A method and system for dynamic monitoring of target organisms in marine ranches

Abstract

The invention discloses a method and system for dynamic monitoring of target organisms in a marine ranch. The method comprises the following steps: receiving echo signals emitted by different sonar array elements in real time, performing distance Fourier transform and azimuth Fourier transform on each echo signal, and obtaining a frequency spectrum of each echo signal; performing phase shift operation and inverse Fourier transform on each frequency spectrum, and obtaining a plurality of transformed signals; performing phase estimation based on each transformed signal, and obtaining a phase error; obtaining a phase compensation function based on the phase error; compensating and coherently superimposing each echo signal according to the phase compensation function, and obtaining an image to be identified; inputting the image to be identified into a constructed target identification network for identification, and obtaining a target organism characteristic map, so that personnel can judge the category and state of the target organism according to the target organism characteristic map. The method realizes high-precision identification of target organisms in marine organisms, thereby improving the accuracy of monitoring.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method and system for dynamically monitoring target organisms in a marine ranch. Background Art

[0002] Marine ranching is an ecological aquaculture model based on the principle of marine ecosystem. It builds or repairs places for marine organisms to reproduce, grow, feed or avoid enemies in specific sea areas through measures such as artificial fish reef placement, proliferation and release, and ecological aquaculture, so as to increase and conserve fishery resources and achieve sustainable development of fisheries. With the continuous advancement of science and technology, the construction and management of marine ranches can achieve real-time monitoring and intelligent management of marine ranch environment and biological resources through the use of technologies such as the Internet of Things, big data, and artificial intelligence.

[0003] At present, in the intelligent management of marine ranches, sonar technology is used to detect and monitor organisms in marine ranches by using the propagation and reflection characteristics of sound waves in water. However, traditional imaging algorithms such as RD have problems such as large data volume and high sampling rate, and algorithms such as OMP, SL0 and BCS have low imaging accuracy and poor noise resistance. In the face of complex and changeable underwater environments, unstable platform motion states, and a large proportion of rocky reef organisms in the marine ranch marine biological community, the image signal-to-noise ratio is low due to the obstruction of the reef, the imaging quality is low, the false alarm is high, and the target organisms cannot be accurately identified and monitored, which makes it difficult to judge their abnormal conditions. Summary of the invention

[0004] In order to solve the above technical problems, the embodiments of the present invention provide a method and system for dynamic monitoring of target organisms in a marine ranch, which realizes high-precision identification of target organisms in marine organisms, thereby improving the accuracy of monitoring.

[0005] A first aspect of an embodiment of the present invention provides a method for dynamically monitoring target organisms in a marine ranch, the method comprising:

[0006] Receive echo signals emitted by different sonar array elements in real time, perform distance Fourier transform and azimuth Fourier transform on each echo signal, and obtain the spectrum of each echo signal;

[0007] Performing a phase shift operation on each spectrum to obtain each phase-shifted spectrum, and performing an inverse Fourier transform on each phase-shifted spectrum to obtain a plurality of transformed signals;

[0008] Performing phase estimation based on each transformed signal to obtain a phase error, obtaining a phase compensation function based on the phase error, compensating each echo signal according to the phase compensation function to obtain a compensated echo signal corresponding to each echo signal, performing imaging processing based on each compensated echo signal to obtain multiple sub-images, and coherently superimposing each sub-image to obtain an image to be identified;

[0009] The image to be identified is input into the constructed target recognition network, so that the target recognition network uses the feature extraction module to perform multi-scale feature extraction on the image to be identified, and obtains multiple feature maps. The feature fusion of each feature map is performed through the introduced attention module to obtain the target biological feature map, so that personnel can judge the category and state of the target organism based on the target biological feature map.

[0010] In a possible implementation manner of the first aspect, performing a distance Fourier transform and an azimuth Fourier transform on each echo signal to obtain a frequency spectrum of each echo signal includes:

[0011] Extracting the range information and azimuth information from each echo signal to obtain range data and azimuth data;

[0012] The Fourier transform algorithm is used to perform a distance Fourier transform on the distance data of each echo signal to obtain a distance Fourier transform result, wherein the expression of the distance Fourier transform result is:

[0013]

[0014] In the formula, is the distance data, is the sampling point number in the distance direction, is the distance sampling length, is the distance to Fourier transform result, is the frequency index in the distance direction;

[0015] The Fourier transform algorithm is used to perform a range Fourier transform on the azimuth data of each echo signal to obtain the azimuth Fourier transform result, where the expression of the azimuth Fourier transform result is:

[0016]

[0017] In the formula, is the azimuth data, is the sampling point number in the azimuth direction, is the azimuth sampling length, is the azimuth Fourier transform result, is the frequency index in the azimuth direction;

[0018] According to the distance Fourier transform results and the azimuth Fourier transform results of each echo signal, the distance frequency spectrum and the azimuth frequency spectrum of each echo signal are obtained.

[0019] In a possible implementation manner of the first aspect, performing a phase shift operation on each spectrum to obtain each phase-shifted spectrum includes:

[0020] Extracting the range spectrum and azimuth spectrum of each echo signal to obtain the range phase and azimuth phase of each echo signal;

[0021] According to the preset first phase shift amount, the range phase of each echo signal is adjusted to obtain the range spectrum of each echo signal after phase shift, wherein the adjusted range phase is:

[0022]

[0023] In the formula, is the adjusted range phase, is the first phase shift;

[0024] According to the preset second phase shift amount, the azimuth phase of each echo signal is adjusted to obtain the azimuth spectrum of each echo signal after phase shift, wherein the adjusted azimuth phase is:

[0025]

[0026] In the formula, is the adjusted azimuth phase, is the second phase shift.

[0027] In a possible implementation manner of the first aspect, performing an inverse Fourier transform on each phase-shifted spectrum to obtain a plurality of transformed signals includes:

[0028] The range spectrum of each echo signal after phase shift is subjected to inverse Fourier transform to obtain a plurality of range-transformed signals, wherein the range-transformed signals are:

[0029]

[0030] In the formula, is the distance data, is the sampling point number in the distance direction, is the distance sampling length, is the distance to Fourier transform result, is the frequency index in the distance direction;

[0031] The azimuth spectrum of each echo signal after phase shift is subjected to inverse Fourier transform to obtain a plurality of azimuth transformed signals, wherein the azimuth transformed signals are:

[0032]

[0033] In the formula, is the azimuth data, is the sampling point number in the azimuth direction, is the azimuth sampling length, is the azimuth Fourier transform result, is the frequency index in the azimuth direction;

[0034] The signals after the distance transformation and the signals after the azimuth transformation of each echo signal are combined to obtain the transformed signals of each echo signal.

[0035] In a possible implementation manner of the first aspect, performing phase estimation based on each transformed signal to obtain a phase error includes:

[0036] The autocorrelation function is defined and used to calculate the autocorrelation function of each transformed signal, where the autocorrelation function is:

[0037]

[0038] In the formula, is the autocorrelation function of each transformed signal, is the amplitude of each transformed signal, is the function of the phase of each transformed signal changing with time, is the time-varying parameter;

[0039] The autocorrelation functions under different time-varying parameters are calculated to obtain phase values ​​of multiple autocorrelation functions. The phase values ​​of each autocorrelation function are fitted using the least squares method to obtain the actual phase change rate. The phase error is obtained based on the target phase change rate and the actual phase change rate.

[0040] In a possible implementation of the first aspect, the feature fusion of each different feature map is performed by introducing an attention module to obtain a target biological feature map, including:

[0041] Concatenate and add each feature map to obtain a fused feature map;

[0042] The attention module is used to perform average pooling on the fused image to obtain a compressed feature map. Multiple fully connected layers are used to perform linear transformation on the compressed feature map to obtain multiple transformation results. The softmax function is used to convert the multiple transformation results into probability estimates of each category. After weighted calculation based on each probability estimate and the corresponding feature map, the target biological feature map is obtained.

[0043] In a possible implementation manner of the first aspect, the target recognition network is obtained based on ResNet network training, wherein the training process is:

[0044] Acquire a synthetic aperture sonar image data set, perform energy normalization processing on the synthetic aperture sonar image, and obtain a processed image;

[0045] The processed image is input into the constructed target recognition network, and the feature extraction module is used to extract multi-scale features of the image to be recognized to obtain multiple feature maps, and each feature map is spliced ​​and added to obtain a fused feature map;

[0046] The attention module is used to perform average pooling on the fused image to obtain a compressed feature map. Multiple fully connected layers are used to perform linear transformation on the compressed feature map to obtain multiple transformation results.

[0047] The softmax function is used to convert multiple transformation results into probability estimates of each category, and the output result is obtained after weighted calculation based on each probability estimate and the corresponding feature map;

[0048] According to the output results, the preset loss function is used for back propagation until the loss value obtained according to the loss function is less than the preset value, and the back propagation is stopped to obtain the final network parameters. According to the final network parameters, the trained target recognition network is obtained, where the loss function is:

[0049]

[0050] In the formula, To output the result, is the norm of the network parameter vector of the target identification network, is the attenuation coefficient, is the loss value.

[0051] A second aspect of an embodiment of the present invention provides a target biological dynamic monitoring system for a marine ranch, including a receiving module, a transformation module, a processing module and an identification module.

[0052] The receiving module is used to receive the echo signals emitted by different sonar array elements in real time, perform distance Fourier transform and azimuth Fourier transform on each echo signal, and obtain the spectrum of each echo signal;

[0053] The transformation module is used to perform a phase shift operation on each spectrum to obtain each phase-shifted spectrum, and perform an inverse Fourier transform on each phase-shifted spectrum to obtain a plurality of transformed signals;

[0054] The processing module is used to perform phase estimation based on each transformed signal to obtain a phase error, obtain a phase compensation function based on the phase error, compensate each echo signal according to the phase compensation function to obtain a compensated echo signal corresponding to each echo signal, perform imaging processing according to each compensated echo signal to obtain multiple sub-images, and coherently superimpose each sub-image to obtain an image to be identified;

[0055] The recognition module is used to input the image to be recognized into the constructed target recognition network, so that the target recognition network uses the feature extraction module to perform multi-scale feature extraction on the image to be recognized, obtains multiple feature maps, and performs feature fusion on each feature map through the introduced attention module to obtain the target biological feature map, so that personnel can judge the category and status of the target organism based on the target biological feature map.

[0056] In a possible implementation manner of the second aspect, the receiving module includes an extraction unit, a first transformation unit, a second transformation unit, and a spectrum processing unit.

[0057] The extraction unit is used to extract the range information and azimuth information in each echo signal to obtain the range data and azimuth data;

[0058] The first transformation unit is used to perform a distance Fourier transform on the distance data of each echo signal using a Fourier transform algorithm to obtain a distance Fourier transform result, wherein the expression of the distance Fourier transform result is:

[0059]

[0060] In the formula, is the distance data, is the sampling point number in the distance direction, is the distance sampling length, is the distance to Fourier transform result, is the frequency index in the distance direction;

[0061] The second transformation unit is used to perform a range Fourier transform on the azimuth data of each echo signal using a Fourier transform algorithm to obtain an azimuth Fourier transform result, wherein the expression of the azimuth Fourier transform result is:

[0062]

[0063] In the formula, is the azimuth data, is the sampling point number in the azimuth direction, is the azimuth sampling length, is the azimuth Fourier transform result, is the frequency index in the azimuth direction;

[0064] The spectrum processing unit is used to obtain the range spectrum and the azimuth spectrum of each echo signal according to the range Fourier transform result and the azimuth Fourier transform result of each echo signal.

[0065] In a possible implementation manner of the second aspect, the transformation module includes a phase extraction unit, a first adjustment unit, and a second adjustment unit.

[0066] The phase extraction unit is used to extract the range spectrum and azimuth spectrum of each echo signal to obtain the range phase and azimuth phase of each echo signal;

[0067] The first adjustment unit is used to adjust the range phase of each echo signal according to a preset first phase shift amount to obtain a range spectrum of each echo signal after phase shift, wherein the adjusted range phase is:

[0068]

[0069] In the formula, is the adjusted range phase, is the first phase shift;

[0070] The second adjustment unit is used to adjust the azimuth phase of each echo signal according to a preset second phase shift amount to obtain an azimuth spectrum of each echo signal after phase shift, wherein the adjusted azimuth phase is:

[0071]

[0072] In the formula, is the adjusted azimuth phase, is the second phase shift. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 A monitoring flow diagram of an embodiment of a method for dynamic monitoring of target organisms in a marine ranch provided by the present invention;

[0074] Figure 2 A schematic diagram of a feature fusion process of an embodiment of a method for dynamic monitoring of target organisms in a marine ranch provided by the present invention;

[0075] Figure 3 A system block diagram of an embodiment of a method for dynamically monitoring target organisms in a marine ranch provided by the present invention. DETAILED DESCRIPTION

[0076] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0077] Please refer to Figure 1 , which is a flow chart of an embodiment of a method for dynamically monitoring target organisms in a marine ranch provided by an embodiment of the present invention, including steps S101 to S104, each of which is specifically as follows:

[0078] S101, receiving echo signals emitted by different sonar array elements in real time, performing distance Fourier transform and azimuth Fourier transform on each echo signal, and obtaining a frequency spectrum of each echo signal.

[0079] In this embodiment, first, by setting up a distributed array in the marine ranch, the distributed array can disperse the sonar array elements at different locations in the marine ranch, and these locations can be determined according to the topography of the ranch, the distribution of breeding facilities or the key monitoring areas. This arrangement can flexibly adapt to the complex marine ranch environment and improve the monitoring coverage of the entire ranch area. For example, in a marine ranch with multiple reef areas or artificial reefs, the sonar array elements are arranged around these areas respectively. By reasonably arranging the array elements, the monitoring blind spots can be reduced, and the density and distribution of the array elements can be adjusted according to the importance and characteristics of different areas.

[0080] Then the sonar receiving system is used to receive the echo signals emitted by different sonar array elements in real time. During the receiving process, it is ensured that the receiving system has sufficient bandwidth and sensitivity to accurately capture the echo signals of each array element. At the same time, in order to distinguish the signals of different array elements, the signals of each array element need to be marked or transmitted to the subsequent processing unit separately through multiplexing.

[0081] In some embodiments, in step S101, "performing a distance Fourier transform and an azimuth Fourier transform on each echo signal to obtain a spectrum of each echo signal" includes but is not limited to the following steps:

[0082] Extracting the range information and azimuth information from each echo signal to obtain range data and azimuth data;

[0083] The Fourier transform algorithm is used to perform a distance Fourier transform on the distance data of each echo signal to obtain a distance Fourier transform result, wherein the expression of the distance Fourier transform result is:

[0084]

[0085] In the formula, is the distance data, is the sampling point number in the distance direction, is the distance sampling length, is the distance to Fourier transform result, is the frequency index in the distance direction;

[0086] The Fourier transform algorithm is used to perform a range Fourier transform on the azimuth data of each echo signal to obtain the azimuth Fourier transform result, where the expression of the azimuth Fourier transform result is:

[0087]

[0088] In the formula, is the azimuth data, is the sampling point number in the azimuth direction, is the azimuth sampling length, is the azimuth Fourier transform result, is the frequency index in the azimuth direction;

[0089] According to the distance Fourier transform results and the azimuth Fourier transform results of each echo signal, the distance frequency spectrum and the azimuth frequency spectrum of each echo signal are obtained.

[0090] In this embodiment, the range information and the azimuth information in each echo signal are extracted to obtain the range data and the azimuth data;

[0091] The Fourier transform algorithm is used to perform a distance Fourier transform on the distance data of each echo signal to obtain a distance Fourier transform result, wherein the expression of the distance Fourier transform result is:

[0092]

[0093] In the formula, is the distance data, is the sampling point number in the distance direction, is the distance sampling length, is the distance to Fourier transform result, is the frequency index in the distance direction;

[0094] The Fourier transform algorithm is used to perform a range Fourier transform on the azimuth data of each echo signal to obtain the azimuth Fourier transform result, where the expression of the azimuth Fourier transform result is:

[0095]

[0096] In the formula, is the azimuth data, is the sampling point number in the azimuth direction, is the azimuth sampling length, is the azimuth Fourier transform result, is the frequency index in the azimuth direction;

[0097] According to the distance Fourier transform results and the azimuth Fourier transform results of each echo signal, the distance frequency spectrum and the azimuth frequency spectrum of each echo signal are obtained.

[0098] In this embodiment, the signal after pulse compression is subjected to FFT (Fast Fourier Transform) to convert the signal from the time domain to the frequency domain. In the frequency domain, the distance information of the target is extracted according to the corresponding relationship between frequency and distance. By coherently processing the echo signals received at different positions, an equivalent large aperture antenna is synthesized to improve the azimuth resolution, thereby extracting the azimuth information of the target.

[0099] The Fourier transform algorithm is used to perform a distance Fourier transform on the extracted distance data to obtain the distance Fourier transform result, where the expression of the distance Fourier transform result is:

[0100]

[0101] In the formula, is the distance data, is the sampling point number in the distance direction, is the distance sampling length, is the distance to Fourier transform result, is the frequency index in the distance direction;

[0102] The azimuth data of the Fourier transform algorithm is used to perform a distance Fourier transform to obtain an azimuth Fourier transform result, wherein the expression of the azimuth Fourier transform result is:

[0103]

[0104] In the formula, is the azimuth data, is the sampling point number in the azimuth direction, is the azimuth sampling length, is the azimuth Fourier transform result, is the frequency index in the azimuth direction;

[0105] The distance Fourier transform result and the azimuth Fourier transform result are the distance spectrum and azimuth spectrum of each echo signal.

[0106] S102 , performing a phase shift operation on each frequency spectrum to obtain each phase-shifted frequency spectrum, and performing an inverse Fourier transform on each phase-shifted frequency spectrum to obtain a plurality of transformed signals.

[0107] In this embodiment, by performing phase shift operation on the sonar signal spectrum of different receiving array elements, accurate estimation of the target azimuth can be achieved. For example, when there is a noise source emitted by an underwater target, there are differences in the time when the signal reaches different array elements, which will result in different signal phases. By performing phase shift operation on the signal spectrum of each array element, it is superimposed in phase in a specific direction, so that the direction of the target can be determined. After performing phase shift operation on each spectrum, each phase-shifted spectrum is inverse Fourier transformed to obtain multiple transformed signals. The distribution of the echo signal in time can be obtained by inverse Fourier transform. Then, the distance between the target and the radar and the motion state of the target are analyzed. Specifically, the distance between the target and the radar is calculated by the echo time, and the change in the echo time is used to determine whether the target is approaching or moving away.

[0108] In some embodiments, the step S102 of “performing a phase shift operation on each spectrum to obtain each phase-shifted spectrum” includes but is not limited to the following steps:

[0109] Extracting the range spectrum and azimuth spectrum of each echo signal to obtain the range phase and azimuth phase of each echo signal;

[0110] According to the preset first phase shift amount, the range phase of each echo signal is adjusted to obtain the range spectrum of each echo signal after phase shift, wherein the adjusted range phase is:

[0111]

[0112] In the formula, is the adjusted range phase, is the first phase shift;

[0113] According to the preset second phase shift amount, the azimuth phase of each echo signal is adjusted to obtain the azimuth spectrum of each echo signal after phase shift, wherein the adjusted azimuth phase is:

[0114]

[0115] In the formula, is the adjusted azimuth phase, is the second phase shift.

[0116] In this embodiment, the range spectrum and azimuth spectrum of each echo signal are extracted to obtain the range phase and azimuth phase of each echo signal, and the range phase in each echo signal is adjusted to obtain the range spectrum of each echo signal after phase shift, wherein the adjusted range phase is:

[0117]

[0118] In the formula, is the adjusted range phase, is the phase shift in the distance direction.

[0119] The azimuth phase of each echo signal is adjusted to obtain the azimuth spectrum of each echo signal after phase shift, wherein the adjusted azimuth phase is:

[0120]

[0121] In the formula, is the adjusted azimuth phase, is the phase shift in the azimuth direction.

[0122] In some embodiments, the step S102 of “performing an inverse Fourier transform on each of the phase-shifted spectra to obtain a plurality of transformed signals” includes but is not limited to the following steps:

[0123] The range spectrum of each echo signal after phase shift is subjected to inverse Fourier transform to obtain a plurality of range-transformed signals, wherein the range-transformed signals are:

[0124]

[0125] In the formula, is the distance data, is the sampling point number in the distance direction, is the distance sampling length, is the distance to Fourier transform result, is the frequency index in the distance direction;

[0126] The azimuth spectrum of each echo signal after phase shift is subjected to inverse Fourier transform to obtain a plurality of azimuth transformed signals, wherein the azimuth transformed signals are:

[0127]

[0128] In the formula, is the azimuth data, is the sampling point number in the azimuth direction, is the azimuth sampling length, is the azimuth Fourier transform result, is the frequency index in the azimuth direction;

[0129] The signals after the distance transformation and the signals after the azimuth transformation of each echo signal are combined to obtain the transformed signals of each echo signal.

[0130] In this embodiment, after the phase-shifted spectrum data is obtained, since the spectrum data is usually in complex form and contains amplitude and phase information, for example, in digital signal processing, the spectrum data may be obtained by fast Fourier transform (FFT), and after the phase shift operation, it is prepared for inverse Fourier transform.

[0131] The range spectrum of each echo signal after phase shift is subjected to inverse Fourier transform to obtain a plurality of range-transformed signals, wherein the range-transformed signals are:

[0132]

[0133] In the formula, is the distance data, is the sampling point number in the distance direction, is the distance sampling length, is the distance to Fourier transform result, is the frequency index in the distance direction;

[0134] The azimuth spectrum of each echo signal after phase shift is subjected to inverse Fourier transform to obtain a plurality of azimuth transformed signals, wherein the azimuth transformed signals are:

[0135]

[0136] In the formula, is the azimuth data, is the sampling point number in the azimuth direction, is the azimuth sampling length, is the azimuth Fourier transform result, is the frequency index in the azimuth direction;

[0137] The signals after the distance transformation and the signals after the azimuth transformation of each echo signal are combined to obtain the transformed signals of each echo signal.

[0138] S103, performing phase estimation based on each transformed signal to obtain a phase error, obtaining a phase compensation function based on the phase error, compensating each echo signal according to the phase compensation function to obtain a compensated echo signal corresponding to each echo signal, performing imaging processing based on each compensated echo signal to obtain multiple sub-images, coherently superimposing each sub-image to obtain an image to be identified.

[0139] In this embodiment, after the signal emitted by the array sonar is received, the round-trip distance of the signal will cause a phase change. The phase error will affect the accuracy of the distance calculated based on the phase change. After the phase error is corrected, the position of the target organism can be calculated more accurately. Therefore, it is necessary to perform phase estimation based on the transformed signal to estimate the phase error. Then, a phase compensation function is obtained based on the phase error. After each echo signal is compensated according to the phase compensation function, imaging processing is performed using the compensated echo signal to obtain a sub-image obtained by each sonar array element. After each sub-image is coherently superimposed, a SAR image is finally obtained, and the SAR image is then input into the constructed target recognition network for recognition.

[0140] In some embodiments, the step S103 of “performing phase estimation based on each of the transformed signals to obtain a phase error” includes but is not limited to the following steps:

[0141] The autocorrelation function is defined and used to calculate the autocorrelation function of each transformed signal, where the autocorrelation function is:

[0142]

[0143] In the formula, is the autocorrelation function of each transformed signal, is the amplitude of each transformed signal, is the function of the phase of each transformed signal changing with time, is the time-varying parameter;

[0144] The autocorrelation functions under different time-varying parameters are calculated to obtain phase values ​​of multiple autocorrelation functions. The phase values ​​of each autocorrelation function are fitted using the least squares method to obtain the actual phase change rate. The phase error is obtained based on the target phase change rate and the actual phase change rate.

[0145] In this embodiment, since the signal collected by the synthetic aperture radar is a continuous signal, the autocorrelation function of each transformed signal is calculated according to the definition of the autocorrelation function, where the autocorrelation function is:

[0146]

[0147] In the formula, is the autocorrelation function of each transformed signal, is the amplitude of each transformed signal, is the function of the phase of each transformed signal changing with time, is the time-varying parameter;

[0148] Then, the autocorrelation functions under different time-varying parameters are calculated to obtain the phase values ​​of multiple autocorrelation functions. The phase values ​​of each autocorrelation function are then fitted using the least squares method to obtain the actual phase change rate. The phase error is obtained based on the target phase change rate and the actual phase change rate.

[0149] After obtaining the phase error, the statistical information of the phase error changing with time can be seen, so the phase compensation function can be set as:

[0150]

[0151] In the formula, is the amplitude of each transformed signal, is the function of the phase of each transformed signal changing with time, is a time-varying parameter.

[0152] S104. Input the image to be identified into the constructed target recognition network, so that the target recognition network uses the feature extraction module to perform multi-scale feature extraction on the image to be identified, and obtains multiple feature maps. The feature maps are then fused by the introduced attention module to obtain a target biological feature map, so that personnel can judge the category and state of the target organism based on the target biological feature map.

[0153] In this embodiment, target recognition is an important process for interpreting SAR images. There is a large amount of coherent speckle noise in SAR images, which is caused by the coherent characteristics of radar waves and the scattering characteristics of the target surface. Coherent speckle noise can cause the gray value of the image to fluctuate, reduce the contrast and resolution of the image, blur the edges and details of the target, and increase the difficulty of target recognition. In addition, the imaging principle of SAR images leads to geometric distortion, such as perspective contraction, overlap and shadow. These geometric distortions can cause the shape, size and position of the target to deform in the image, affecting the accurate recognition and positioning of the target. For example, in areas with dense reefs, perspective contraction and overlap will be very obvious, which may cause the target to be blocked or distorted in shape. CNN (convolutional neural network) can automatically extract features from SAR images, avoiding the tediousness and subjectivity of manual feature extraction. By using the feature extraction module to perform multi-scale feature extraction on the SAR image, multiple feature maps are obtained. The multi-scale residual feature extraction module proposed in this paper includes convolution layers with convolution kernel sizes of 1x1, 3x3, 5x5, and 7x7 to extract features at different scales. After obtaining the feature maps at each scale, the feature maps are fused through the introduced attention module to obtain the target biological feature map, so that personnel can judge the category and state of the target organism based on the target biological feature map.

[0154] It should be noted that SAR images refer to images to be identified.

[0155] In some embodiments, in step S104, “by introducing an attention module to fuse the features of the different feature maps to obtain a target biological feature map” includes but is not limited to the following steps:

[0156] Concatenate and add each feature map to obtain a fused feature map;

[0157] The attention module is used to perform average pooling on the fused image to obtain a compressed feature map. Multiple fully connected layers are used to perform linear transformation on the compressed feature map to obtain multiple transformation results. The softmax function is used to convert the multiple transformation results into probability estimates of each category. After weighted calculation based on each probability estimate and the corresponding feature map, the target biological feature map is obtained.

[0158] In this embodiment, each feature map is first concatenated and added to obtain a fused feature map, and the fused image is compressed using an attention module, that is, average pooling is performed to obtain a compressed feature map. The compressed feature map is linearly transformed using multiple fully connected layers to obtain multiple transformation results. The softmax function is used to convert the multiple transformation results into probability estimates of each category to obtain weights, and then the weights are weightedly calculated with the corresponding feature maps to obtain the target biological feature map M.

[0159] In some embodiments, in step S104, “the target recognition network is obtained based on ResNet network training” includes but is not limited to the following steps:

[0160] Acquire a synthetic aperture sonar image data set, perform energy normalization processing on the synthetic aperture sonar image, and obtain a processed image;

[0161] The processed image is input into the constructed target recognition network, and the feature extraction module is used to extract multi-scale features of the image to be recognized to obtain multiple feature maps, and each feature map is spliced ​​and added to obtain a fused feature map;

[0162] The attention module is used to perform average pooling on the fused image to obtain a compressed feature map. Multiple fully connected layers are used to perform linear transformation on the compressed feature map to obtain multiple transformation results.

[0163] The softmax function is used to convert multiple transformation results into probability estimates of each category, and the output result is obtained after weighted calculation based on each probability estimate and the corresponding feature map;

[0164] According to the output results, the preset loss function is used for back propagation until the loss value obtained according to the loss function is less than the preset value, and the back propagation is stopped to obtain the final network parameters. According to the final network parameters, the trained target recognition network is obtained, where the loss function is:

[0165]

[0166] In the formula, To output the result, is the norm of the network parameter vector of the target identification network, is the attenuation coefficient, is the loss value.

[0167] In this embodiment, some data from a comprehensive ocean database is used for the experiment, such as some large ocean databases established by relevant international ocean research organizations, which contain SAR images of some marine fish. The input images are energy normalized, and the normalized images are input into the constructed target recognition network. The batch size is set to 32, and the SGD optimizer is used for training. The initial learning rate is 0.001, and the learning rate is halved every 50 epochs. The loss function used is:

[0168]

[0169] In the formula, To output the result, is the norm of the network parameter vector of the target identification network, is the attenuation coefficient, is the loss value.

[0170] During training, a feature extraction module is used to perform multi-scale feature extraction on the image to be identified to obtain multiple feature maps. The feature maps are first concatenated and added to obtain a fused feature map. The fused image is compressed by an attention module, that is, average pooling is performed to obtain a compressed feature map. The compressed feature map is linearly transformed using multiple fully connected layers to obtain multiple transformation results. The softmax function is used to convert the multiple transformation results into probability estimates of each category to obtain weights. The weights are then weightedly calculated with the corresponding feature maps to obtain output results. According to the output results, a preset loss function is used for back propagation until the loss value obtained according to the loss function is less than the preset value. The back propagation is stopped to obtain the final network parameters, and the trained target recognition network is obtained according to the final network parameters.

[0171] In some embodiments, Figure 3As shown, it shows a block diagram of a target biological dynamic monitoring system 300 of a marine ranch provided in an embodiment of the present application, including: a receiving module 301, a transformation module 302, a processing module 303 and an identification module 304,

[0172] The receiving module 301 is used to receive the echo signals emitted by different sonar array elements in real time, perform distance Fourier transform and azimuth Fourier transform on each echo signal, and obtain the spectrum of each echo signal;

[0173] The transformation module 302 is used to perform a phase shift operation on each spectrum to obtain each phase-shifted spectrum, and perform an inverse Fourier transform on each phase-shifted spectrum to obtain a plurality of transformed signals;

[0174] The processing module 303 is used to perform phase estimation based on each transformed signal to obtain a phase error, obtain a phase compensation function based on the phase error, compensate each echo signal according to the phase compensation function to obtain a compensated echo signal corresponding to each echo signal, perform imaging processing according to each compensated echo signal to obtain multiple sub-images, and coherently superimpose each sub-image to obtain an image to be identified;

[0175] The recognition module 304 is used to input the image to be recognized into the constructed target recognition network, so that the target recognition network uses the feature extraction module to perform multi-scale feature extraction on the image to be recognized, obtains multiple feature maps, and performs feature fusion on each feature map through the introduced attention module to obtain a target biological feature map, so that personnel can judge the category and status of the target organism based on the target biological feature map.

[0176] The specific implementation of the target organism dynamics monitoring system for marine ranching is basically the same as the specific implementation of the target organism dynamics monitoring method for marine ranching mentioned above, and will not be repeated here.

[0177] Compared with the prior art, the present invention has the following beneficial effects:

[0178] The present invention receives echo signals emitted by different sonar array elements in real time, performs distance Fourier transform and azimuth Fourier transform on each echo signal to obtain the spectrum of each echo signal; performs phase shift operation on each spectrum to obtain each phase-shifted spectrum, performs inverse Fourier transform on each phase-shifted spectrum to obtain multiple transformed signals; performs phase estimation based on each transformed signal to obtain a phase error, obtains a phase compensation function based on the phase error, compensates each echo signal according to the phase compensation function to obtain a compensated echo signal corresponding to each echo signal, performs imaging processing according to each compensated echo signal to obtain multiple sub-images, performs coherent superposition on each sub-image to obtain an image to be identified; inputs the image to be identified into a constructed target recognition network, so that the target recognition network uses a feature extraction module to perform multi-scale feature extraction on the image to be identified to obtain multiple feature maps, and performs feature fusion on each feature map through an introduced attention module to obtain a target biological feature map, so that personnel can judge the category and state of the target biological according to the target biological feature map. The method realizes high-precision recognition of the target biological in the marine organisms, thereby improving the accuracy of monitoring.

[0179] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0180] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for dynamic monitoring of target organisms in marine ranches, characterized in that: include: Receiving echo signals emitted by different sonar array elements in real time, performing distance Fourier transform and azimuth Fourier transform on each echo signal, and obtaining the frequency spectrum of each echo signal; Performing a phase shift operation on each of the frequency spectra to obtain each phase-shifted frequency spectra, and performing an inverse Fourier transform on each of the phase-shifted frequency spectra to obtain a plurality of transformed signals; Performing phase estimation based on each of the transformed signals to obtain a phase error, obtaining a phase compensation function based on the phase error, compensating each of the echo signals according to the phase compensation function to obtain a compensated echo signal corresponding to each of the echo signals, performing imaging processing based on each of the compensated echo signals to obtain a plurality of sub-images, and coherently superimposing each of the sub-images to obtain an image to be identified; Input the image to be identified into the constructed target recognition network, so that the target recognition network uses the feature extraction module to perform multi-scale feature extraction on the image to be identified, obtains multiple feature maps, and performs feature fusion on each of the feature maps through the introduced attention module to obtain a target biological feature map, so that personnel can judge the category and state of the target organism according to the target biological feature map; The performing of distance Fourier transform and azimuth Fourier transform on each of the echo signals to obtain a frequency spectrum of each of the echo signals comprises: Extracting the range information and the azimuth information from each of the echo signals to obtain range data and azimuth data; The distance data of each echo signal is subjected to a distance Fourier transform using a Fourier transform algorithm to obtain a distance Fourier transform result, wherein the expression of the distance Fourier transform result is: ； In the formula, is the distance data, is the sampling point number in the distance direction, is the distance sampling length, is the distance to Fourier transform result, is the frequency index in the distance direction; The azimuth data of each echo signal is subjected to a range Fourier transform using a Fourier transform algorithm to obtain an azimuth Fourier transform result, wherein the expression of the azimuth Fourier transform result is: ； In the formula, is the azimuth data, is the sampling point number in the azimuth direction, is the azimuth sampling length, is the azimuth Fourier transform result, is the frequency index in the azimuth direction; Obtaining a frequency spectrum in the range and a frequency spectrum in the azimuth of each echo signal according to a Fourier transform result in the range and a Fourier transform result in the azimuth of each echo signal; The performing a phase shift operation on each of the frequency spectra to obtain each phase-shifted frequency spectra includes: Extracting the range spectrum and the azimuth spectrum of each echo signal to obtain the range phase and the azimuth phase of each echo signal; According to the preset first phase shift amount, the range phase of each echo signal is adjusted to obtain the range spectrum of each echo signal after phase shift, wherein the adjusted range phase is: ； In the formula, is the adjusted range phase, is the first phase shift; According to the preset second phase shift amount, the azimuth phase of each echo signal is adjusted to obtain the azimuth spectrum of each echo signal after phase shift, wherein the adjusted azimuth phase is: ； In the formula, is the adjusted azimuth phase, is the second phase shift; The performing phase estimation based on each of the transformed signals to obtain a phase error comprises: The autocorrelation function of each of the transformed signals is calculated using the autocorrelation function definition, wherein the autocorrelation function is: ； In the formula, is the autocorrelation function of each of the transformed signals, is the amplitude of each of the transformed signals, is a function of the phase of each transformed signal changing with time, is the time-varying parameter; The autocorrelation functions under different time-varying parameters are calculated to obtain phase values ​​of multiple autocorrelation functions, and the phase values ​​of each autocorrelation function are fitted using the least squares method to obtain the actual phase change rate, and the phase error is obtained according to the target phase change rate and the actual phase change rate.

2. The target organism dynamic monitoring method of marine ranch as claimed in claim 1, characterized in that: The inverse Fourier transform is performed on each of the phase-shifted spectra to obtain a plurality of transformed signals, including: Perform inverse Fourier transform on the range spectrum of each phase-shifted echo signal to obtain a plurality of range-transformed signals, wherein the range-transformed signals are: ； In the formula, is the distance data, is the sampling point number in the distance direction, is the distance sampling length, is the inverse Fourier transform result of the distance. is the frequency index in the distance direction; Performing inverse Fourier transform on the azimuth spectrum of each of the phase-shifted echo signals to obtain a plurality of azimuth-transformed signals, wherein the azimuth-transformed signals are: ； In the formula, is the azimuth data, is the sampling point number in the azimuth direction, is the azimuth sampling length, is the result of the inverse Fourier transform in azimuth direction, is the frequency index in the azimuth direction; The signals after the distance transformation and the signals after the azimuth transformation of each of the echo signals are combined to obtain the signals after the transformation of each of the echo signals.

3. The target organism dynamic monitoring method of marine ranch as claimed in claim 1, characterized in that: The introduced attention module performs feature fusion on each of the different feature maps to obtain a target biological feature map, including: Concatenate and add the feature maps to obtain a fused feature map; An attention module is used to perform average pooling on the fused image to obtain a compressed feature map. Multiple fully connected layers are used to perform linear transformation on the compressed feature map to obtain multiple transformation results. The softmax function is used to convert the multiple transformation results into probability estimates of each category. After weighted calculation is performed on each probability estimate and the corresponding feature map, the target biological feature map is obtained.

4. The target organism dynamic monitoring method of marine ranch as claimed in claim 1, characterized in that: The target recognition network is obtained based on ResNet network training, wherein the training process is: Acquire a synthetic aperture sonar image data set, and perform energy normalization processing on the synthetic aperture sonar image to obtain a processed image; Input the processed image into the constructed target recognition network, use the feature extraction module to extract multi-scale features of the image to be recognized, obtain multiple feature maps, and splice and add the feature maps to obtain a fused feature map; An attention module is used to perform average pooling on the fused image to obtain a compressed feature map, and multiple fully connected layers are used to perform linear transformation on the compressed feature map to obtain multiple transformation results; The softmax function is used to convert the multiple transformation results into probability estimates of each category, and the output result is obtained after weighted calculation based on each probability estimate and the corresponding feature map; According to the output result, a preset loss function is used for back propagation until the loss value obtained according to the loss function is less than the preset value, and the back propagation is stopped to obtain the final network parameters. A trained target recognition network is obtained according to the final network parameters, wherein the loss function is: ； In the formula, To output the result, is the norm of the network parameter vector of the target identification network, is the attenuation coefficient, is the loss value.

5. A target biological dynamic monitoring system for marine ranches, characterized in that: It includes a receiving module, a transforming module, a processing module and an identifying module. The receiving module is used to receive echo signals emitted by different sonar array elements in real time, perform distance Fourier transform and azimuth Fourier transform on each echo signal, and obtain the frequency spectrum of each echo signal; The transformation module is used to perform a phase shift operation on each of the frequency spectra to obtain each phase-shifted frequency spectra, and perform an inverse Fourier transform on each of the phase-shifted frequency spectra to obtain a plurality of transformed signals; The processing module is used to perform phase estimation based on each of the transformed signals to obtain a phase error, obtain a phase compensation function based on the phase error, compensate each of the echo signals according to the phase compensation function to obtain a compensated echo signal corresponding to each of the echo signals, perform imaging processing according to each of the compensated echo signals to obtain a plurality of sub-images, and coherently superimpose each of the sub-images to obtain an image to be identified; The recognition module is used to input the image to be recognized into the constructed target recognition network, so that the target recognition network uses the feature extraction module to perform multi-scale feature extraction on the image to be recognized, obtain multiple feature maps, and perform feature fusion on each of the feature maps through the introduced attention module to obtain a target biological feature map, so that personnel can judge the category and state of the target organism according to the target biological feature map; The receiving module includes an extraction unit, a first transformation unit, a second transformation unit and a spectrum processing unit. The extraction unit is used to extract the range information and the azimuth information from each of the echo signals to obtain the range data and the azimuth data; The first transformation unit is used to perform a distance Fourier transform on the distance data of each echo signal using a Fourier transform algorithm to obtain a distance Fourier transform result, wherein the expression of the distance Fourier transform result is: ； In the formula, is the distance data, is the sampling point number in the distance direction, is the distance sampling length, is the distance to Fourier transform result, is the frequency index in the distance direction; The second transformation unit is used to perform a range Fourier transform on the azimuth data of each echo signal using a Fourier transform algorithm to obtain an azimuth Fourier transform result, wherein the expression of the azimuth Fourier transform result is: ； In the formula, is the azimuth data, is the sampling point number in the azimuth direction, is the azimuth sampling length, is the azimuth Fourier transform result, is the frequency index in the azimuth direction; The spectrum processing unit is used to obtain the range spectrum and the azimuth spectrum of each echo signal according to the range Fourier transform result and the azimuth Fourier transform result of each echo signal; The transformation module includes a phase extraction unit, a first adjustment unit and a second adjustment unit. The phase extraction unit is used to extract the range spectrum and azimuth spectrum of each echo signal to obtain the range phase and azimuth phase of each echo signal; The first adjustment unit is used to adjust the range phase of each of the echo signals according to a preset first phase shift amount to obtain a range spectrum of each of the echo signals after phase shift, wherein the adjusted range phase is: ； In the formula, is the adjusted range phase, is the first phase shift; The second adjustment unit is used to adjust the azimuth phase of each of the echo signals according to a preset second phase shift amount to obtain an azimuth spectrum of each of the echo signals after phase shift, wherein the adjusted azimuth phase is: ； In the formula, is the adjusted azimuth phase, is the second phase shift; The performing phase estimation based on each of the transformed signals to obtain a phase error comprises: The autocorrelation function of each of the transformed signals is calculated using the autocorrelation function definition, wherein the autocorrelation function is: ； In the formula, is the autocorrelation function of each of the transformed signals, is the amplitude of each of the transformed signals, is a function of the phase of each transformed signal changing with time, is the time-varying parameter; The autocorrelation functions under different time-varying parameters are calculated to obtain phase values ​​of multiple autocorrelation functions, and the phase values ​​of each autocorrelation function are fitted using the least squares method to obtain the actual phase change rate, and the phase error is obtained according to the target phase change rate and the actual phase change rate.

Images