dimensional surface, and each plane is made up of multiple independent neuron.The concept exemplary view of convolutional neural networks as shown in Figure 2.
Input picture is by carrying out convolution with three trainable wave filters with being biased, three Feature Mapping figure are produced at C1 layer after convolution, then four pixels often organized in Feature Mapping figure are sued for peace again, weighted value, be biased, obtained the Feature Mapping figure of three S2 layers by a Sigmoid function.These mapping graphs entered filtering again and obtained C3 layer.This hierarchical structure is the same with S2 again produces S4.Finally, these pixel values are rasterized, and connect into a vector and be input to traditional neural network, exported.
Usually, C layer is feature extraction layer, and each neuronic input is also connected with the local experiences of front one deck, and extracts the feature of this local, once after this local feature is extracted, the position relationship between it and other features is also decided thereupon; S layer is Feature Mapping layer, and each computation layer of network is made up of multiple Feature Mapping, and each Feature Mapping is a plane, and in plane, all neuronic weights are equal.Feature Mapping structure adopts sigmoid function that influence function core is little as the activation function of convolutional network, makes Feature Mapping have shift invariant.
In addition, because the neuron on a mapping face shares weights, thus decrease the number of freedom of network parameter, reduce the complexity that network parameter is selected.Each feature extraction layer (C-layer) in convolutional neural networks is used for asking the computation layer (S-layer) of local average and second extraction followed by one, and this distinctive twice feature extraction structure makes network have higher distortion tolerance when identifying to input amendment.
The specific implementation that the present invention carries out vehicle characteristics extraction is as follows:
All vehicle sample images are normalized to 320 × 320 onesize images, and now input layer is made up of 320 × 320 sensing nodes, receives original image.Then calculation process hockets between the sampling of Convolution sums, as shown in Figure 3.
First hidden layer carries out convolution, and it is made up of 8 Feature Mapping, and each Feature Mapping is made up of 271 × 271 neurons, and each neuron specifies the acceptance domain of 50 × 50.
Second hidden layer realizes son sampling and local average, namely carries out lower down-sampled.It is made up of 8 Feature Mapping equally, but its each Feature Mapping is made up of 136 × 136 neurons.Each neuron has the acceptance domain of 2 × 2, can train coefficient for one, can train biased and a sigmoid activation function for one.Coefficient and the neuronic operating point of biased control can be trained.
3rd hidden layer carries out second time convolution, and it is made up of 20 Feature Mapping, and each Feature Mapping is made up of 10 × 10 neurons.Each neuron in this hidden layer may have the Synaptic junction be connected with the several Feature Mapping of next hidden layer, and it operates in the mode similar to first convolutional layer.
4th hidden layer carries out the second second son sampling and the calculation of local average juice.It is made up of 20 Feature Mapping, but each Feature Mapping is made up of 64 × 64 neurons, and it is to operate to similar mode of sampling for the first time.
5th hidden layer realizes the final stage of convolution, and it is made up of 120 neurons, and each neuron specifies the acceptance domain of 5 × 5.
Be finally a full articulamentum, obtain output vector, the whole similar features of the vehicle namely extracted.
Train vehicle characteristics algorithm specific as follows about CNN.
Convolutional network is a kind of mapping being input to output in itself, it can learn the mapping relations between a large amount of constrained input, and without any need for the accurate mathematic(al) representation between input and output, as long as trained convolutional network by known pattern, network just has the mapping ability between inputoutput pair.What convolutional network performed is have tutor to train, thus its sample set be by shape as: the vector of (input vector, desirable output vector) is to forming.All these vectors are right, should be all to derive from the actual " RUN " result that network is about to the system of simulation.They can gather from actual motion system.Before starting training, all power all should carry out initialization with some different " little random number "." little random number " is used for ensureing that network can not enter state of saturation because weights are excessive, thus causes failure to train, and " difference " is used for ensureing that network can normally learn.In fact, if with identical several deinitialization weight matrixs, then network impotentia study.
Training algorithm mainly comprises two benches, four step rule:
First stage, forward direction.The first step gets a sample (X, Yp) from sample set, and wherein X is the feature of vehicle extraction.X is inputted network; Second step calculates corresponding actual output Op.
In this stage, information through conversion step by step, is sent to output layer from input layer.This process is also the process that network performs during normal operation after completing training.In the process, what network performed is calculate (be in fact exactly input and the weight matrix phase dot product of every layer, obtain last Output rusults);
Subordinate phase, back-propagation.3rd step calculation actual output Op and corresponding ideal export the difference of Yp; 4th step is by the method backpropagation adjustment weight matrix of minimization error.
The relevant information of vehicle in image just can be identified by above-mentioned steps convolutional neural networks, as license plate number, car mark model, vehicle color, vehicle, vehicle annual inspection mark, the relevant informations such as automobile interior.Finally by the information of vehicles that identifies as license plate number Zhejiang A12345 etc., car mark is as east wind, Mitsubishi etc., body color is as sapphire blue, dark green etc., vehicle is as truck, car etc., and the non-digitalization information unification such as vehicle annual test time and automobile interior information convert digital information to according to same format and store.So often open the high dimension vector that vehicle image just can obtain a relevant dimension after convolutional neural networks is trained.High dimension vector is that follow-up image retrieval makes coupling foundation.
Quick high similarity nearest neighbor search algorithm (FLANN) is adopted to realize as follows with the method for scheming to search figure.Next just need to carry out image retrieval after carrying out feature extraction training to image, to realize quick high similarity to scheme to search the function of figure.The vehicle characteristics dimension that degree of depth study is above extracted is usually all higher, and what traditional force search carried out is linear sweep, at this moment will calculate input example and each trains the distance of example, and when training set is very large, very consuming time, this method is infeasible.In order to improve the search efficiency of conventional linear scan method, the feature that vehicle training sample extracts uses special storage organization to store training data by the present invention, to reduce the number of times calculating distance.What FLANN of the present invention taked is carry out the structure of k-d tree to high dimensional feature and carry out corresponding improvement in the later stage to it.Specifically comprise the content of six aspects.
1, higher-dimension k-d sets and builds:
Kd-Tree, i.e. K-dimensionaltree are binary trees, and what store in tree is some K dimension data.The set of a K dimension data builds the division that a Kd-Tree represents the K dimension space to this K dimension data set formation, the hypermatrix region of a K dimension that each node namely in tree is just corresponding.For the such binary tree of Kd-tree, first we need to determine how to divide left subtree and right subtree, and namely what a K dimension data be divided into left subtree or right subtree according to.Dimension partitioning standards adopted in the present invention is varimax, when namely we select dimension to divide at every turn, all selects to have maximum variance dimension.Determine partition dimension just to need to consider how to divide in this dimension guarantee to divide the quantity of two subclass obtained as far as possible equal, the node number namely in left subtree and right subtree is as far as possible equal.We adopt feature array intermediate value to divide in the present invention, and what some dimensions were less than intermediate value is divided in subset A, is greater than being divided in subset B of intermediate value.Two that obtain so sub-collective data numbers are just substantially identical.
2, the developing algorithm of Kd-Tree:
(1) in the set of K dimension data, select the dimension k with maximum variance, in this dimension, then select intermediate value m to be that pivot divides this data acquisition, obtain two subclass; Create a tree node node, for storing simultaneously.
(2) all subclass two subclass repeated to the process of (1) step, until only all can not be further subdivided into; If when certain subclass can not divide again, then the data in this subclass are saved in leafy node (leafnode).
3, Kd-Tree carries out arest neighbors and searches:
After building a Kd-Tree, just need to utilize Kd-Tree to carry out arest neighbors and search, method is as follows:
(1) by data query Q from root node, access Kd-Tree downwards according to the comparative result of Q and each node, until reach leafy node.
Wherein the value compared in the k dimension that refers to and to be corresponded to by Q in node of Q and node compares with m, if Q (k) is <m, then accesses left subtree, otherwise accesses right subtree.When reaching leafy node, calculate the distance between data that Q and leafy node are preserved, record the data point that minor increment is corresponding, be designated as current " nearest neighbor point " Pcur and minor increment Dcur.
(2) carry out recalling (Backtracking) operation, this operation in order to find from Q more close to " nearest neighbor point ".Namely whether also have in the branch judging not accessed mistake from Q more close to point, the distance between them is less than Dcur.
If the distance between the branch of the not accessed mistake under Q and its father node is less than Dcur, then think to exist in this branch from P more close to data, enter this node, carry out the search procedure that (1) step is the same, if find nearer data point, then be updated to current " nearest neighbor point " Pcur, and upgrade Dcur.
If the distance between the branch of the not accessed mistake under Q and its father node is greater than Dcur, then illustrate in this branch there is not the point nearer with Q.
The deterministic process of backtracking is carried out, from the bottom up until there is not the branch nearer with P when tracing back to root node.
4, k-d tree algorithm is improved:
Kd-tree is (such as: K≤30) when dimension is less, the search efficiency of algorithm is very high, but when Kd-tree is used for high dimensional data (such as: K >=100) is carried out to index and searches, be just faced with dimension disaster problem, search efficiency can decline rapidly along with the increase of dimension.Usually, in practical application, the data that we usually process all have the feature of higher-dimension, therefore, index to high dimensional data is met in order to Kd-tree can be allowed, Kd-treewithBBF(BestBinFirst of the present invention) algorithm, this algorithm can realize the fast search of approximate k nearest neighbor, under the prerequisite of precision is necessarily searched in guarantee, make seek rate very fast.Kd-tree algorithm is because too much backtracking number of times causes algorithm search efficiency to decline in higher dimensional space, therefore just can limit the number of times upper limit carrying out when searching recalling, thus avoid search efficiency to decline.After restriction backtracking number of times, if we still conduct interviews in turn according to original retrogressive method, the precision of that lookup result last obviously just depends on the distribution of data to a great extent and recalls number of times.The problem of the method for accessing in turn is to think that the probability that there is arest neighbors in each tree branch to be recalled is the same, so to all, it treats that Backward Tree branch makes no exception.In fact, treat in Backward Tree branch at these, the possibility that some tree branch exists arest neighbors is higher than other tree branches because tree divides distance between fragmented Q point or crossing degree to be different, from Q more close to tree branch to there is the possibility of the arest neighbors of Q higher.Therefore, we need to be treated differently each tree branch to be recalled, and namely adopt certain priority orders to visit these and treat Backward Tree branch, make to find the possibility of the arest neighbors of Q very high in limited backtracking number of times.
5, the Kd-Tree approximate KNN based on BBF is searched:
(1) the current nearest neighbor point P of Q is searched
A. from the root node of KT, Q and intermediate node node (k, m) is compared, select certain tree branch Branch(according to comparative result or be called Bin); And position in the tree at another tree branch (UnexploredBranch) place non-selected and it are saved in Queue in a priority query together with the distance between Q;
B. according to the process of step a, as above comparison and selection is carried out to tree branch Branch, until have access to leafy node, then calculate the distance between the data of preserving in Q and leafy node, and record the data P of minor increment D and correspondence.
(2) based on the backtracking of BBF, known: maximum traceback number of times BTmax:
If the number of times of a. current backtracking is less than BTmax, and Queue is not empty, then proceed as follows:
From Queue, take out the Branch that minor increment is corresponding, then access this Branch until reach leafy node according to this section (1) .a. step; To calculate in Q and leafy node each data pitch from, if there be the value less than D, then this value is assigned to D, these data are then considered to the current approximate KNN point of Q;
B. repeat a step, until backtracking number of times is greater than BTmax or Queue for time empty, search end, the data P now obtained and distance D is exactly the approximate KNN point of Q and the distance between them.
6, above-mentioned algorithm is utilized to carry out the concrete enforcement of vehicle image retrieval
(1) vehicle image feature is tieed up to the N width M that convolutional neural networks extracts
,
...,
utilize above-mentioned higher-dimension k-d to set developing algorithm and carry out higher-dimension k-d structure.
(2) utilize based on the Kd-Tree approximate KNN lookup algorithm of BBF the feature of a query image
the feature of image is tieed up with above-mentioned N width M
,
...,
mate, find out the most similar K width image fast and arrange according to the size of similarity, the image that similarity is larger is forward, and the less image of similarity rearward, finally by manual confirmation similar vehicle with timely, empty arrangement, thus realize the retrieval of relevant fee evasion vehicle.
Claims (6)
1. the method for a quick-searching highway fee evasion vehicle high similarity graph picture, it is characterized in that: the convolutional neural networks model utilizing computing machine, Feature extraction and recognition is carried out to the sample image of the violation vehicle collected, and corresponding K-d tree structure is carried out to the feature identified, then convolutional neural networks is utilized to carry out feature extraction to the unknown images collected, and utilize quick high similarity nearest neighbor search algorithm to mate sample characteristics, to realize the quick-searching of high similarity graph picture.
2. the method for quick-searching highway fee evasion vehicle high similarity graph picture according to claim 1, is characterized in that, specifically comprise the following steps:
1. collect vehicle sample image in violation of rules and regulations, set up vehicle management database;
2. sample image is normalized, unified vehicle image form;
3. utilize convolutional neural networks model, Feature extraction and recognition is carried out to vehicle, obtains sample characteristics;
4. corresponding K-d is carried out to sample characteristics and set structure;
5. feature is extracted to obtain to the unknown images collected, set on basis at the K-d built, quick for these characteristic use high similarity nearest neighbor search algorithm is gone to mate with sample characteristics, to realize scheming to search figure effect;
6. according to the matching degree size that 5. step obtains, export similarity graph picture according to time sequencing, finally carry out manual confirmation.
3. the method for quick-searching highway fee evasion vehicle high similarity graph picture according to claim 2, is characterized in that, the described step 3. middle feature extracted comprises:
(1) characters on license plate feature;
(2) vehicle feature;
(3) car mark, body color feature;
(4) vehicle facial characteristics;
(5) external form of driver and conductor and facial characteristics.
4. the method for quick-searching highway fee evasion vehicle high similarity graph picture according to claim 2, is characterized in that: described step 5. in quick high similarity nearest neighbor search algorithm be carry out by back tracking operation the method improved when utilizing K-d to set to carry out neighbor searching.
5. the method for quick-searching highway fee evasion vehicle high similarity graph picture according to claim 2, it is characterized in that: described step 3. in the convolutional neural networks model that uses be obtain after carrying out convolutional neural networks training as sample set using the vector of the formation of input vector and desirable output vector, and before training, use the every weights of little random number to convolutional neural networks to carry out initialization.
6. the method for quick-searching highway fee evasion vehicle high similarity graph picture according to claim 5, is characterized in that: the concrete grammar of training comprises the following steps:
The first step, gets a sample input convolutional neural networks from sample set;
Second step, calculates corresponding actual output;
3rd step, calculates the difference that actual output exports with corresponding ideal;
4th step, by the method backpropagation adjustment weight matrix of minimization error.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510714633.3A CN105354273A (en) | 2015-10-29 | 2015-10-29 | Method for fast retrieving high-similarity image of highway fee evasion vehicle |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510714633.3A CN105354273A (en) | 2015-10-29 | 2015-10-29 | Method for fast retrieving high-similarity image of highway fee evasion vehicle |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105354273A true CN105354273A (en) | 2016-02-24 |
Family
ID=55330246
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510714633.3A Pending CN105354273A (en) | 2015-10-29 | 2015-10-29 | Method for fast retrieving high-similarity image of highway fee evasion vehicle |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105354273A (en) |
Cited By (38)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105808732A (en) * | 2016-03-10 | 2016-07-27 | 北京大学 | Integration target attribute identification and precise retrieval method based on depth measurement learning |
| CN106156750A (en) * | 2016-07-26 | 2016-11-23 | 浙江捷尚视觉科技股份有限公司 | A kind of based on convolutional neural networks to scheme to search car method |
| CN106203493A (en) * | 2016-07-04 | 2016-12-07 | 何广森 | A kind of food identification device and recognition methods |
| CN106295124A (en) * | 2016-07-27 | 2017-01-04 | 广州麦仑信息科技有限公司 | Utilize the method that multiple image detecting technique comprehensively analyzes gene polyadenylation signal figure likelihood probability amount |
| CN106446949A (en) * | 2016-09-26 | 2017-02-22 | 成都通甲优博科技有限责任公司 | Vehicle model identification method and apparatus |
| CN106778777A (en) * | 2016-11-30 | 2017-05-31 | 成都通甲优博科技有限责任公司 | A kind of vehicle match method and system |
| CN106875693A (en) * | 2017-03-29 | 2017-06-20 | 广西信路威科技发展有限公司 | A kind of method and system of vehicle feature recognition |
| CN106897677A (en) * | 2017-02-06 | 2017-06-27 | 桂林电子科技大学 | A kind of vehicle characteristics classification and retrieval system and method |
| CN107341506A (en) * | 2017-06-12 | 2017-11-10 | 华南理工大学 | A kind of Image emotional semantic classification method based on the expression of many-sided deep learning |
| CN107423379A (en) * | 2017-07-13 | 2017-12-01 | 西安电子科技大学 | Image search method based on CNN feature words trees |
| CN107545239A (en) * | 2017-07-06 | 2018-01-05 | 南京理工大学 | A kind of deck detection method matched based on Car license recognition with vehicle characteristics |
| CN107644251A (en) * | 2016-07-22 | 2018-01-30 | 北京市商汤科技开发有限公司 | Neuronal activation methods, devices and systems and object classification method and system |
| CN107679078A (en) * | 2017-08-29 | 2018-02-09 | 银江股份有限公司 | A kind of bayonet socket image vehicle method for quickly retrieving and system based on deep learning |
| CN107688819A (en) * | 2017-02-16 | 2018-02-13 | 平安科技（深圳）有限公司 | The recognition methods of vehicle and device |
| CN107833183A (en) * | 2017-11-29 | 2018-03-23 | 安徽工业大学 | A kind of satellite image based on multitask deep neural network while super-resolution and the method for coloring |
| CN108038423A (en) * | 2017-11-22 | 2018-05-15 | 广东数相智能科技有限公司 | The recognition methods of automotive type based on image recognition and device |
| CN108197326A (en) * | 2018-02-06 | 2018-06-22 | 腾讯科技（深圳）有限公司 | A kind of vehicle retrieval method and device, electronic equipment, storage medium |
| CN108305461A (en) * | 2017-12-20 | 2018-07-20 | 大唐软件技术股份有限公司 | A kind of determination method and apparatus for evading expense suspected vehicles |
| CN108595576A (en) * | 2018-04-17 | 2018-09-28 | 高新兴科技集团股份有限公司 | It is a kind of based on the vehicle of database to scheme to search drawing method |
| CN108830379A (en) * | 2018-05-23 | 2018-11-16 | 电子科技大学 | A kind of neuromorphic processor shared based on parameter quantization |
| CN109033107A (en) * | 2017-06-09 | 2018-12-18 | 腾讯科技（深圳）有限公司 | Image search method and device, computer equipment and storage medium |
| CN109360280A (en) * | 2018-08-22 | 2019-02-19 | 东软集团股份有限公司 | Establish method, apparatus, storage medium and the electronic equipment for escaping payment omitted identification model |
| CN109426831A (en) * | 2017-08-30 | 2019-03-05 | 腾讯科技（深圳）有限公司 | The method, apparatus and computer equipment of picture Similarity matching and model training |
| CN109948734A (en) * | 2019-04-02 | 2019-06-28 | 北京旷视科技有限公司 | Image clustering method, device and electronic equipment |
| CN110019876A (en) * | 2017-12-25 | 2019-07-16 | 深圳云天励飞技术有限公司 | Data query method, electronic equipment and storage medium |
| CN110044374A (en) * | 2018-01-17 | 2019-07-23 | 南京火眼猴信息科技有限公司 | A kind of method and odometer of the monocular vision measurement mileage based on characteristics of image |
| CN110363180A (en) * | 2019-07-24 | 2019-10-22 | 厦门云上未来人工智能研究院有限公司 | A kind of method and apparatus and equipment that statistics stranger's face repeats |
| WO2019241224A1 (en) * | 2018-06-11 | 2019-12-19 | Raytheon Company | Architectures for vehicle tolling |
| CN111079602A (en) * | 2019-12-06 | 2020-04-28 | 长沙千视通智能科技有限公司 | Vehicle fine granularity identification method and device based on multi-scale regional feature constraint |
| CN111191713A (en) * | 2019-12-27 | 2020-05-22 | 大象慧云信息技术有限公司 | User portrait method and device based on invoice data |
| CN111401466A (en) * | 2020-03-26 | 2020-07-10 | 广州紫为云科技有限公司 | Traffic sign detection and identification marking method and device and computer equipment |
| CN111694981A (en) * | 2020-06-21 | 2020-09-22 | 深圳天海宸光科技有限公司 | Gas station fee evasion alarm system and method based on machine vision |
| CN112000940A (en) * | 2020-09-11 | 2020-11-27 | 支付宝(杭州)信息技术有限公司 | User identification method, device and equipment under privacy protection |
| CN113570738A (en) * | 2021-07-06 | 2021-10-29 | 哈尔滨工业大学 | ETC passage loss-of-credit behavior electronic evidence obtaining and classified management method and system |
| CN113705469A (en) * | 2021-08-30 | 2021-11-26 | 平安科技（深圳）有限公司 | Face recognition method and device, electronic equipment and computer readable storage medium |
| CN114387572A (en) * | 2022-01-11 | 2022-04-22 | 杭州电子科技大学 | AI technology-based red light running illegal vehicle matching method in traffic law enforcement image |
| CN115601967A (en) * | 2022-09-30 | 2023-01-13 | 广州天长信息技术有限公司（Cn） | Data analysis method for UJ type driving fee evasion behavior of expressway |
| CN117312598A (en) * | 2023-11-27 | 2023-12-29 | 广东利通科技投资有限公司 | Evidence obtaining method, device, computer equipment and storage medium for fee evasion auditing |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104463241A (en) * | 2014-10-31 | 2015-03-25 | 北京理工大学 | Vehicle type recognition method in intelligent transportation monitoring system |
-
2015
- 2015-10-29 CN CN201510714633.3A patent/CN105354273A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104463241A (en) * | 2014-10-31 | 2015-03-25 | 北京理工大学 | Vehicle type recognition method in intelligent transportation monitoring system |
Non-Patent Citations (4)
| Title |
|---|
| D. MENOTTI等: "Vehicle License Plate Recognition With", 《2014 27TH SIBGRAPI CONFERENCE ON GRAPHICS,PATTERNS AND IMAGES》 * |
| YUE HUANG等: "Vehicle Logo Recognition System Based", 《IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS》 * |
| 刘超: "多姿态车型识别算法涉及应用研究", 《万方数据企业知识服务平台》 * |
| 彭博等: "基于深度学习的车标识别方法研究", 《计算机科学》 * |
Cited By (63)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105808732B (en) * | 2016-03-10 | 2019-05-17 | 北京大学 | A kind of integrated Target attribute recognition and precise search method based on depth measure study |
| CN105808732A (en) * | 2016-03-10 | 2016-07-27 | 北京大学 | Integration target attribute identification and precise retrieval method based on depth measurement learning |
| CN106203493B (en) * | 2016-07-04 | 2019-07-12 | 何广森 | A kind of food identification device and recognition methods |
| CN106203493A (en) * | 2016-07-04 | 2016-12-07 | 何广森 | A kind of food identification device and recognition methods |
| CN107644251B (en) * | 2016-07-22 | 2020-09-18 | 北京市商汤科技开发有限公司 | Object classification method, device and system |
| CN107644251A (en) * | 2016-07-22 | 2018-01-30 | 北京市商汤科技开发有限公司 | Neuronal activation methods, devices and systems and object classification method and system |
| CN106156750A (en) * | 2016-07-26 | 2016-11-23 | 浙江捷尚视觉科技股份有限公司 | A kind of based on convolutional neural networks to scheme to search car method |
| CN106295124A (en) * | 2016-07-27 | 2017-01-04 | 广州麦仑信息科技有限公司 | Utilize the method that multiple image detecting technique comprehensively analyzes gene polyadenylation signal figure likelihood probability amount |
| CN106295124B (en) * | 2016-07-27 | 2018-11-27 | 广州麦仑信息科技有限公司 | The method of a variety of image detecting technique comprehensive analysis gene subgraph likelihood probability amounts |
| CN106446949A (en) * | 2016-09-26 | 2017-02-22 | 成都通甲优博科技有限责任公司 | Vehicle model identification method and apparatus |
| CN106446949B (en) * | 2016-09-26 | 2019-05-21 | 成都通甲优博科技有限责任公司 | A kind of vehicle model recognition methods and device |
| CN106778777A (en) * | 2016-11-30 | 2017-05-31 | 成都通甲优博科技有限责任公司 | A kind of vehicle match method and system |
| CN106897677A (en) * | 2017-02-06 | 2017-06-27 | 桂林电子科技大学 | A kind of vehicle characteristics classification and retrieval system and method |
| CN106897677B (en) * | 2017-02-06 | 2020-04-10 | 桂林电子科技大学 | Vehicle feature classification retrieval system and method |
| US10740927B2 (en) | 2017-02-16 | 2020-08-11 | Ping An Technology (Shenzhen) Co., Ltd. | Method and device for vehicle identification |
| WO2018149302A1 (en) * | 2017-02-16 | 2018-08-23 | 平安科技（深圳）有限公司 | Vehicle identification method and apparatus |
| CN107688819A (en) * | 2017-02-16 | 2018-02-13 | 平安科技（深圳）有限公司 | The recognition methods of vehicle and device |
| US20190139262A1 (en) * | 2017-02-16 | 2019-05-09 | Ping An Technology (Shenzhen) Co., Ltd. | Method and device for vehicle identification |
| CN106875693A (en) * | 2017-03-29 | 2017-06-20 | 广西信路威科技发展有限公司 | A kind of method and system of vehicle feature recognition |
| CN109033107A (en) * | 2017-06-09 | 2018-12-18 | 腾讯科技（深圳）有限公司 | Image search method and device, computer equipment and storage medium |
| CN109033107B (en) * | 2017-06-09 | 2021-09-17 | 腾讯科技（深圳）有限公司 | Image retrieval method and apparatus, computer device, and storage medium |
| CN107341506A (en) * | 2017-06-12 | 2017-11-10 | 华南理工大学 | A kind of Image emotional semantic classification method based on the expression of many-sided deep learning |
| CN107545239A (en) * | 2017-07-06 | 2018-01-05 | 南京理工大学 | A kind of deck detection method matched based on Car license recognition with vehicle characteristics |
| CN107545239B (en) * | 2017-07-06 | 2021-01-12 | 南京理工大学 | Fake plate detection method based on license plate recognition and vehicle characteristic matching |
| CN107423379A (en) * | 2017-07-13 | 2017-12-01 | 西安电子科技大学 | Image search method based on CNN feature words trees |
| CN107423379B (en) * | 2017-07-13 | 2019-10-11 | 西安电子科技大学 | Image search method based on CNN feature words tree |
| CN107679078A (en) * | 2017-08-29 | 2018-02-09 | 银江股份有限公司 | A kind of bayonet socket image vehicle method for quickly retrieving and system based on deep learning |
| CN107679078B (en) * | 2017-08-29 | 2020-01-10 | 银江股份有限公司 | Bayonet image vehicle rapid retrieval method and system based on deep learning |
| CN109426831A (en) * | 2017-08-30 | 2019-03-05 | 腾讯科技（深圳）有限公司 | The method, apparatus and computer equipment of picture Similarity matching and model training |
| CN109426831B (en) * | 2017-08-30 | 2022-12-13 | 腾讯科技（深圳）有限公司 | Image similarity matching and model training method and device and computer equipment |
| CN108038423B (en) * | 2017-11-22 | 2022-03-04 | 广东数相智能科技有限公司 | Automobile type identification method and device based on image identification |
| CN108038423A (en) * | 2017-11-22 | 2018-05-15 | 广东数相智能科技有限公司 | The recognition methods of automotive type based on image recognition and device |
| CN107833183B (en) * | 2017-11-29 | 2021-05-25 | 安徽工业大学 | Method for simultaneously super-resolving and coloring satellite image based on multitask deep neural network |
| CN107833183A (en) * | 2017-11-29 | 2018-03-23 | 安徽工业大学 | A kind of satellite image based on multitask deep neural network while super-resolution and the method for coloring |
| CN108305461A (en) * | 2017-12-20 | 2018-07-20 | 大唐软件技术股份有限公司 | A kind of determination method and apparatus for evading expense suspected vehicles |
| CN110019876B (en) * | 2017-12-25 | 2023-07-28 | 深圳云天励飞技术有限公司 | Data query method, electronic device and storage medium |
| CN110019876A (en) * | 2017-12-25 | 2019-07-16 | 深圳云天励飞技术有限公司 | Data query method, electronic equipment and storage medium |
| CN110044374B (en) * | 2018-01-17 | 2022-12-09 | 宽衍(北京)科技发展有限公司 | Image feature-based monocular vision mileage measurement method and odometer |
| CN110044374A (en) * | 2018-01-17 | 2019-07-23 | 南京火眼猴信息科技有限公司 | A kind of method and odometer of the monocular vision measurement mileage based on characteristics of image |
| CN108197326A (en) * | 2018-02-06 | 2018-06-22 | 腾讯科技（深圳）有限公司 | A kind of vehicle retrieval method and device, electronic equipment, storage medium |
| CN108595576A (en) * | 2018-04-17 | 2018-09-28 | 高新兴科技集团股份有限公司 | It is a kind of based on the vehicle of database to scheme to search drawing method |
| CN108830379B (en) * | 2018-05-23 | 2021-12-17 | 电子科技大学 | Neural morphology processor based on parameter quantification sharing |
| CN108830379A (en) * | 2018-05-23 | 2018-11-16 | 电子科技大学 | A kind of neuromorphic processor shared based on parameter quantization |
| US11062529B2 (en) | 2018-06-11 | 2021-07-13 | Raytheon Company | Architectures for vehicle tolling |
| WO2019241224A1 (en) * | 2018-06-11 | 2019-12-19 | Raytheon Company | Architectures for vehicle tolling |
| CN109360280A (en) * | 2018-08-22 | 2019-02-19 | 东软集团股份有限公司 | Establish method, apparatus, storage medium and the electronic equipment for escaping payment omitted identification model |
| CN109948734A (en) * | 2019-04-02 | 2019-06-28 | 北京旷视科技有限公司 | Image clustering method, device and electronic equipment |
| CN110363180A (en) * | 2019-07-24 | 2019-10-22 | 厦门云上未来人工智能研究院有限公司 | A kind of method and apparatus and equipment that statistics stranger's face repeats |
| CN111079602B (en) * | 2019-12-06 | 2024-02-09 | 长沙千视通智能科技有限公司 | Vehicle fine granularity identification method and device based on multi-scale regional feature constraint |
| CN111079602A (en) * | 2019-12-06 | 2020-04-28 | 长沙千视通智能科技有限公司 | Vehicle fine granularity identification method and device based on multi-scale regional feature constraint |
| CN111191713A (en) * | 2019-12-27 | 2020-05-22 | 大象慧云信息技术有限公司 | User portrait method and device based on invoice data |
| CN111401466A (en) * | 2020-03-26 | 2020-07-10 | 广州紫为云科技有限公司 | Traffic sign detection and identification marking method and device and computer equipment |
| CN111694981A (en) * | 2020-06-21 | 2020-09-22 | 深圳天海宸光科技有限公司 | Gas station fee evasion alarm system and method based on machine vision |
| US11277258B1 (en) * | 2020-09-11 | 2022-03-15 | Alipay (Hangzhou) Information Technology Co., Ltd. | Privacy protection-based user recognition methods, apparatuses, and devices |
| CN112000940B (en) * | 2020-09-11 | 2022-07-12 | 支付宝(杭州)信息技术有限公司 | User identification method, device and equipment under privacy protection |
| CN112000940A (en) * | 2020-09-11 | 2020-11-27 | 支付宝(杭州)信息技术有限公司 | User identification method, device and equipment under privacy protection |
| CN113570738A (en) * | 2021-07-06 | 2021-10-29 | 哈尔滨工业大学 | ETC passage loss-of-credit behavior electronic evidence obtaining and classified management method and system |
| CN113705469A (en) * | 2021-08-30 | 2021-11-26 | 平安科技（深圳）有限公司 | Face recognition method and device, electronic equipment and computer readable storage medium |
| CN114387572A (en) * | 2022-01-11 | 2022-04-22 | 杭州电子科技大学 | AI technology-based red light running illegal vehicle matching method in traffic law enforcement image |
| CN114387572B (en) * | 2022-01-11 | 2024-04-09 | 杭州电子科技大学 | Traffic law enforcement image based on AI technology for matching vehicles running red light and illegal |
| CN115601967A (en) * | 2022-09-30 | 2023-01-13 | 广州天长信息技术有限公司（Cn） | Data analysis method for UJ type driving fee evasion behavior of expressway |
| CN117312598A (en) * | 2023-11-27 | 2023-12-29 | 广东利通科技投资有限公司 | Evidence obtaining method, device, computer equipment and storage medium for fee evasion auditing |
| CN117312598B (en) * | 2023-11-27 | 2024-04-09 | 广东利通科技投资有限公司 | Evidence obtaining method, device, computer equipment and storage medium for fee evasion auditing |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105354273A (en) | Method for fast retrieving high-similarity image of highway fee evasion vehicle | |
| US10984532B2 (en) | Joint deep learning for land cover and land use classification | |
| Othman et al. | Domain adaptation network for cross-scene classification | |
| Zhuang et al. | Innovative method for traffic data imputation based on convolutional neural network | |
| Shi et al. | Branch feature fusion convolution network for remote sensing scene classification | |
| CN104244113B (en) | A kind of video abstraction generating method based on depth learning technology | |
| CN110059581A (en) | People counting method based on depth information of scene | |
| Zhou et al. | Image semantic segmentation based on FCN-CRF model | |
| CN113240688A (en) | Integrated flood disaster accurate monitoring and early warning method | |
| CN114937151A (en) | Lightweight target detection method based on multi-receptive-field and attention feature pyramid | |
| CN112232151B (en) | Iterative polymerization neural network high-resolution remote sensing scene classification method embedded with attention mechanism | |
| CN104504395A (en) | Method and system for achieving classification of pedestrians and vehicles based on neural network | |
| CN111639587B (en) | Hyperspectral image classification method based on multi-scale spectrum space convolution neural network | |
| CN113297972B (en) | Transformer substation equipment defect intelligent analysis method based on data fusion deep learning | |
| He et al. | A robust method for wheatear detection using UAV in natural scenes | |
| Lu et al. | Multi-object detection method based on YOLO and ResNet hybrid networks | |
| CN113592894A (en) | Image segmentation method based on bounding box and co-occurrence feature prediction | |
| CN106355210A (en) | Method for expressing infrared image features of insulators on basis of depth neuron response modes | |
| Lee et al. | Fast object localization using a CNN feature map based multi-scale search | |
| CN112668662B (en) | Outdoor mountain forest environment target detection method based on improved YOLOv3 network | |
| CN112418262A (en) | Vehicle re-identification method, client and system | |
| Li et al. | A new algorithm of vehicle license plate location based on convolutional neural network | |
| CN114913475B (en) | Urban gridding management method and system based on GIS and machine vision | |
| CN117011566A (en) | Target detection method, detection model training method, device and electronic equipment | |
| KR102492843B1 (en) | Apparatus and method for analyzing spatio temporal data for geo-location |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160224 |