Unmanned driving image recognition method and system based on intelligent optimization

Abstract

The present invention proposes an unmanned driving image recognition method and system based on intelligent optimization, which belongs to the field of image processing technology. The present invention adopts a refraction reverse learning mechanism to improve the initialization of the wave search algorithm, utilizes LEVY flight to improve the global search of the wave search algorithm, and combines the golden sine algorithm to improve the electromagnetic wave emission process in the local optimization of the wave search algorithm to obtain an improved wave search algorithm; the improved wave search algorithm is used to train the three weights added in the Triplet Attention mechanism of the convolutional neural network structure, and an improved convolutional neural network structure is obtained for the recognition of unmanned driving images. The present invention improves the accuracy of unmanned driving image recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an unmanned driving image recognition method and system based on intelligent optimization. Background Art

[0002] The statements in this section merely present background art related to the present invention and do not necessarily constitute prior art.

[0003] With the vigorous development of artificial intelligence technology, driverless driving has gradually entered a stage of rapid development. Compared with manned driving, driverless driving is safer and can ease traffic pressure and reduce traffic accidents. Driverless driving technology consists of three major parts: environmental perception, planning and decision-making, and trajectory tracking, and environmental perception is the primary link and key step of driverless driving. The road is the realization environment of environmental perception. It can accurately identify targets in the road environment and fully obtain obstacle information, which is of great significance for the realization of driverless driving. The current driverless driving technology lacks the ability to recognize small targets. How to improve the ability to recognize small targets during driverless driving has become an important problem to be solved in this field.

[0004] As an important branch of deep learning, convolutional neural networks have achieved remarkable results in the field of autonomous driving image recognition. However, with the increasing complexity of application scenarios and the continuous growth of data volume, traditional convolutional neural network structures face many challenges, such as low accuracy in small target recognition, and cannot be effectively applied to high-precision recognition of autonomous driving images. Summary of the invention

[0005] In order to solve the shortcomings of the prior art, the present invention proposes an unmanned driving image recognition method and system based on intelligent optimization, adopts the refraction reverse learning mechanism to improve the initialization of the wave search algorithm, uses LEVY flight to improve the global search of the wave search algorithm, combines the golden sine algorithm to improve the electromagnetic wave emission process in the local optimization of the wave search algorithm, and obtains an improved wave search algorithm. The improved wave search algorithm is used to add three weights (w 1 , w 2 , w 3 ) is trained to obtain an improved convolutional neural network structure, which in turn improves the accuracy of unmanned driving image recognition.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention proposes an unmanned driving image recognition method based on intelligent optimization.

[0008] An unmanned driving image recognition method based on intelligent optimization includes the following processes:

[0009] Add trainable first weight, second weight and third weight to the three branches in the Triplet Attention mechanism of the convolutional neural network structure, where the sum of the first weight, the second weight and the third weight is 1;

[0010] The refraction reverse learning mechanism is used to improve the initialization of the wave search algorithm. The global search of the wave search algorithm is improved by using LEVY flight. The process of transmitting electromagnetic waves in the local optimization of the wave search algorithm is improved by combining the golden sine algorithm to obtain an improved wave search algorithm. The first weight, the second weight and the third weight are trained using the improved wave search algorithm to obtain an improved convolutional neural network structure.

[0011] Based on the acquired unmanned driving images and the improved convolutional neural network structure, the final unmanned driving image recognition results are obtained.

[0012] In the second aspect, the present invention proposes an unmanned driving image recognition system based on intelligent optimization.

[0013] An unmanned driving image recognition system based on intelligent optimization, comprising:

[0014] A weight adding unit is configured to: add a trainable first weight, a second weight, and a third weight to three branches in the Triplet Attention mechanism of the convolutional neural network structure, respectively, wherein the sum of the first weight, the second weight, and the third weight is 1;

[0015] The weight optimization unit is configured to: adopt a refraction reverse learning mechanism to improve the initialization of the wave search algorithm, use LEVY flight to improve the global search of the wave search algorithm, combine the golden sine algorithm to improve the process of transmitting electromagnetic waves in the local optimization of the wave search algorithm, obtain an improved wave search algorithm, use the improved wave search algorithm to train the first weight, the second weight and the third weight, and obtain an improved convolutional neural network structure;

[0016] The image recognition unit is configured to obtain a final unmanned driving image recognition result based on the acquired unmanned driving image and the improved convolutional neural network structure.

[0017] In a third aspect, the present invention provides a computer device, comprising: a processor and a computer-readable storage medium;

[0018] a processor adapted to execute a computer program;

[0019] A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by the processor, the unmanned driving image recognition method based on intelligent optimization as described in the first aspect of the present invention is implemented.

[0020] In a fourth aspect, the present invention proposes a computer-readable storage medium, which stores a computer program, and the computer program is suitable for being loaded by a processor and executing the unmanned driving image recognition method based on intelligent optimization as described in the first aspect of the present invention.

[0021] In a fifth aspect, the present invention proposes a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the unmanned driving image recognition method based on intelligent optimization as described in the first aspect of the present invention.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] 1. The present invention adopts an improved Triplet Attention mechanism. The Triplet Attention mechanism uses three different branches to process the cross-dimensional interaction information between the channel C, width W and height H. The improved Triplet Attention mechanism adds trainable weights w to the three branches respectively. 1 、w 2 and w 3 , satisfying w 1 +w 2 +w 3 =1, each weight will be dynamically adjusted according to the importance of the three branches during training, which improves the recognition accuracy.

[0024] 2. The present invention adopts a refraction reverse learning mechanism to improve the initialization process of the wave search algorithm to seek better candidate solutions; utilizes LEVY flight to improve the global search of the wave search algorithm to expand the exploration scope of the algorithm and increase the optimization speed; combines the golden sine algorithm to improve the process of emitting electromagnetic waves in the local optimization of the wave search algorithm, combines the respective advantages of the wave search algorithm and the golden sine algorithm, and thereby obtains an improved wave search algorithm, making the algorithm more accurate and faster.

[0025] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0027] Figure 1 A schematic diagram of the process of the unmanned driving image recognition method based on intelligent optimization provided in Example 1 of the present invention;

[0028] Figure 2 An example of an unmanned driving image recognition system based on intelligent optimization provided in Embodiment 2 of the present invention;

[0029] Figure 3 This is an example of a computer device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0030] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0031] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0032] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0033] Embodiment 1:

[0034] This implementation proposes an unmanned driving image recognition method based on intelligent optimization, such as Figure 1 As shown, the following process is included:

[0035] S1: Add trainable weights w to the three branches of the Triplet Attention mechanism of the convolutional neural network structure 1 (i.e. the first weight), w 2 (i.e. the second weight) and w 3 (i.e. the third weight), and the weight satisfies w 1 +w 2 +w 3 =1, during training, it will be dynamically adjusted according to the importance of the three branches;

[0036] S2: The refraction reverse learning mechanism is used to improve the initialization of the wave search algorithm. The global search of the wave search algorithm is improved by LEVY flight. The electromagnetic wave emission process in the local optimization of the wave search algorithm is improved by combining the golden sine algorithm to obtain an improved wave search algorithm. The weight w added in the Triplet Attention mechanism of the convolutional neural network structure is used by the improved wave search algorithm.1 、w 2 and w 3 Training is performed to optimize its weights according to the recognition accuracy, thereby obtaining an improved convolutional neural network structure;

[0037] S3: Real-time acquisition of images during the unmanned driving process, and recognition using an improved convolutional neural network structure for use in the intelligent driving process.

[0038] In step S1 of this implementation, the Triplet Attention mechanism of the convolutional neural network structure is improved, and three different branches are used to process the cross-dimensional interaction information between the channel C, width W and height H; two branches are responsible for capturing the cross-dimensional interaction between the channel C and the space H or W, and the other branch is used to capture the dependency between the spatial dimensions. In the three branches, the input feature vector dimension is converted to the dimension that needs to be calculated. For example, when calculating the mutual information between the C dimension and the W dimension, the input feature vector dimension needs to be changed from [B, C, H, W] to [B, H, C, W], where B (batch) is the batch dimension, indicating the number of samples, and the dimension does not need to be changed when calculating the dimension H and the dimension W. In this implementation, after the maximum pooling layer and the average pooling layer of the Z-Pool module, the channel dimension is superimposed and the number of channels is changed to 1. The dimension is converted to the initial state after normalization using the sigmoid function, and finally the average value of the feature map of the three branches is output. This implementation adds trainable weights w to the three branches respectively. 1 、w 2 and w 3 , and the weight satisfies w 1 +w 2 +w 3 =1, which will be dynamically adjusted according to the importance of the three branches during training.

[0039] In step S2 of this implementation, it is assumed that the number of electromagnetic wave particles in the wave search algorithm is n, the dimension of the optimization problem is d, t is the current iteration number, T is the total iteration number, and the position of the electromagnetic wave particle W is simulated by a matrix:

[0040]

[0041] Use f([W n1 ,W n2 ,...,W nd ]) represents the fitness value of the nth individual, and then prepare n random numbers k between 0 and 1 1 , k 2 ,···,k n , initialize the particle position:

[0042]

[0043] W=lb+y * (ub-lb) (3);

[0044] In the formula, y i is the uniform k i ,y * is a random value in y, lb and ub are the upper and lower boundaries of the search space respectively.

[0045] The traditional wave search algorithm includes a global development phase and a local exploration phase. The global development phase includes:

[0046] W i new =W min +r 1 ·(W max -W min ) (4);

[0047]

[0048] Where W min is a vector composed of the minimum values ​​of each dimension of W, W max is the vector of the maximum values ​​of each dimension of W, f mean is the average fitness value of all particles, r 1 Is a random number between 0 and 1.

[0049] The local exploration stage includes:

[0050] (1) Emitting electromagnetic waves:

[0051]

[0052] Where λ is the waveform size control coefficient, and The elements in m are column vectors that follow a normal distribution and are arranged in order. best is the current optimal position, W l For W according to W best The position matrix after rearrangement of the proximity, W li W l The i-th value of max is the maximum fitness value in the population.

[0053] (2) Reflected electromagnetic waves, including:

[0054] W i new =-ε·r 2 ·cos(πi / n w2 )·(W best -Wfi )+W fi (8);

[0055]

[0056] Where ε is the waveform size control coefficient, and r 2 is a random number between 0 and 1, n w2 is the number of particles simulating reflected electromagnetic waves, W fi is the position matrix after W is rearranged in order of fitness value from small to large.

[0057] (3) Receiving electromagnetic waves, including:

[0058]

[0059]

[0060] Where ρ is the receiving control coefficient, and η is a random number that follows a normal distribution, r 3 、r 4 、r 5 、r 6 is a random number between 0 and 1, n w3 is the number of particles simulating the reception of electromagnetic waves, ξ is the correction factor, ξ=(2t / T-0.7) / (0.78+|2t / T-0.7|)+1, is the uniform optimal position obtained by convolution of the historical optimal position, Among them I 1 =I 2 =I 3 =…=I t =1 / t.

[0061] This implementation introduces a fitting gradient descent method based on the central difference method, and its mathematical expression is:

[0062] g i =(f(W +Δi )-f(W -Δi )) / 2Δ (12);

[0063] W i =W i -step·g i (13);

[0064] Where W +Δi =W i +Δ,W -Δi =W i -Δ, Δ=10 -6, g represents the gradient, gi represents the gradient of the i-th particle, step represents the step coefficient, and the initial value step 0 =0.3, and the final step length is determined after step length tests.

[0065] In step S2 of this implementation, a mechanism for seeking a better candidate solution is proposed in combination with the refraction reverse learning mechanism, which can effectively improve the performance of the algorithm to a certain extent. The principle of the refraction reverse learning mechanism is: assuming that the search interval of the solution on the X-axis is [a, b], o is the midpoint of the interval [a, b], the Y-axis represents the normal, and the intersection of the X-axis and the Y-axis is o, α represents the incident angle, β represents the refraction angle, and the projection of the incident point on the X-axis is x, and the projection of the refraction point on the X-axis after the incident light is refracted is x * , l and l * represent the lengths of the incident light and the refracted light respectively.

[0066] According to the refractive index formula We can get:

[0067]

[0068] make Substituting into the above formula, we get:

[0069]

[0070] Transforming the above formula and extending it to multi-dimensional space, the solution of refraction reverse learning can be obtained as:

[0071]

[0072] In the formula, x i,j is the position of the i-th individual in the population on the j-th dimension, is x i,j The reverse solution of refraction, a j and b j They represent the lower and upper limits of the search space on the j-th dimension respectively.

[0073] Specifically in the wave search algorithm of the present application, after the wave search algorithm is initialized, the population W is brought into the formula for solving the refraction reverse learning to obtain a new population Q, the formula is:

[0074]

[0075] In formula (17), W i is the i-th particle, W i *represents the refraction reverse solution of the i-th particle, lb and ub are the upper and lower boundaries of the search space respectively, the original population of the wave search algorithm and the newly obtained population Q are sorted according to the fitness value, and the first n individuals are selected to form a new population.

[0076] In step S2 of this implementation, the LEVY flight strategy is a non-Gaussian random transformation process. The characteristics of the LEVY flight strategy are long-term random walks with small steps and occasional sudden jumps in direction with larger steps. These characteristics are simplified and Fourier transformed to obtain:

[0077] LEVY~μ=t -λ (18);

[0078] 1<λ<3 (19);

[0079] The flight trajectory of LEVY is simulated using the following formula:

[0080]

[0081] Among them, s is the step size of LEVY flight, μ and v are two random variables in LEVY flight strategy, and the calculation method of μ and v variables is as follows:

[0082]

[0083] σ μ and σ v The calculation formula is as follows:

[0084]

[0085] Where Γ is the gama distribution function.

[0086] Specifically in the wave search algorithm of the present application, LEVY flight is used to improve the global search of the wave search algorithm, and the LEVY flight strategy is added to the global development stage of the wave search algorithm, and the new formula is obtained:

[0087] W i new =W min +r 1 ·(W max -W min )·s (23);

[0088]

[0089] Where W min is a vector composed of the minimum values ​​of each dimension of W, W max is the vector of the maximum values ​​of each dimension of W, f mean is the average fitness value of all particles, r1 is a random number between 0 and 1, s is the step length of LEVY flight, and the calculation formulas of specific parameters are shown in formula (20), formula (21) and formula (22).

[0090] In step S2 of this implementation, the golden sine algorithm uses the characteristics of the sine function in the trigonometric function and the golden section coefficient to continuously iterate until the optimal value is found. It has the characteristics of few parameters, good robustness and easy implementation. The specific steps are:

[0091] First, initialize the population:

[0092] V=rand(n,d)·(ub-lb)+lb (25);

[0093] Where rand(n,d) is the initial position of the n populations in the d-dimensional space; ub is the upper limit of the solution space, and lb is the lower limit of the solution space.

[0094] Secondly, in order to find the global optimal value, the golden section search method is used to update the position of the population. The position update method of the i-th individual is:

[0095] V i t+1 =V i t |sin(R 1 )|+R 2 ·sin(R 1 )·|a 1 ·P i t -a 2 ·V i t | (26);

[0096] In the formula, R 1 represents the distance moved by the i-th individual in the next iteration and R 1 ∈[0,2π]; R 2 ∈[0,π] determines the direction of position update of the i-th individual in the next iteration; and is the coefficient obtained by the golden section method, represents the golden ratio; P i t is the optimal position of the ith individual in the tth iteration.

[0097] Specifically, in the wave search algorithm of the present application, the golden sine algorithm is combined to improve the process of transmitting electromagnetic waves in the local optimization of the wave search algorithm, and the golden sine algorithm is used to improve the process of transmitting electromagnetic waves in the local exploration stage of the wave search algorithm. The improved formula is:

[0098]

[0099] In the formula, R 1 ∈[0,2π], R 2 ∈[0,π], and is the coefficient obtained by the golden section method, Represents the golden ratio.

[0100] The formulas for the reflected electromagnetic wave and received electromagnetic wave processes in the local exploration phase of the wave search algorithm remain unchanged, that is, the specific formulas of formula (8), formula (9), formula (10) and formula (11) are adopted.

[0101] Unmanned driving images are divided into road motion images and construction site static images. Among them, road motion images are related images during road driving, and the images should include large targets such as vehicles and obstacles, as well as small targets such as pedestrians and road signs; construction site static images are images related to construction machinery on the construction site, and the images should include large targets such as trucks, excavators, rollers, bulldozers, as well as small targets such as workers.

[0102] In this implementation, an improved wave search algorithm is used to improve the trainable weights w added to the three branches in the Triplet Attention mechanism. 1 、w 2 and w 3 When searching for the best value, f(x) = 0.6 × mAP 小目标 +0.4×mAP 大目标 As the fitness function, an improved convolutional neural network based on the attention mechanism is obtained, which is used to identify different types of targets in unmanned driving images.

[0103] Embodiment 2:

[0104] like Figure 2 As shown, this implementation method proposes an unmanned driving image recognition system based on intelligent optimization, including:

[0105] The weight adding unit is configured to: add a trainable first weight, a second weight and a third weight to the three branches in the Triplet Attention mechanism of the convolutional neural network structure, respectively, wherein the sum of the first weight, the second weight and the third weight is 1; the specific working process of this unit is the same as the process of step S1 in embodiment 1, which will not be repeated here;

[0106] The weight optimization unit is configured to: adopt a refraction reverse learning mechanism to improve the initialization of the wave search algorithm, use LEVY flight to improve the global search of the wave search algorithm, combine the golden sine algorithm to improve the process of transmitting electromagnetic waves in the local optimization of the wave search algorithm, obtain an improved wave search algorithm, use the improved wave search algorithm to train the first weight, the second weight and the third weight, and obtain an improved convolutional neural network structure; the specific working process of this unit is the same as the process of step S2 in embodiment 1, which will not be repeated here;

[0107] The image recognition unit is configured to obtain a final unmanned driving image recognition result based on the acquired unmanned driving image and the improved convolutional neural network structure.

[0108] It is understandable that the above-mentioned various units can be respectively or all merged into one or several other units to constitute, or some (some) units therein can also be split into multiple smaller units in function to constitute, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present application. The above-mentioned units are divided based on logical functions. In practical applications, the function of a unit can also be realized by multiple units, or the function of multiple units is realized by one unit. In other embodiments of the present application, the image recognition system can also include other units. In practical applications, these functions can also be assisted by other units to achieve, and can be achieved by the collaboration of multiple units.

[0109] According to another embodiment of the present application, the system described in this embodiment can be constructed and the image recognition method of the embodiment of the present application can be implemented by running a computer program (including program code) capable of executing the steps involved in the corresponding method described in Example 1 on a general computing device such as a computer including processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM). The computer program can be recorded on, for example, a computer-readable recording medium, and loaded into the above-mentioned computing device through the computer-readable recording medium and run therein.

[0110] Embodiment 3:

[0111] like Figure 3 As shown, this implementation proposes a computer device, which includes a processor 1001, a communication interface 1002, and a computer-readable storage medium 1003. The processor 1001, the communication interface 1002, and the computer-readable storage medium 1003 may be connected via a bus or other means.

[0112] Among them, the communication interface 1002 is used to receive and send data, the computer-readable storage medium 1003 can be stored in the memory of the electronic device, the computer-readable storage medium 1003 is used to store a computer program, the computer program includes program instructions, and the processor 1001 is used to execute the program instructions stored in the computer-readable storage medium 1003.

[0113] Processor 1001 (or CPU (Central Processing Unit)) is the computing core and control core of the electronic device, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to implement corresponding method processes or corresponding functions.

[0114] The processor 1001 is configured to execute the following process:

[0115] Add a trainable first weight, a second weight, and a third weight to the three branches in the Triplet Attention mechanism of the convolutional neural network structure, respectively, wherein the sum of the first weight, the second weight, and the third weight is 1; the specific working process is the same as the process of step S1 in Example 1, and will not be repeated here;

[0116] The refraction reverse learning mechanism is used to improve the initialization of the wave search algorithm, the LEVY flight is used to improve the global search of the wave search algorithm, and the golden sine algorithm is combined to improve the process of transmitting electromagnetic waves in the local optimization of the wave search algorithm to obtain an improved wave search algorithm. The improved wave search algorithm is used to train the first weight, the second weight and the third weight to obtain an improved convolutional neural network structure; the specific working process is the same as the process of step S2 in Example 1, which will not be repeated here;

[0117] Based on the acquired unmanned driving images and the improved convolutional neural network structure, the final unmanned driving image recognition results are obtained.

[0118] Embodiment 4:

[0119] This implementation method proposes a computer-readable storage medium (Memory), which is a memory device in an electronic device for storing programs and data. It is understandable that the computer-readable storage medium here can include both built-in storage media in the electronic device and, of course, extended storage media supported by the electronic device. The computer-readable storage medium provides a storage space that stores the processing system of the electronic device.

[0120] In addition, the storage space also stores one or more instructions suitable for being loaded and executed by the processor, and these instructions may be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here may be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as at least one disk storage; optionally, it may also be at least one computer-readable storage medium located away from the aforementioned processor.

[0121] In one embodiment, the computer-readable storage medium stores one or more instructions; the processor loads and executes the one or more instructions stored in the computer-readable storage medium to implement the following process:

[0122] Add a trainable first weight, a second weight, and a third weight to the three branches in the Triplet Attention mechanism of the convolutional neural network structure, respectively, wherein the sum of the first weight, the second weight, and the third weight is 1; the specific working process is the same as the process of step S1 in Example 1, and will not be repeated here;

[0123] The refraction reverse learning mechanism is used to improve the initialization of the wave search algorithm, the LEVY flight is used to improve the global search of the wave search algorithm, and the golden sine algorithm is combined to improve the process of transmitting electromagnetic waves in the local optimization of the wave search algorithm to obtain an improved wave search algorithm. The improved wave search algorithm is used to train the first weight, the second weight and the third weight to obtain an improved convolutional neural network structure; the specific working process is the same as the process of step S2 in Example 1, which will not be repeated here;

[0124] Based on the acquired unmanned driving images and the improved convolutional neural network structure, the final unmanned driving image recognition results are obtained.

[0125] Embodiment 5:

[0126] The present implementation provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device performs the following process:

[0127] Add a trainable first weight, a second weight, and a third weight to the three branches in the Triplet Attention mechanism of the convolutional neural network structure, respectively, wherein the sum of the first weight, the second weight, and the third weight is 1; the specific working process is the same as the process of step S1 in Example 1, and will not be repeated here;

[0128] The refraction reverse learning mechanism is used to improve the initialization of the wave search algorithm, the LEVY flight is used to improve the global search of the wave search algorithm, and the golden sine algorithm is combined to improve the process of transmitting electromagnetic waves in the local optimization of the wave search algorithm to obtain an improved wave search algorithm. The improved wave search algorithm is used to train the first weight, the second weight and the third weight to obtain an improved convolutional neural network structure; the specific working process is the same as the process of step S2 in Example 1, which will not be repeated here;

[0129] Based on the acquired unmanned driving images and the improved convolutional neural network structure, the final unmanned driving image recognition results are obtained.

[0130] A person skilled in the art can appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0131] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loading and executing computer program instructions on a computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instruction can be stored in a computer-readable storage medium or transmitted by a computer-readable storage medium. The computer instruction can be transmitted from a website site, a computer, a server or a data center to another website site, a computer, a server or a data center by wired (for example, coaxial cable, optical fiber, digital line (DSL)) or wireless (for example, infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that a computer can access or a data processing device such as a server, a data center, etc. that contains one or more available media integration. Available media can be magnetic media (for example, floppy disk, hard disk, tape), optical media (for example, DVD), or semiconductor media (for example, solid-state drive (Solid State Disk, SSD)) and the like.

[0132] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An unmanned driving image recognition method based on intelligent optimization, characterized in that: The process includes: Add trainable first weight, second weight and third weight to the three branches in the Triplet Attention mechanism of the convolutional neural network structure, where the sum of the first weight, the second weight and the third weight is 1; The refraction reverse learning mechanism is used to improve the initialization of the wave search algorithm. The global search of the wave search algorithm is improved by using LEVY flight. The process of transmitting electromagnetic waves in the local optimization of the wave search algorithm is improved by combining the golden sine algorithm to obtain an improved wave search algorithm. The first weight, the second weight and the third weight are trained using the improved wave search algorithm to obtain an improved convolutional neural network structure. Adding the LEVY flight strategy to the global development phase of the wave search algorithm, the new formula is: In the formula, W min for W A vector consisting of the minimum values ​​in each dimension, W max is a vector of the maximum values ​​of each dimension of W, f mean is the average fitness value of all particles, r 1 is a random number between 0 and 1, s is the step length of LEVY flight, and the specific formula is: in, and v are two random variables in the LEVY flight strategy, and v The variables are calculated as follows: and The calculation formula is as follows: in, is the gama distribution function; The golden sine algorithm is used to improve the electromagnetic wave transmission process in the local exploration stage of the wave search algorithm. The improved formula for transmitting electromagnetic waves is: In the formula, R 1∈[0,2π], R 2∈[0,π], and is the coefficient obtained by the golden section method, represents the golden ratio coefficient, is the waveform size control coefficient, and , m The elements in are column vectors that follow a normal distribution and are arranged in order. is the current optimal position, for W According to The position matrix after rearrangement of the proximity, for No. values, fmax is the maximum fitness value in the population; Based on the acquired unmanned driving images and the improved convolutional neural network structure, the final unmanned driving image recognition results are obtained.

2. The unmanned driving image recognition method based on intelligent optimization according to claim 1, characterized in that: Three branches are used to process channels C ,width W and height H In order to obtain the cross-dimensional interaction information between the two, in the three branches, the input feature vector dimension is converted to the dimension that needs to be calculated. After passing through the maximum pooling layer and the average pooling layer of the Z-Pool module, it is superimposed on the channel dimension and the number of channels is changed to 1. After normalization using the sigmoid function, the dimension is converted to the initial state, and the average value of the feature maps of the three branches is output.

3. The unmanned driving image recognition method based on intelligent optimization according to claim 1, characterized in that: The refraction reverse learning mechanism is used to improve the initialization of the wave search algorithm, including: assuming that the search interval of the solution on the X-axis is [ a , b ], and represent the lengths of the incident and refracted light rays, respectively. After the wave search algorithm is initialized, the population W Substitute it into the formula for solving refraction reverse learning to obtain a new population , the formula is: in, For the Particles, Indicates The reverse solution of the refraction of a particle is, and are the upper and lower boundaries of the search space respectively, and the original population of the wave search algorithm is compared with the newly obtained population , sort by fitness value and select the first n Individuals form a new population.

4. The unmanned driving image recognition method based on intelligent optimization according to claim 1, characterized in that: The fitting gradient descent method based on the central difference method is introduced, including: In the formula, , , , g represents the gradient, Indicates i The gradient of a particle, step Represents the step size coefficient, initial value 0.3, the final step length is determined after step length experiments.

5. An unmanned driving image recognition system based on intelligent optimization, characterized in that: include: The weight adding unit is configured to: add a trainable first weight, a second weight, and a third weight to three branches in the Triplet Attention mechanism of the convolutional neural network structure, respectively, wherein the sum of the first weight, the second weight, and the third weight is 1; The weight optimization unit is configured to: adopt a refraction reverse learning mechanism to improve the initialization of the wave search algorithm, use LEVY flight to improve the global search of the wave search algorithm, combine the golden sine algorithm to improve the process of transmitting electromagnetic waves in the local optimization of the wave search algorithm, obtain an improved wave search algorithm, use the improved wave search algorithm to train the first weight, the second weight and the third weight, and obtain an improved convolutional neural network structure; Adding the LEVY flight strategy to the global development phase of the wave search algorithm, the new formula is: In the formula, W min for W A vector consisting of the minimum values ​​in each dimension, W max is a vector of the maximum values ​​of each dimension of W, f mean is the average fitness value of all particles, r 1 is a random number between 0 and 1, s is the step length of LEVY flight, and the specific formula is: in, and v are two random variables in the LEVY flight strategy, and v The variables are calculated as follows: and The calculation formula is as follows: in, is the gama distribution function; The golden sine algorithm is used to improve the electromagnetic wave transmission process in the local exploration stage of the wave search algorithm. The improved formula for transmitting electromagnetic waves is: In the formula, R 1∈[0,2π], R 2∈[0,π], and is the coefficient obtained by the golden section method, represents the golden ratio coefficient, is the waveform size control coefficient, and , m The elements in are column vectors that follow a normal distribution and are arranged in order. is the current optimal position, for W According to The position matrix after rearrangement of the proximity, for No. values, fmax is the maximum fitness value in the population; The image recognition unit is configured to obtain a final unmanned driving image recognition result based on the acquired unmanned driving image and the improved convolutional neural network structure.

6. A computer device, characterized in that: include: a processor and a computer readable storage medium; a processor adapted to execute a computer program; A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by the processor, the unmanned driving image recognition method based on intelligent optimization as described in any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which is suitable for being loaded by a processor and executing the unmanned driving image recognition method based on intelligent optimization as described in any one of claims 1 to 4.

8. A computer program product, characterized in that The computer program product includes a computer program, and when the computer program is executed by a processor, it implements the unmanned driving image recognition method based on intelligent optimization as described in any one of claims 1 to 4.

Images