CN114494947B - Traffic video vehicle classification method based on quantum optimization algorithm - Google Patents
Traffic video vehicle classification method based on quantum optimization algorithm Download PDFInfo
- Publication number
- CN114494947B CN114494947B CN202111662993.5A CN202111662993A CN114494947B CN 114494947 B CN114494947 B CN 114494947B CN 202111662993 A CN202111662993 A CN 202111662993A CN 114494947 B CN114494947 B CN 114494947B
- Authority
- CN
- China
- Prior art keywords
- vehicle
- traffic video
- quantum
- algorithm
- video vehicle
- 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
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 75
- 238000005457 optimization Methods 0.000 title claims abstract description 50
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000001514 detection method Methods 0.000 claims abstract description 30
- 238000013145 classification model Methods 0.000 claims abstract description 17
- 238000012360 testing method Methods 0.000 claims abstract description 15
- 238000013135 deep learning Methods 0.000 claims abstract description 9
- 238000012706 support-vector machine Methods 0.000 claims abstract description 9
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 239000002245 particle Substances 0.000 claims description 38
- 230000006870 function Effects 0.000 claims description 16
- 238000000137 annealing Methods 0.000 claims description 15
- 239000013598 vector Substances 0.000 claims description 10
- 238000011176 pooling Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 7
- 238000004458 analytical method Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 4
- 230000001131 transforming effect Effects 0.000 claims description 4
- FGUUSXIOTUKUDN-IBGZPJMESA-N C1(=CC=CC=C1)N1C2=C(NC([C@H](C1)NC=1OC(=NN=1)C1=CC=CC=C1)=O)C=CC=C2 Chemical compound C1(=CC=CC=C1)N1C2=C(NC([C@H](C1)NC=1OC(=NN=1)C1=CC=CC=C1)=O)C=CC=C2 FGUUSXIOTUKUDN-IBGZPJMESA-N 0.000 claims description 2
- 230000001360 synchronised effect Effects 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 description 6
- 239000002096 quantum dot Substances 0.000 description 6
- 230000003321 amplification Effects 0.000 description 4
- 238000003199 nucleic acid amplification method Methods 0.000 description 4
- 230000001133 acceleration Effects 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 230000005283 ground state Effects 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000010845 search algorithm Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000002028 premature Effects 0.000 description 2
- 238000002360 preparation method Methods 0.000 description 2
- 230000005610 quantum mechanics Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N10/00—Quantum computing, i.e. information processing based on quantum-mechanical phenomena
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Condensed Matter Physics & Semiconductors (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a traffic video vehicle classification method based on a quantum optimization algorithm, which comprises the following steps: step 1, performing unstructured traffic video vehicle coarse-granularity target detection by using a target detection algorithm; step 2, extracting the brand features of the target vehicle by using a deep learning network algorithm, preprocessing the extracted brand features of the vehicle, and dividing the preprocessed data into a training data set and a test data set according to 70%/30%; step 3, constructing and training a traffic video vehicle feature classification model by utilizing the training data set in the step 2 and combining a quantum optimization algorithm with a traditional support vector machine; and 4, performing classification test on the test data set in the step 2 by using the trained traffic video vehicle classification model to obtain a traffic video vehicle prediction classification result, comparing and analyzing the traffic video vehicle prediction classification result with a real vehicle brand result, and outputting the recognition accuracy.
Description
Technical Field
The invention relates to a traffic video vehicle classification method, in particular to a traffic video vehicle classification method based on a quantum optimization algorithm.
Background
In recent years, quantum computing technology has developed rapidly, and positive progress has been made in the aspects of quantum computing hardware design and preparation, quantum algorithm software development and optimization and the like. Quantum computing application research involves many aspects, of which optimization is an important research direction, and is widely applied to many fields such as traffic optimization, image analysis, financial transactions, logistics lines, engineering design and management, high-end equipment manufacturing, and the like. In solving the optimization problem, the acceleration realized by quantum computing is one of the research directions of great attention at present, namely quantum optimization, and the corresponding quantum optimization algorithm comprises a Grover quantum search algorithm, a quantum approximation optimization algorithm, a HHT solution linear equation set quantum algorithm, a quantum annealing algorithm, hamiltonian simulation and the like.
Along with the continuous and deep construction of smart cities, multi-dimensional and all-dimensional sensor systems are gradually built in the ground traffic fields of roads, railways, urban rails and the like in China, and a strong data support is built for the smart traffic. In the face of all-weather traffic data collected by sensors, the traditional mode cannot complete large data analysis. The quantum information technology is based on quantum mechanics theory, has strong parallel computing capability by virtue of quantum superposition state and entanglement characteristics, can realize quantum acceleration in specific fields, and particularly can realize high-efficiency processing and analysis of data by adopting a quantum optimization algorithm in combination with a traditional method in the face of a modern traffic network system with increasingly large data scale and complex structure, thereby having important significance for construction and management of smart cities.
In order to solve the above problems, an ideal technical solution is always sought.
Disclosure of Invention
The invention aims at overcoming the defects of the prior art, and provides a traffic video vehicle classification method based on a quantum optimization algorithm.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows: a traffic video vehicle classification method based on a quantum optimization algorithm comprises the following steps:
Step 1, performing unstructured traffic video vehicle coarse-granularity target detection by using a target detection algorithm, and extracting a target vehicle picture in a traffic video;
Step 2, extracting vehicle brand features from the target vehicle pictures by using a deep learning network algorithm, preprocessing the extracted vehicle brand features, and dividing preprocessed data into a training data set and a test data set according to 70%/30%;
step 3, constructing and training a traffic video vehicle feature classification model by utilizing the training data set in the step 2 and combining a quantum optimization algorithm with a traditional support vector machine;
And 4, carrying out classification test on the test data set in the step 2 by using the trained traffic video vehicle classification model to obtain a traffic video vehicle prediction classification result, and comparing and analyzing the traffic video vehicle prediction classification result with a real vehicle brand result to verify the correctness of the traffic video vehicle feature classification model.
Based on the above, the target detection algorithm in step1 is YOLOv detection algorithm.
Based on the above, before the unstructured traffic video vehicle coarse-granularity target detection is performed by using YOLOv detection algorithm, the super-parameter learning rate lr0, cosine annealing super-parameter lrf, learning rate momentum and weight attenuation coefficient weight_decay of YOLOv detection algorithm need to be synchronously and comprehensively optimized by using quantum particle swarm optimization QPSO algorithm to obtain the optimal super-parameter learning rate lr0, optimal cosine annealing super-parameter lrf, optimal learning rate momentum and optimal weight attenuation coefficient weight_decay.
Based on the above, in step 2, the deep learning network algorithm selected when the vehicle brand feature of the target vehicle is extracted by using the deep learning network algorithm is a base network layer of FASTER RCNN network structure, which comprises 13 convolution layers and 4 pooling layers, wherein a ReLU function process is arranged behind each convolution layer, the convolution kernel of the convolution layer is 3*3, the step length is 1, the pooling size of the pooling layer is 2 x 2, and the step length is 2;
The 4096-dimensional features are obtained after processing by the base network layer, and the first 512-dimensional features are selected as vehicle brand features.
Compared with the prior art, the invention has outstanding substantive characteristics and remarkable progress, and specifically, the invention carries out mark recognition on unstructured traffic video vehicles through a target detection algorithm, extracts vehicle characteristic information to construct a sample data set, carries out training and classification test of a traffic video vehicle brand characteristic classification model on the sample data set by combining a support vector machine based on a quantum optimization algorithm, and can realize correct classification recognition of vehicle brands in traffic video information.
According to the invention, the quantum particle swarm optimization QPSO algorithm is utilized to synchronously and comprehensively optimize the super-parameter learning rate lr0, the cosine annealing super-parameter lrf, the learning rate momentum and the weight attenuation coefficient weight_decay of the target detection algorithm, so that the defects of premature convergence, poor global optimization calculation precision and the like caused by gradual reduction of particle swarm diversity due to multiple optimization iterations of the particle swarm optimization algorithm can be effectively overcome, and the convergence speed of the training process of the target detection algorithm is improved.
Drawings
Fig. 1 is a flow chart of a traffic video vehicle classification method based on a quantum optimization algorithm of the present invention.
Fig. 2 is a flow chart of feature value extraction by a deep learning object detection algorithm.
Fig. 3 is a flow chart of a quantum particle swarm optimization algorithm.
Fig. 4 is a quantum circuit diagram of the HHL algorithm solving linear equations.
Fig. 5 is an enlarged schematic of the Grover quantum search algorithm amplitude.
Detailed Description
The technical scheme of the invention is further described in detail through the following specific embodiments.
Example 1
As shown in fig. 1, the present embodiment provides a traffic video vehicle classification method based on a quantum optimization algorithm, which includes the following steps:
Step 1, performing unstructured traffic video vehicle coarse-granularity target detection by using a target detection algorithm, and extracting a target vehicle picture in a traffic video;
Because the real-time requirement of the traffic video vehicle analysis is higher, the target detection algorithm adopts YOLOv algorithm with higher real-time; the YOLOv algorithm detects faster and the network is lighter in weight than the previous YOLO version with comparable accuracy.
Step 2, extracting vehicle brand features from the target vehicle pictures by using a deep learning network algorithm, preprocessing the extracted vehicle brand features, and dividing preprocessed data into a training data set and a test data set according to 70%/30%;
In specific implementation, extracting vehicle brand features from a vehicle target picture by utilizing FASTER RCNN networks;
Specifically, as shown in fig. 2, when the brand features of the vehicle are extracted, a base network layer of FASTER RCNN network structures is selected and comprises 13 convolution layers and 4 pooling layers, wherein a ReLU function is arranged behind each convolution layer, the convolution kernel of the convolution layer is 3*3, the step length is 1, the pooling size of the pooling layers is 2 x 2, and the step length is 2.
The 4096-dimensional features are obtained after the processing of the basic network layer, and the clustering analysis of the features finds that the features of the first 512-dimensional have the same data distribution state as the features of the 4096-dimensional features, so that the features of the first 512-dimensional are used for replacing the features of the 4096-dimensional features as the brand features of the vehicle and are stored in a data set, and the calculated amount and the memory can be reduced.
The 512-dimensional feature is selected here in order to select a portion of the features in 4096 dimensions as the identification feature. Through the clustering analysis of the features, the features of the first 512 dimensions have the same data distribution state as the features of the 4096 dimensions, so that the features of the first 512 dimensions are used for replacing the features of the 4096 dimensions as the unique identification features of the picture.
The vehicle brand features in the data set are preprocessed, the specific preprocessing comprises data normalization, standardization and data dimension reduction processing, the data preprocessing process is favorable for realizing acceleration convergence of a vehicle feature classification model solving process of the traffic video, the calculation accuracy is improved, 70% of data in the processed new data set is used as a training data set for training a vehicle feature classification recognition model of the traffic video, and the rest is used as a test data set for verifying the vehicle feature classification recognition model of the traffic video.
Step 3, constructing and training a traffic video vehicle feature classification model by utilizing the training data set in the step 2 and combining a quantum optimization algorithm with a traditional support vector machine;
specifically, the construction steps of the traffic video vehicle feature classification model are as follows:
The sample form of the training dataset is set as follows: Wherein the method comprises the steps of For 512-dimensional feature vectors of training samples, x j is 512-dimensional attribute feature data, y j is a real vehicle brand corresponding to the training samples, N is feature vector dimensions of the training samples in the training dataset and n=512, m is the total number of samples in the training dataset;
the specific objective of the method is to seek the maximum interval hyperplane to realize the optimal brand feature classification of the vehicle, and obtain an objective function after conversion as follows Is a weight vector; regarding noise fault tolerance data as approximate linear separable, and searching soft boundary hyperplane; introducing a relaxation variable e j for indicating the allowable data point deviation, the objective function isWherein gamma is a penalty coefficient, b is a bias constant, and the constraint is constrained by inequalityConversion to equality constraints
Introducing Lagrangian function multipliersConstruction
Alpha j is greater than or equal to 0 and corresponds toIs a product of the lagrangian multipliers of (c),Is [ alpha 1,α2,…,αj,…,αM ],Is a relaxation vector;
based on KKT condition pair Obtaining partial derivativeI.e.Thereby converting the quadratic programming optimization problem into a problem of solving a linear equation set;
Wherein the method comprises the steps of AndInput vectors for the jth and kth models, j, k=1, 2, …, M, respectively; e is an identity matrix;
solving a linear equation set to obtain an optimal hyperplane parameter And b, thereby constructing a support vector machine model y (x) = Σ j＝1αjk(x,xj) +b for traffic video vehicle feature analysis,I.e., traffic video vehicle feature classification model.
And 4, carrying out classification test on the test data set in the step 2 by using the trained traffic video vehicle classification model to obtain a traffic video vehicle prediction classification result, and comparing and analyzing the traffic video vehicle prediction classification result with a real vehicle brand result to verify the correctness of the traffic video vehicle feature classification model.
Example 2
This embodiment differs from embodiment 1 in that: before carrying out coarse-granularity target detection on unstructured traffic video vehicles by using YOLOv detection algorithm, the super-parameter learning rate lr0, cosine annealing super-parameter lrf, learning rate momentum and weight attenuation coefficient weight_decay of the YOLOv detection algorithm are required to be synchronously and comprehensively optimized by using quantum particle swarm optimization QPSO algorithm to obtain the optimal super-parameter learning rate lr0, optimal cosine annealing super-parameter lrf, optimal learning rate momentum and optimal weight attenuation coefficient weight_decay.
It can be appreciated that the result of the loss function affects the accuracy of target recognition; whereas Yolov's loss function is multimodal, with multiple locally optimal solutions in addition to globally optimal solutions, the gradient descent algorithm may sink to local minima during training, oscillate around the minima, and may not even converge. In order to solve the problem, the super-parameter learning rate lr0, the cosine annealing super-parameter lrf, the learning rate momentum and the weight attenuation coefficient weight_decay selected by the quantum particle swarm optimization algorithm are adopted to enable the loss function to be converged to the global minimum value in the shortest time, so that the defects of premature convergence, poor global optimization calculation precision and the like caused by gradual reduction of particle swarm diversity due to multiple optimization iterations of the particle swarm optimization algorithm can be effectively overcome, and the convergence speed of the algorithm training process is improved.
Based on the characteristics of microscopic particle superposition in quantum mechanics theory and the like, the quantum particle swarm algorithm utilizes the quantum rotation gate to transform the quantum bit probability amplitude, so that the transformation of the position of the particles in the solution space is realized, and the simultaneous change of two positions in the solution space can be realized when the position of the quantum particles is transformed according to the quantum rotation gate, so that the search range can be expanded by adopting quantum bit coding to perform particle optimization, most of the particle positions can be traversed as much as possible, and finally the accuracy and the efficiency of optimizing calculation are improved.
Specifically, as shown in fig. 3, the specific steps of performing synchronous comprehensive optimization on the super-parameter learning rate lr0, the cosine annealing super-parameter lrf, the learning rate momentum and the weight attenuation coefficient weight_decay of the YOLOv detection algorithm by using the quantum particle swarm optimization QPSO algorithm are as follows:
a. initializing four super parameters to be optimized, namely a learning rate lr0, a cosine annealing super parameter lrf, a learning rate momentum and a weight attenuation coefficient weight_decay of a YOLOv detection algorithm, and setting a reasonable optimizing interval;
b. the quantum particle swarm optimization initial parameter setting comprises particle swarm size and dimension;
c. initializing probability amplitude codes by using individual position information of a quantum particle swarm;
d. transforming the initialized randomly generated particle position probability amplitude into a quantum solution space;
e. Calculating fitness values of all particles based on the objective function to obtain an initial individual optimal value and a global optimal value;
f. Transforming the quantum state probability amplitude by utilizing the quantum rotating gate to realize particle position updating and generate a new particle swarm;
g. Calculating the fitness value of each particle in the new population based on the objective function to obtain a new individual optimal value and a global optimal value;
h. comparing and judging the information with the original optimal value, if the information is superior to the original optimal value, updating and replacing the information, otherwise, keeping the information of the original optimal value;
i. judging whether iteration is finished, if so, finishing, otherwise, returning to the execution step f;
j. And finally outputting a global individual optimal particle value, wherein the global individual optimal particle value comprises a learning rate lr0, cosine annealing super parameters lrf, learning rate momentum and a weight attenuation coefficient weight_decay.
Example 3
The difference between this embodiment and embodiment 1 is that the HIL quantum algorithm and the amplitude amplification equivalent quantum optimization algorithm are used to perform the linear equation setIs a solution to (c).
As shown in fig. 5, a system of linear equationsThe solving steps of (a) are specifically as follows:
Preparation of input Quantum states And set upI u j > is the eigen state of matrix F (EIGENSTATES);
Converting matrix F into unitary operation
Quantum phase estimation is applied to the clock register and the input register, so that the clock register and the input register respectively obtainEigenvalue |λ j > and eigenvector
The clock register is used as a control quantum bit to rotate an auxiliary quantum bit, and the auxiliary quantum bit is converted into an overlapped state of |0> and |1 >; auxiliary qubit controlled rotation operation extracts lambda j to probability amplitude from ground state |lambda j > respectivelyAndWherein the method comprises the steps ofThe quantum state in all three registers will become
The amplitude amplification module is used for increasing the amplitude of the quantum state in the register to be |1>;
the auxiliary register is measured, if the result is |1>, the result input to the register will be AND And (5) a proportional calculation result, and finishing calculation.
Wherein the amplitude amplification (Amplitude Amplification) is an important kernel of the Grover search algorithm, as shown in fig. 5, the Grover algorithm geometry angle is explained as follows: placing N elements into N qubits (N=2 n), wherein M are correctly solved, M is more than or equal to 1 and less than or equal to N, and applying to N qubits in initial stateThe equal weight uniform superposition state of all the calculated ground states |0 > can be obtained:
Where x is a numeric string consisting of |0 >, |1 >, let n=2 n, the above-mentioned uniform superimposed state |ψ >:
If x is a correct solution, f (x) =1, if x is an incorrect solution, f (x) =0, the correct solution can be identified by quantum Oracle, |x > → (-1) f(x) |x >, and the description form of the superposition state is:
the above formula consists of two parts f (x) =0 and f (x) =1, wherein, Corresponding to all x-constituent quantum states of f (x) =0,Corresponding to the quantum states of all x components of f (x) =1, the superposition state description is:
Is provided with Let G= (2|ψ > < ψ| -I) O, where |ψ > < ψ| is the corresponding operator of the system uniform superposition state, I is the unit operator, and O is the Oracle operator
The result after 1 iteration of G operator for the uniform superposition state |ψ > is as follows:
the result after k G operator iterations for the uniform superposition |ψ > is as follows:
The G operator can be used for multiple times to enable |ψ > to be continuously close to |β >, and finally the |ψ > can be measured to obtain a ground state of |β > with high probability, namely, a correct solution corresponding to f (x) =1.
After all the above processes are finished, the input register will be composed ofBecomes into Wherein the method comprises the steps ofTherefore, solving of a linear equation set is achieved, and the maximum interval hyperplane is constructed.
And (3) training the support vector machine model based on the quantum optimization algorithm by using the training data set in the step (2), so as to construct the traffic video vehicle feature classification recognition model for unstructured.
According to the invention, the data sets formed by different vehicle brand information are adopted for preliminary test, the accuracy of vehicle brand identification in the traffic video information can reach 96.2% based on the support vector machine model of the quantum optimization algorithm, and the accurate classification identification of the traffic video vehicle characteristics is basically realized.
Finally, it should be noted that the above-mentioned embodiments are only for illustrating the technical scheme of the present invention and are not limiting; while the invention has been described in detail with reference to the preferred embodiments, those skilled in the art will appreciate that: modifications may be made to the specific embodiments of the present invention or equivalents may be substituted for part of the technical features thereof; without departing from the spirit of the invention, it is intended to cover the scope of the invention as claimed.
Claims (5)
1. The traffic video vehicle classification method based on the quantum optimization algorithm is characterized by comprising the following steps of:
Step 1, performing unstructured traffic video vehicle coarse-granularity target detection by using a target detection algorithm, and extracting a target vehicle picture in a traffic video;
Step 2, extracting vehicle brand features from the target vehicle pictures by using a deep learning network algorithm, preprocessing the extracted vehicle brand features, and dividing preprocessed data into a training data set and a test data set according to 70%/30%;
step 3, constructing and training a traffic video vehicle feature classification model by utilizing the training data set in the step 2 and combining a quantum optimization algorithm with a traditional support vector machine;
the construction steps of the traffic video vehicle feature classification model are as follows:
The sample form of the training dataset is set as follows: Wherein the method comprises the steps of For 512-dimensional feature vectors of training samples, x j is 512-dimensional attribute feature data, y j is a real vehicle brand corresponding to the training samples, N is feature vector dimensions of the training samples in the training dataset and n=512, m is the total number of samples in the training dataset;
the specific objective of the method is to seek the maximum interval hyperplane to realize the optimal brand feature classification of the vehicle, and obtain an objective function after conversion as follows Is a weight vector; regarding noise fault tolerance data as approximate linear separable, and searching soft boundary hyperplane; introducing a relaxation variable e j for indicating the allowable data point deviation, the objective function isWherein gamma is a penalty coefficient, b is a bias constant, and the constraint is constrained by inequalityConversion to equality constraints
Introducing Lagrangian function multipliersConstruction
Alpha j is greater than or equal to 0 and corresponds toIs a product of the lagrangian multipliers of (c),Is [ alpha 1,α2,…,αj,…,αM ], Is a relaxation vector;
based on KKT condition pair Obtaining partial derivativeI.e. Thereby converting the quadratic programming optimization problem into a problem of solving a linear equation set;
Wherein the method comprises the steps of I＝[1,1,…,1]， AndInput vectors for the jth and kth models, j, k=1, 2, …, M, respectively; e is an identity matrix;
solving a linear equation set to obtain an optimal hyperplane parameter And b, thereby constructing a support vector machine model for traffic video vehicle feature analysis Namely a traffic video vehicle characteristic classification model;
And 4, carrying out classification test on the test data set in the step 2 by using the trained traffic video vehicle classification model to obtain a traffic video vehicle prediction classification result, and comparing and analyzing the traffic video vehicle prediction classification result with a real vehicle brand result to verify the correctness of the traffic video vehicle feature classification model.
2. The traffic video vehicle classification method based on the quantum optimization algorithm according to claim 1, wherein: the target detection algorithm in step1 is YOLOv detection algorithm.
3. The traffic video vehicle classification method based on the quantum optimization algorithm according to claim 2, wherein: before carrying out coarse-granularity target detection on unstructured traffic video vehicles by using YOLOv detection algorithm, the super-parameter learning rate lr0, cosine annealing super-parameter lrf, learning rate momentum and weight attenuation coefficient weight_decay of the YOLOv detection algorithm are required to be synchronously and comprehensively optimized by using quantum particle swarm optimization QPSO algorithm to obtain the optimal super-parameter learning rate lr0, optimal cosine annealing super-parameter lrf, optimal learning rate momentum and optimal weight attenuation coefficient weight_decay.
4. The traffic video vehicle classification method based on a quantum optimization algorithm according to claim 3, wherein the specific steps of performing synchronous comprehensive optimization on the super parameter learning rate lr0, the cosine annealing super parameter lrf, the learning rate momentum and the weight attenuation coefficient weight_decay of the YOLOv detection algorithm by using the quantum particle swarm optimization QPSO algorithm are as follows:
a. initializing four super parameters to be optimized, namely a learning rate lr0, a cosine annealing super parameter lrf, a learning rate momentum and a weight attenuation coefficient weight_decay of a YOLOv detection algorithm, and setting a reasonable optimizing interval;
b. the quantum particle swarm optimization initial parameter setting comprises particle swarm size and dimension;
c. initializing probability amplitude codes by using individual position information of a quantum particle swarm;
d. transforming the initialized randomly generated particle position probability amplitude into a quantum solution space;
e. Calculating fitness values of all particles based on the objective function to obtain an initial individual optimal value and a global optimal value;
f. Transforming the quantum state probability amplitude by utilizing the quantum rotating gate to realize particle position updating and generate a new particle swarm;
g. Calculating the fitness value of each particle in the new population based on the objective function to obtain a new individual optimal value and a global optimal value;
h. comparing and judging the information with the original optimal value, if the information is superior to the original optimal value, updating and replacing the information, otherwise, keeping the information of the original optimal value;
i. judging whether iteration is finished, if so, finishing, otherwise, returning to the execution step f;
j. And finally outputting a global individual optimal particle value, wherein the global individual optimal particle value comprises a learning rate lr0, cosine annealing super parameters lrf, learning rate momentum and a weight attenuation coefficient weight_decay.
5. The traffic video vehicle classification method based on the quantum optimization algorithm according to claim 1, wherein in step 2, a deep learning network algorithm selected when the vehicle brand features are extracted from the target vehicle picture by using the deep learning network algorithm is a base network layer of FASTER RCNN network structure, and comprises 13 convolution layers and 4 pooling layers, wherein one ReLU function process is arranged behind each convolution layer, the convolution kernel of the convolution layer is 3*3, the step length is 1, the pooling size of the pooling layers is 2 x 2, and the step length is 2; the 4096-dimensional features are obtained after processing by the base network layer, and the first 512-dimensional features are selected as vehicle brand features.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111662993.5A CN114494947B (en) | 2021-12-31 | 2021-12-31 | Traffic video vehicle classification method based on quantum optimization algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111662993.5A CN114494947B (en) | 2021-12-31 | 2021-12-31 | Traffic video vehicle classification method based on quantum optimization algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114494947A CN114494947A (en) | 2022-05-13 |
| CN114494947B true CN114494947B (en) | 2024-09-10 |
Family
ID=81508639
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111662993.5A Active CN114494947B (en) | 2021-12-31 | 2021-12-31 | Traffic video vehicle classification method based on quantum optimization algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114494947B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102023100405A1 (en) | 2023-01-10 | 2024-07-11 | Dr. Ing. H.C. F. Porsche Aktiengesellschaft | Method and system for production planning for a vehicle |
| CN117649668B (en) * | 2023-12-22 | 2024-06-14 | 南京天溯自动化控制系统有限公司 | Medical equipment metering certificate identification and analysis method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106371610A (en) * | 2016-09-23 | 2017-02-01 | 重庆金瓯科技发展有限责任公司 | Method for detecting driving fatigue based on electroencephalogram |
| CN109597043A (en) * | 2018-11-16 | 2019-04-09 | 江苏科技大学 | Radar Signal Recognition method based on quantum particle swarm convolutional neural networks |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016201679A1 (en) * | 2015-06-18 | 2016-12-22 | 华为技术有限公司 | Feature extraction method, lip-reading classification method, device and apparatus |
| CN111260118B (en) * | 2020-01-10 | 2022-08-23 | 天津理工大学 | Vehicle networking traffic flow prediction method based on quantum particle swarm optimization strategy |
-
2021
- 2021-12-31 CN CN202111662993.5A patent/CN114494947B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106371610A (en) * | 2016-09-23 | 2017-02-01 | 重庆金瓯科技发展有限责任公司 | Method for detecting driving fatigue based on electroencephalogram |
| CN109597043A (en) * | 2018-11-16 | 2019-04-09 | 江苏科技大学 | Radar Signal Recognition method based on quantum particle swarm convolutional neural networks |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114494947A (en) | 2022-05-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Diallo et al. | Deep embedding clustering based on contractive autoencoder | |
| CN110287800B (en) | Remote sensing image scene classification method based on SGSE-GAN | |
| CN110516095B (en) | Semantic migration-based weak supervision deep hash social image retrieval method and system | |
| CN101551809B (en) | Search method of SAR images classified based on Gauss hybrid model | |
| CN102693299B (en) | System and method for parallel video copy detection | |
| WO2022077646A1 (en) | Method and apparatus for training student model for image processing | |
| CN114494947B (en) | Traffic video vehicle classification method based on quantum optimization algorithm | |
| CN105138973A (en) | Face authentication method and device | |
| CN114169442B (en) | Remote sensing image small sample scene classification method based on double prototype network | |
| CN102663351A (en) | Face characteristic point automation calibration method based on conditional appearance model | |
| CN110082738B (en) | Radar target identification method based on Gaussian mixture and tensor recurrent neural network | |
| CN114511710A (en) | Image target detection method based on convolutional neural network | |
| Li et al. | Quantum support vector machine for classifying noisy data | |
| Tao et al. | Chaotic time series prediction based on radial basis function network | |
| CN109871907A (en) | Radar target high resolution range profile recognition methods based on SAE-HMM model | |
| CN112257817A (en) | Geological geology online semantic recognition method and device and electronic equipment | |
| CN113361476B (en) | Zhang Heng one-number pre-earthquake abnormal signal identification method based on artificial intelligence technology | |
| CN114140524A (en) | Closed loop detection system and method for multi-scale feature fusion | |
| Xie et al. | DECOR: Dynamic Decoupling and Multi-Objective Optimization for Long-tailed Remote Sensing Image Classification | |
| CN118761478B (en) | Method for improving generalization performance of quantum machine learning model | |
| CN110598071A (en) | Nonlinear data classification method based on DPSO-WKELM-IFSVM | |
| Zhao et al. | Industrial fault diagnosis based on few shot learning | |
| CN116071636B (en) | Commodity image retrieval method | |
| CN116883469B (en) | Point cloud registration method based on EIV model description under plane feature constraint | |
| CN118504427B (en) | Accurate segmentation method for remote sensing image |
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 |