A waterway monitoring and early warning method based on high-frequency millimeter-wave radar

Abstract

The present invention discloses a waterway monitoring and early warning method based on high-frequency millimeter-wave radar. The high-frequency millimeter-wave radar is used to detect underwater obstacles in the waterway during the travel of the ship, and the time-frequency image data collected by the radar is obtained; the time-frequency image is randomly reversed and cropped to expand the data set for training; the improved Fire Eagle algorithm is used to optimize the Resnet network model and build a recognition model; the SPM chaotic mapping method is used to make the Fire Eagle algorithm jump out of the local optimum and accelerate the convergence speed; the collected time-frequency image data set is used to train the improved Resnet network model; the model is put into use, and during the travel of the ship, reef obstacles that may cause potential harm are identified and an alarm is issued. Based on the existing technology, the present invention accelerates the convergence speed of the Resnet network model and further improves the accuracy and robustness of the radar recognition system.

Description

Technical Field

[0001] The present invention relates to the field of radar detection technology, and in particular to a waterway monitoring and early warning method based on high-frequency millimeter-wave radar. Background Art

[0002] my country's water transport routes are extensive, connecting all parts of the country. For bulk goods, such as ore, coal, steel, wood, etc., and other goods that do not have strict time requirements, sea transportation can also be a better choice. In general, water transportation plays an indispensable role in my country's cargo transportation with its economic, flexible, and environmentally friendly characteristics.

[0003] China's water transport industry has achieved remarkable development and has also made great improvements in safety, but water transport accidents still occur, posing a threat to the safety of personnel and property. Accidents often occur in inland waterways, where ships collide and run aground. Bad weather can also cause shipwrecks.

[0004] In most cases, boat operators will choose to use radar to scan the underwater environment and warn of possible reefs or obstacles to avoid, but commonly used laser radars have narrow beams and are easily affected by bad weather or environmental interference. Under water, the laser will be offset and the accuracy will be further affected. At the same time, laser radars are expensive and economical.

[0005] In existing technologies, radars are paired with neural network models to enhance image recognition accuracy. Conventional Resnet models take a lot of time to train, which affects their efficiency. Summary of the invention

[0006] Purpose of the invention: In response to the problems mentioned in the background technology, the present invention provides a waterway monitoring and early warning method based on high-frequency millimeter-wave radar. High-frequency millimeter-wave radar is used for detection to improve recognition accuracy and robustness; the SPM chaotic mapping method is used to add a random preference swimming strategy to improve the Fire Eagle algorithm, so that it has the ability to accelerate convergence and jump out of local optimality; the improved Fire Eagle optimization algorithm is combined with the Resnet network model to further accelerate the model training speed; the trained model is used in combination with radar scanning to perform higher-precision recognition of underwater reefs and obstacles when the ship is sailing, and an alarm is sent if an obstacle exists.

[0007] Technical solution: The present invention provides a waterway monitoring and early warning method based on high-frequency millimeter-wave radar, the method comprising:

[0008] S1 controls the radar to detect underwater reef obstacles, uses short-time Fourier transform to perform time-frequency joint analysis on echo data, obtains two-dimensional information in time-frequency and frequency domains, and extracts richer data features;

[0009] S2 randomly reverses the time-frequency images and crops and expands the data set;

[0010] S3 improves the Fire Eagle algorithm. When the Fire Eagle algorithm generates random individuals, it uses the SPM chaotic mapping method and adds a random preference swimming strategy to the algorithm to improve the position update method, so that the algorithm has a certain ability to jump out of the local optimum and accelerate the convergence of the algorithm.

[0011] S4 uses the improved Fire Eagle algorithm to optimize the weights and thresholds of the convolution operation in the Resnet network model, iteratively outputs the optimal solution and substitutes it into the Resnet model, and uses the optimized Resnet network model to build an image recognition model;

[0012] S5 uses the collected data set to train the Resnet network model, and after training, it is put into use on the ship;

[0013] S6 initializes the main control chip and radar chip, controls the radar to scan, inputs the data image obtained by the radar into the central control platform of the integrated recognition model, determines whether there are obstacles in the waters below the ship, and sends an alarm if there are obstacles.

[0014] Furthermore, the radar described in S1 is a high-frequency millimeter-wave radar. When scanning is performed using a high-frequency millimeter-wave radar, the radar transmitting antenna transmits a linear frequency modulation signal, which is scattered after encountering an obstacle, and the receiving antenna receives the scattered echo signal. The echo signal is:

[0015]

[0016] Among them, f 0 is the starting frequency; k is the linear frequency modulation frequency, which indicates the growth rate of the linear frequency modulation signal over time; T is the delay of the echo signal relative to the transmitted signal;

[0017] Furthermore, when performing short-time Fourier transform on the signal, it is necessary to process the sampled and quantized digital signal, and convert the short-time Fourier transform after discretization in the time domain and frequency domain into

[0018]

[0019] Where m and n are sampling points of time and frequency respectively, x(k) represents the sample at time k, h(kt-mt) is a window function, where t is the time variable. The window function is used to slide on the signal to analyze the frequency content of the signal in different time periods. -j2πnfk It is a complex exponential function used to calculate the Fourier coefficients of different frequencies nf. After processing, the time-frequency diagram of the echo signal is obtained. The improved Fire Eagle optimization algorithm combined with the Resnet model is used to identify the time-frequency diagram.

[0020] Furthermore, SPM chaotic mapping method is used in S3 to select the initial position of the fire eagle. The chaotic search idea is mainly to generate a chaotic sequence through some kind of iteration. The formula of SPM chaotic mapping is:

[0021]

[0022] Among them, when η∈(0,1),μ∈(0,1), the system is in a chaotic state, and r is a random number between 0 and 1. In the algorithm iteration, Fire Eagle continuously updates the population position, which is the process of continuously optimizing the weights and thresholds of the convolution module.

[0023] Furthermore, in the process of Firehawk continuously updating the population position, the random preference walk strategy is used to improve the position update. The random preference walk is essentially a weight factor ω(t). The weight factor is used to enhance the individual's self-learning, and the individual position information is borrowed to make the individual have memory. The formula is as follows:

[0024]

[0025]

[0026] in represents the position of the population at time t+1 on dimension j, represents the inertia part, ω(t) represents the inertia weight, represents the individual cognitive part, S 1 represents the individual learning factor, represents the value of population i at latitude j in its historical best position, S 2 ×rand× represents the social cognitive part, where S 2 is the social learning factor, Represents the value of the global best position in dimension i.

[0027] Furthermore, the Fire Eagle algorithm described in S4 optimizes the weights and thresholds in the two convolution blocks in Resnet: initialize the Fire Eagle optimization algorithm, including the population size N, the problem dimension D, the number of prey n, the upper bound of the algorithm search space is lower bound of the algorithm search space, the maximum number of iterations, and the initial values ​​of the weights and thresholds; evaluate the fitness of the newly arrived Fire Eagle and prey to determine whether it is the best solution and whether the maximum number of iterations has been reached. If so, output the optimal solution divided into weights and thresholds; if not, use the random preference walking strategy to continue updating the Fire Eagle position.

[0028] Beneficial effects:

[0029] 1. The present invention uses high-frequency millimeter-wave radar for monitoring, which has advantages in recognition accuracy, cost, and robustness, and is suitable for the needs of ships to avoid obstacles during navigation.

[0030] 2. The present invention further inverts the time-frequency images acquired by the radar and crops them to expand the training data set, thereby improving the generalization ability of the Resnet network model, helping the model to better understand the true distribution of the data and cope with obstacles in different water conditions.

[0031] 3. The present invention adopts the SPM chaotic mapping method and adds a random preference movement strategy to improve the position update method to improve the Fire Eagle optimization algorithm, so that the algorithm has a certain ability to jump out of the local optimum and speeds up the algorithm.

[0032] 4. The present invention uses the improved Firehawk optimization algorithm to further optimize the weights and thresholds of convolution operations in the Resnet network model, saving training time and improving the efficiency of radar image precision processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is the overall flow chart of the present invention;

[0034] Figure 2 It is the principle diagram of the radar in the present invention;

[0035] Figure 3 It is the flow chart of Firehawk optimization algorithm in the present invention;

[0036] Figure 4 This is the structure diagram of the Resnet model in the present invention. DETAILED DESCRIPTION

[0037] The present application is further described in detail below in conjunction with the accompanying drawings. It is necessary to point out here that the following specific implementation methods are only used to further illustrate the present application and cannot be understood as limiting the scope of protection of the present application. Technical personnel in this field can make some non-essential improvements and adjustments to the present application based on the above application content.

[0038] The specific example of the present invention provides a waterway monitoring and early warning method based on high-frequency millimeter wave radar. Figure 1 The process shown in the figure includes the following steps:

[0039] 001 uses high-frequency millimeter-wave radar for scanning. The radar transmitting antenna transmits a linear frequency modulation signal, which is scattered after encountering an obstacle. The receiving antenna receives the scattered echo signal. The echo signal is

[0040]

[0041] Among them, f 0is the starting frequency; k is the linear frequency modulation frequency, which represents the growth rate of the linear frequency modulation signal over time; T is the delay of the echo signal relative to the transmitted signal.

[0042] The 002 transmission signal and the echo signal are processed by the mixer for de-linear frequency modulation to obtain the baseband signal. After the baseband signal is sampled by the ADC, a short-time Fourier transform is performed to obtain the frequency-time Doppler heat map.

[0043] 003 When performing short-time Fourier transform on a signal, it is necessary to process the sampled and quantized digital signal. Therefore, the short-time Fourier transform after discretization of the time domain and frequency domain is:

[0044]

[0045] Where m and n are sampling points of time and frequency respectively. After this processing, the time-frequency diagram of the echo signal can be obtained. The working principle of high-frequency millimeter-wave radar is as follows: Figure 2 After obtaining the time-frequency graph, the improved Fire Eagle optimization algorithm combined with the Resnet model is used to identify the time-frequency graph.

[0046] 004 Fire hawks are birds that attempt to spread fire intentionally by carrying burning sticks in their beaks and claws. They pick up burning sticks and when they see prey, they drop the sticks in their mouths. Insects and small mammals flee, and the hawks take the opportunity to hunt. The fire hawk optimization algorithm based on this inspiration is as follows: Figure 3 shown.

[0047] The 005 Fire Hawk Optimization (FHO) algorithm simulates the foraging behavior of a fire hawk, taking into account the process of making fire, making fire, and catching prey. First, multiple candidate solutions (X) are determined as the position vectors of the fire hawk and prey, and a random initialization process is used to identify the initial positions of these vectors in the search space.

[0048]

[0049]

[0050] Among them, x i represents the i-th candidate solution in the search space; d represents the dimension of the problem under consideration; N is the total number of candidate solutions in the search space; is the jth decision variable of the i-th candidate solution; (0) indicates the initial position of the candidate solution; and are the minimum and maximum bounds of the j-th decision variable of the ith candidate solution; rand is a uniformly distributed random number in the range [0,1].

[0051] 006 The positions of the Fire Hawk (FH) and the Prey (PR) are determined as follows:

[0052]

[0053]

[0054] In the formula, PR k is the kth prey in the search space with respect to the total number of n prey, FH 1 It is the i-th Firehawk among the total number of N Firehawks in the search space.

[0055] The total distance between the 007 Fire Hawk and its prey is:

[0056]

[0057] in, is the total distance between the first fire eagle and the kth prey; m is the total number of prey in the search space; n is the total number of fire eagles in the search space; (x1, y1) and (x2, y2) represent the coordinates of the fire eagle and prey in the search space.

[0058] After the above process of measuring the total distance between the fire hawk and the prey was carried out, the territories of these birds were distinguished by the nearest prey around them. By classifying the fire hawks and prey, the search process of the algorithm was designed. It should be noted that the fire hawk with a better objective function value selects the closest best prey in the search space of its specific territory. Then, the other fire hawks completed the next nearest prey in the search space, which proves that the stronger fire hawks completed more successful hunting than the weaker ones.

[0059] 008In the next phase of the algorithm, the fire hawks collect burning sticks from the main fire in order to light fires in the selected areas. In this phase, each fire hawk picks up a burning stick and throws it to its specific territory, forcing the prey to flee in a hurry. At the same time, some fire hawks are eager to use burning sticks from other fire hawks' territories. Therefore, these two behaviors can be used as the position update process in the main search loop of the FHO algorithm, as shown in the following equation:

[0060]

[0061] in, is the new position vector of the first fire hawk (FH); GB is the current optimal solution of the population in the search space and is considered to be the main fire; FH Near is another fire eagle in the search space; r1 and r2 are random numbers uniformly distributed in the range of (0,1), which are used to determine the movement of the fire eagle to the main fire and other fire eagle territories.

[0062] 009 simulates the movement of prey in each fire eagle's territory and designs the prey position update formula as follows:

[0063]

[0064] To simulate the movement of prey outside the territory of each fire eagle, the prey position update formula is designed as:

[0065]

[0066] in, is the new position vector (PR) of the qth prey surrounded by the first fire eagle q ); FH Alter is another Fire Eagle in the search space: SP here is a safe area outside the territory of the first Fire Eagle: r5 and r6 are random numbers uniformly distributed in the range of (0.1) and are used to determine the movement of prey to other Fire Eagles and safe places outside the territory.

[0067]

[0068]

[0069] SP i It is a safe place for the Fire Eagle, SP is SP i Outside the safe area.

[0070] 010 The Fire Eagle optimization algorithm is used to optimize the weights and thresholds in the two convolution blocks in Resnet:

[0071] Initialize the Firehawk optimization algorithm, including the population size N, the problem dimension D, the number of prey n, the upper bound of the algorithm search space and the lower bound of the algorithm search space, the maximum number of iterations, and the initial values ​​of the weights and thresholds.

[0072] The SPM chaotic mapping method is used to select the initial position of the fire eagle. The chaotic search idea is mainly to generate a chaotic sequence through some kind of iteration. The formula of SPM chaotic mapping is:

[0073]

[0074] Among them, when η∈(0,1),μ∈(0,1), the system is in a chaotic state, and r is a random number between 0 and 1. In the algorithm iteration, Fire Eagle continuously updates the population position, which is the process of continuously optimizing the weights and thresholds of the convolution module.

[0075] 011 When updating the position of the Firehawk, the present invention uses a random preference walk strategy to improve the position update:

[0076] The essence of random preference walk is a weight factor ω(t), which is used to enhance the individual's self-learning and to make the individual have memory by drawing on its own individual position information. The formula is as follows:

[0077]

[0078]

[0079] After many experiments, the improved algorithm has the best comprehensive optimization performance when ωmax=9,ωmin=0.4,k=1000. Maintaining a larger weight value in the early stage of optimization can expand the optimization space and maintain good global optimization performance; a smaller inertia weight in the later stage can enhance local development capabilities, allowing the algorithm to converge quickly, better balance the two stages of the algorithm, and increase the stability of the algorithm.

[0080] 012 The safe position under the first fire eagle area is calculated by formula (9), the updated position of the prey is determined by formula (7), the safe position outside the first fire eagle area is calculated by formula (10), and the updated position of the prey is determined by formula (8).

[0081] 013 Evaluate the fitness of the newly obtained fire eagle and prey to determine whether it is the best solution. Whether the maximum number of iterations has been reached, if so, output the optimal solution divided into weights and thresholds, if not, use the random preference walking strategy to continue updating the position of the fire eagle.

[0082] 014 Use the optimized weights and thresholds to train the Resnet model. After the model training is completed, the model is put into use. If a reef obstacle is detected, an early warning is issued.

[0083] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable people familiar with the technology to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent transformation or modification made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A waterway monitoring and early warning method based on high-frequency millimeter-wave radar, characterized in that: The method comprises the following steps: S1 controls the radar to detect underwater reef obstacles, and uses short-time Fourier transform to perform time-frequency joint analysis on echo data to obtain two-dimensional information in the time and frequency domains; S2 randomly reverses the time-frequency images and crops and expands the data set; S3 improves the Fire Eagle algorithm. When the Fire Eagle algorithm generates random individuals, it uses the SPM chaotic mapping method and adds a random preference swimming strategy to the algorithm to improve the position update method, so that the algorithm has a certain ability to jump out of the local optimum and accelerate the convergence of the algorithm. The essence of random preference walk is a weight factor ω(t). The weight factor is used to enhance the individual's self-learning and to make the individual have memory by drawing on its own individual position information. The formula is as follows: in represents the position of the population at time t+1 on dimension j, represents the inertia part, ω(t) represents the inertia weight, represents the individual cognitive part, S1 represents the individual learning factor, represents the value of population i at latitude j in its historical best position, represents the social cognitive part, where S2 is the social learning factor, Represents the value of the global best position in dimension i; S4 uses the improved Fire Eagle algorithm to optimize the weights and thresholds of the convolution operation in the Resnet network model, iteratively outputs the optimal solution and substitutes it into the Resnet model, and uses the optimized Resnet network model to build an image recognition model; S5 uses the collected data set to train the Resnet network model, and after training, it is put into use on the ship; S6 controls the radar to scan and inputs the data images obtained by the radar into the trained and optimized Resnet network model to determine whether there are obstacles in the waters below the ship. If there are obstacles, an alarm will be sent.

2. The waterway monitoring and early warning method based on high-frequency millimeter-wave radar according to claim 1 is characterized in that: The radar described in S1 is a high-frequency millimeter-wave radar. When scanning is performed using a high-frequency millimeter-wave radar, the radar transmitting antenna transmits a linear frequency modulation signal, which is scattered after encountering an obstacle. The receiving antenna receives the scattered echo signal, and the echo signal is: Wherein, f0 is the starting frequency; k is the linear frequency modulation frequency and represents the growth rate of the linear frequency modulation signal over time; and T is the time delay of the echo signal relative to the transmitted signal.

3. The waterway monitoring and early warning method based on high-frequency millimeter-wave radar according to claim 2 is characterized in that: When performing short-time Fourier transform on a signal, it is necessary to process the sampled and quantized digital signal, and the short-time Fourier transform after discretization of the time domain and frequency domain is: Where m and n are sampling points of time and frequency respectively, x(k) represents the sample at time k, h(kt-mt) is a window function, where t is the time variable. The window function is used to slide on the signal to analyze the frequency content of the signal in different time periods. -j2πnfk It is a complex exponential function used to calculate the Fourier coefficients of different frequencies nf. After processing, the time-frequency diagram of the echo signal is obtained. The improved Fire Eagle optimization algorithm combined with the Resnet model is used to identify the time-frequency diagram.

4. The waterway monitoring and early warning method based on high-frequency millimeter-wave radar according to claim 1 is characterized in that: In S3, the SPM chaotic mapping method is used to select the initial position of the fire eagle. The chaotic search idea is mainly to generate a chaotic sequence through some kind of iteration. The formula of SPM chaotic mapping is: Among them, when η∈(0,1),μ∈(0,1), the system is in a chaotic state, and r is a random number between 0 and 1. In the algorithm iteration, Fire Eagle continuously updates the population position, which is the process of continuously optimizing the weights and thresholds of the convolution module.

5. The waterway monitoring and early warning method based on high-frequency millimeter-wave radar according to claim 1 is characterized in that: The Fire Eagle algorithm described in S4 optimizes the weights and thresholds in the two convolutional blocks in Resnet: Initialize the Fire Eagle optimization algorithm, including the population size N, the problem dimension D, the number of prey n, the upper bound of the algorithm search space is lower bound of the algorithm search space, the maximum number of iterations, and the initial values ​​of weights and thresholds; evaluate the fitness of the newly obtained Fire Eagle and prey to determine whether it is the best solution and whether the maximum number of iterations has been reached. If so, output the optimal solution divided into weights and thresholds; if not, use the random preference walking strategy to continue updating the Fire Eagle position.

Images