A method, system, device and medium for detecting underwater cultural relics based on hyperspectral

Abstract

The present invention discloses a method, system, electronic device and storage medium for detecting underwater cultural relics based on hyperspectral. The scheme acquires underwater environmental data and enhanced hyperspectral data with a detection area, constructs an original data set according to the underwater environmental data and enhanced hyperspectral data, performs calculations according to the original data set and a trained recognition network model, and obtains a predicted probability value set of the area to be detected; determines whether the area to be detected is an area with cultural relics according to the calculated predicted probability value set and a preset threshold. The seabed is detected using the spectrum, and the reflection of the spectrum by seabed materials and cultural relics is detected, without the damage of acoustic vibration to the cultural relics, and the influence of the seabed environment on the spectrum transmission is considered at the same time, so as to realize noise-free pollution detection of the seabed; the embodiments of the present invention can be widely used in the field of data processing technology.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method, system, electronic equipment and storage medium for detecting underwater cultural relics based on hyperspectral. Background Art

[0002] Undersea cultural relics are usually covered by silt or sandy sediments, and it is necessary to detect areas where cultural relics may exist in order to determine the areas where cultural relics exist. Existing technologies usually use geophysical detection methods, such as multi-beam detection of the seabed. Since multi-beam detection uses sound wave detection, the vibration of sound waves can cause damage to cultural relics and produce ocean noise pollution. Summary of the invention

[0003] The main purpose of the embodiments of the present invention is to provide a method, system, electronic device and storage medium for detecting underwater cultural relics based on hyperspectral, which can reduce damage to cultural relics and reduce underwater noise pollution.

[0004] To achieve the above object, an embodiment of the present invention provides a method for detecting underwater cultural relics based on hyperspectral, the method comprising:

[0005] Acquire seabed environmental data and enhanced hyperspectral data of the area to be detected;

[0006] Constructing an original data set according to the seabed environment data and the enhanced hyperspectral data, and performing calculations according to the original data set and the trained recognition network model to obtain a probability value set;

[0007] The area with cultural relics is determined according to the probability value set and a preset threshold.

[0008] In some embodiments, the enhanced hyperspectral data is obtained by processing a dynamically updated filter, and the filter is dynamically updated by the following method:

[0009] Acquire current sensor data according to a preset cycle; wherein the current sensor data includes any one or more of light intensity data, water depth data, suspended particle concentration data, temperature data, salinity data or pH data;

[0010] Calculating according to a preset linear model and the current sensor data to obtain target parameters of the filter; wherein the target parameters include the bandwidth and center frequency of the filter;

[0011] The filter is updated according to the target parameter.

[0012] In some embodiments, the calculation based on the original data set and the trained recognition network model to obtain the probability value set specifically includes:

[0013] Constructing first dynamic graph data according to the original data set; wherein the first dynamic graph data includes node features and a first adjacency matrix;

[0014] updating node features of the first dynamic graph data according to the recognition network model to obtain a node feature set; calculating a normalization coefficient set according to the node feature set; calculating according to the node feature set and a first preset function to determine a matrix parameter adjustment value set, and updating the first adjacency matrix according to the matrix parameter adjustment value set to obtain a second adjacency matrix;

[0015] Updating the first dynamic graph data according to the node feature set, the normalization coefficient set and the second adjacency matrix, and comparing the current update number with a preset number;

[0016] If the current update number is less than the preset number, returning to execute the node feature update of the first dynamic graph data according to the trained recognition network model until the current update number is equal to the preset number;

[0017] If the current update number is equal to the preset number, full connection processing is performed according to the recognition network model and the first dynamic graph data updated last time to obtain the probability value set.

[0018] In some embodiments, updating the node features of the first dynamic graph data according to the recognition network model to obtain a node feature set specifically includes:

[0019] Determine the current node characteristics and the neighbor node characteristic set respectively according to the first dynamic graph data; wherein the node characteristics include spectral parameters and ocean environment parameters;

[0020] Performing a first nonlinear calculation based on the recognition network model and the current node feature to obtain a first parameter; performing a calculation based on the recognition network model and the neighbor node feature set to obtain a second parameter;

[0021] The current node feature is updated according to the first parameter and the second parameter to obtain the node feature set.

[0022] In some embodiments, the calculating of the normalization coefficient set according to the node feature set specifically includes:

[0023] Determine the current node feature and the neighbor node feature set respectively according to the node feature set; wherein the neighbor node feature set includes feature vectors of several nodes connected to the current node;

[0024] Calculating according to the identification network model and the current node feature to obtain a first vector; calculating according to the identification network model and the neighbor node feature set to obtain a second vector set;

[0025] sequentially connect the second vector in the second vector set with the first vector to obtain a third vector set; perform a second nonlinear calculation based on the third vector set, the attention parameter vector and the recognition network model to obtain a non-normalized attention coefficient set;

[0026] The non-normalized attention coefficient set is normalized according to a preset function to obtain the normalized coefficient set.

[0027] In some embodiments, the first preset function is:

[0028] ,

[0029] in, Adjust the values ​​for the matrix parameters, is the feature vector of node i at time t, is the feature vector of node j at time t, is the norm of the feature vector of node i at time t, is the norm of the feature vector of node j at time t.

[0030] In some embodiments, determining the area with cultural relics according to the probability value set and a preset threshold value specifically includes:

[0031] Comparing the maximum probability value in the probability value set with the preset threshold;

[0032] If the maximum probability value is less than the preset threshold, the area to be detected is marked as an area without cultural relics;

[0033] If the maximum probability value is greater than the preset threshold, the area to be detected is marked as the area with cultural relics.

[0034] To achieve the above object, another aspect of an embodiment of the present invention provides a seabed cultural relic detection system based on hyperspectral, comprising:

[0035] The first module is used to obtain seabed environmental data and enhanced hyperspectral data of the area to be detected;

[0036] The second module is used to construct an original data set according to the seabed environment data and the enhanced hyperspectral data, and to perform calculations according to the original data set and the trained recognition network model to obtain a probability value set;

[0037] The third module is used to determine the area with cultural relics according to the probability value set and a preset threshold.

[0038] To achieve the above-mentioned purpose, another aspect of an embodiment of the present application provides an electronic device, the electronic device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the above-mentioned method when executing the computer program.

[0039] To achieve the above objective, another aspect of an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described above is implemented.

[0040] The implementation of the embodiment of the present invention includes the following beneficial effects: This embodiment provides a method, system, electronic device and storage medium for detecting underwater cultural relics based on hyperspectral. The scheme acquires underwater environmental data and enhanced hyperspectral data with a detection area, constructs an original data set based on the underwater environmental data and enhanced hyperspectral data, calculates based on the original data set and the trained recognition network model, and obtains a predicted probability value set of the area to be detected; determines whether the area to be detected is an area with cultural relics based on the calculated predicted probability value set and a preset threshold. Detection using the spectrum reduces the damage to cultural relics caused by acoustic vibrations, and noise-free pollution detection of the seabed is achieved by reflecting the spectrum of the seabed material and considering the influence of the seabed environment on the spectrum transmission. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic flow chart of the steps of a method for detecting underwater cultural relics based on hyperspectral provided by an embodiment of the present invention;

[0042] Figure 2 Schematic diagram of an application scenario of a method for detecting underwater cultural relics based on hyperspectral spectrum provided by an embodiment of the present invention;

[0043] Figure 3 It is a schematic flow chart of the steps of adjusting dynamic parameters of a filter in a method for detecting underwater cultural relics based on hyperspectral spectrum provided by an embodiment of the present invention;

[0044] Figure 4 It is a schematic flow chart of the steps of obtaining a probability value set in a method for detecting underwater cultural relics based on hyperspectral provided by an embodiment of the present invention;

[0045] Figure 5 It is a schematic flow chart of steps for obtaining a node feature set in a method for detecting underwater cultural relics based on hyperspectral spectrum provided by an embodiment of the present invention;

[0046] Figure 6 It is a schematic flow chart of steps for obtaining a normalized coefficient set in a method for detecting underwater cultural relics based on hyperspectral spectrum provided by an embodiment of the present invention;

[0047] Figure 7 It is a schematic flow chart of the steps of determining an area with cultural relics in a method for detecting underwater cultural relics based on hyperspectral spectrum provided by an embodiment of the present invention;

[0048] Figure 8 It is a schematic flow chart of the steps of training and obtaining a recognition network model in a method for detecting underwater cultural relics based on hyperspectral spectrum provided by an embodiment of the present invention;

[0049] Fig. 9 It is a schematic diagram of a step flow of a specific embodiment provided by an embodiment of the present invention;

[0050] Fig.10 It is a structural block diagram of a seabed cultural relic detection system based on hyperspectral provided by an embodiment of the present invention;

[0051] Fig.11 is a schematic diagram of the hardware structure of an electronic device provided by an embodiment of the present invention;

[0052] Among them, 1 is an underwater hyperspectral device, 2 is an underwater cultural relic, 3 is muddy sediment, 4 is sandy sediment, and 5 is seawater. DETAILED DESCRIPTION

[0053] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only provided for the convenience of explanation and description, and the order between the steps is not limited in any way. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0054] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0055] In the following description, the terms "first\second\third" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present invention described herein can be implemented in an order other than that illustrated or described herein.

[0056] Unless otherwise defined, all technical and scientific terms used in the embodiments of the present invention have the same meanings as those commonly understood by those skilled in the art of the present invention. The terms used in the embodiments of the present invention are only for the purpose of describing the embodiments of the present invention and are not intended to limit the present invention.

[0057] In the marine environment, cultural relics will be covered by silt or sandy sediments on the seabed, and it is necessary to detect the seabed area to determine whether there are cultural relics in the seabed area. In the existing technology, geophysical detection methods are usually used for detection, such as gravity detection, magnetic detection, etc. In the field of seabed cultural relic detection, multi-beam detection is usually used to detect the seabed through sonar. Since sonar uses ultrasonic waves for detection, it will cause damage to the underwater cultural relics. At the same time, the sediments on the seabed surface diffusely reflect the ultrasonic waves emitted by the sonar, which will cause ocean noise pollution and further affect the seabed environment.

[0058] In view of this, the embodiments of the present application provide a method, system, electronic device and storage medium for detecting underwater cultural relics based on hyperspectral spectrum. The scheme transmits a detection beam to the seabed through an underwater hyperspectral device. The detection beam is reflected after encountering the sediment on the seabed surface. The type of sediment on the seabed surface and whether there are cultural relics under the seabed sediment will affect the spectrum of the reflected light of the detection beam. By analyzing the spectral characteristics of the reflected light and combining the influence of the seabed environment on the spectral characteristics of the reflected light, it is determined whether there are cultural relics under the seabed sediment. Since the detection is carried out through the light beam, the reflected light generated by the reflection of the sediment on the seabed surface will be gradually absorbed by the marine environment, and no marine noise pollution will be generated, and no damage will be caused to the underwater cultural relics.

[0059] Figure 1 is an optional flow chart of a method for detecting underwater cultural relics based on hyperspectral provided by an embodiment of the present invention. Figure 1 The method may include but is not limited to steps S101 to S103.

[0060] Step S101, acquiring seabed environmental data and enhanced hyperspectral data of the area to be detected;

[0061] Step S102, constructing an original data set according to the seabed environment data and the enhanced hyperspectral data, and performing calculations according to the original data set and the trained recognition network model to obtain a probability value set;

[0062] Step S103, determining the area with cultural relics according to the probability value set and the preset threshold.

[0063] Steps S101 to S103 shown in the embodiment of the present application are applied to Figure 2In the detection scenario shown, the underwater hyperspectral device irradiates the detection beam to the detection area on the seabed and receives the reflected beam according to a certain set interval; at the same time, the seabed environmental data is collected by the real-time environment perception module installed on the underwater hyperspectral device; in this embodiment, the real-time environment perception module is provided with underwater light sensors, pressure sensors, turbidity sensors, temperature sensors, salinity sensors and pH sensors, etc., and the influence of the seabed environment on the propagation, attenuation, refraction and other characteristics of the detection beam is analyzed through the real-time collected seabed environmental data, and then the filter parameters are adjusted, the hyperspectral data is enhanced, and the quality of the hyperspectral data is optimized; the collected seabed environmental data and the data-enhanced hyperspectral data are sent to the processor, and the processor classifies the seabed environmental data and the hyperspectral data by labels, and constructs the original data matrix as the input of the recognition network model Data; input the constructed original data matrix into the trained recognition network model, the recognition network model constructs dynamic graph data according to the original data matrix, and performs multiple iterations of feature learning on the dynamic graph data, extracts the marine environment characteristics and spectral characteristics of the area to be detected in the original data matrix, and makes probability predictions on the possible results in the area to be detected according to the extracted characteristics, including cultural relics covered by sandy sediments, cultural relics covered by muddy sediments, sandy sediments without cultural relics, and muddy sediments without cultural relics; the recognition network model outputs a predicted probability set, and determines the area with cultural relics in the area to be detected according to the probability set output by the recognition network model and the set threshold, and can further divide the boundaries of the determined area with cultural relics based on the probability set output by the recognition network model; wherein, the preset threshold can be freely set according to the actual application, and this embodiment does not make specific restrictions.

[0064] In step S101 of some embodiments, the quality of the spectral data may be optimized by dynamically adjusting parameters of a filter for receiving spectral data. The spectral data may also be optimized by a model such as a neural network, but is not limited thereto.

[0065] See also Figure 3 In some embodiments, step S101 may include but is not limited to steps S301 to S303:

[0066] Step S301, obtaining current sensor data according to a preset cycle; wherein the current sensor data includes any one or more of light intensity data, water depth data, suspended particle concentration data, temperature data, salinity data or pH data;

[0067] Step S302, performing calculations based on a preset linear model and current sensor data to obtain target parameters of the filter; wherein the target parameters include bandwidth and center frequency of the filter;

[0068] Step S303: update the filter according to the target parameters.

[0069] In step S301 of some embodiments, the parameters of the filter are adjusted regularly to cope with the impact of dynamic changes in the seabed environment on the spectral data; the current seabed environment data is collected by a real-time environment perception module provided in the underwater hyperspectral device; in this embodiment, light intensity data, water depth data, suspended particle concentration data, temperature data, salinity data and pH data are collected by the real-time environment perception module.

[0070] In step S302 of some embodiments, after collecting the current seabed environment data, the target parameters of the filter are calculated according to the preset linear model and the current seabed environment data. The target parameters indicate that the filter can obtain high-quality spectral data under the current seabed environment. In some embodiments, the linear model can be established by collecting feedback data of sensors under different seabed environments and the optimal filter parameters through multiple experiments for fitting analysis. The filter parameters include the bandwidth and center frequency of the filter. In this embodiment, the following linear model is established:

[0071] ,

[0072] ,

[0073] in, and is a coefficient determined by experiment, i is the type of data of the seabed ocean environment fed back by the sensor, and is a constant term determined by experiments, L is the light intensity, H is the water depth, T is the suspended particle concentration, is temperature, S is salinity, For pH.

[0074] In step S303 of some embodiments, within the set filter parameter update cycle, the system adjusts the filter according to the calculated target parameters; in this embodiment, a real-time loop mechanism is set in the system to continuously read the environmental data fed back by the real-time sensor module, calculate the target parameters according to the set linear model, and apply them to the filter in real time; at the same time, the system is provided with a data caching mechanism, which can cache the sensor data for several times. Based on the data caching mechanism, the system can maintain the continuity and stability of the dynamic parameter update of the filter when processing delays or data loss occurs; illustratively, the system sets a software filter, and the dynamic parameters are adjusted through the software filter. The software filter adopts Using a hybrid filter structure, when filtering spectral data, preliminary filtering is first performed through an infinite impulse response filter (IIR filter) to quickly remove obvious noise and interference; then fine filtering is performed through a finite impulse response filter (FIR filter) to further optimize signal quality and ensure linear phase characteristics; the system uses a high-precision timer to set the interval for cyclically updating filter parameters. Within the cyclic time interval, the system completes the parameter update of the filter based on the data fed back by the sensor, thereby improving the real-time requirements of the system; the system is also equipped with an exception capture mechanism. Based on the exception capture mechanism, the system can recover or restart automatically in time when communication failures, data errors or mechanical abnormalities occur.

[0075] See also Figure 4 In some embodiments, step S102 may include but is not limited to steps S401 to S405:

[0076] Step S401, constructing first dynamic graph data according to the original data set; wherein the first dynamic graph data includes node features and a first adjacency matrix;

[0077] Step S402, updating node features of the first dynamic graph data according to the recognition network model to obtain a node feature set; calculating a normalization coefficient set according to the node feature set; calculating according to the node feature set and the first preset function to determine a matrix parameter adjustment value set, and updating the first adjacency matrix according to the matrix parameter adjustment value set to obtain a second adjacency matrix;

[0078] Step S403, updating the first dynamic graph data according to the node feature set, the normalization coefficient set and the second adjacency matrix, and comparing the current update number with the preset number;

[0079] Step S404, if the current update number is less than the preset number, return to execute node feature update of the first dynamic graph data according to the trained recognition network model until the current update number is equal to the preset number;

[0080] Step S405: If the current update number is equal to the preset number, full connection processing is performed according to the identification network model and the first dynamic graph data updated last time to obtain a probability value set.

[0081] In step S401 of some embodiments, the system constructs an n-row matrix based on the seabed environmental data and spectral data fed back by the sensor, and each element in the matrix represents the seabed environmental data and spectral data collected and fed back by the underwater hyperspectral device each time; creates dynamic graph data based on the constructed matrix, and the initial features of each node in the dynamic graph structure correspond to each element in the matrix; determines the correlation between nodes based on the static similarity measurement between different nodes, and determines whether there is an edge between two nodes based on the correlation, and sets a numerical weight for the existing edge to quantify the relationship strength of the edge between the nodes; and matrixes the set of weighted edges between the nodes to obtain the adjacency matrix of the dynamic graph.

[0082] In step S402 of some embodiments, the constructed dynamic graph data includes characteristic data of the seabed environment and spectrum detected by the underwater hyperspectral device, and the constructed dynamic graph data is input into the trained recognition network model for feature extraction, and the detection area is predicted based on the extracted features; in this embodiment, the input dynamic graph data is feature extracted by a long short-term memory neural network (LSTM); LSTM realizes feature extraction by continuously updating the feature vectors of the nodes; at the same time, this embodiment sets a dynamic update mechanism, and when the dynamic graph node features are updated, the elements of the adjacency matrix are adaptively updated as the node features are updated, and the weights of the edges are adjusted; at the same time, when the dynamic graph node features are updated, the adjacency relationship between the nodes is updated according to the similarity between different nodes, that is, the adjacency matrix is ​​updated; LSTM uses an enhanced attention mechanism to update the dynamic graph, so as to more effectively capture complex node relationships and feature changes, and calculates a normalization coefficient for each pair of nodes to reflect the relative importance between the nodes, thereby enhancing the feature learning ability of the model and the accuracy of the prediction results; therefore, after the features of the dynamic graph nodes are updated, the normalization coefficients of each pair of nodes are recalculated, and then the adjacency matrix of the dynamic graph data is updated.

[0083] In step S403 of some embodiments, after the LSTM model completes an update of the node features, normalization coefficients and adjacency matrix of the dynamic graph, the LSTM performs a feature extraction on the dynamic graph, and updates the dynamic graph based on the extracted features for the next feature extraction; after completing a feature extraction, the system determines whether the LSTM has ended feature extraction based on a preset number of iterations and performs a prediction.

[0084] In step S404 of some embodiments, the system determines that the number of updates of the dynamic graph data by LSTM has not reached the preset number, and LSTM has not yet learned the key features of the seabed environmental data and spectral data collected by the underwater hyperspectral device. LSTM continues to learn features of the dynamic graph data, and the system records the number of dynamic graph updates.

[0085] In step S405 of some embodiments, after the LSTM updates the features of the dynamic graph a preset number of times, the LSTM inputs the dynamic graph data obtained from the last update into a fully connected layer, and calculates the probability of cultural relics detection for each node feature of the dynamic graph through the fully connected layer; in this embodiment, the LSTM uses a softmax activation function to calculate the last dynamic graph data.

[0086] See also Figure 5 In some embodiments, step S402 may include but is not limited to steps S501 to S503:

[0087] Step S501, determining the current node characteristics and the neighbor node characteristic set respectively according to the first dynamic graph data; wherein the node characteristics include spectral parameters and ocean environment parameters;

[0088] Step S502, performing a first nonlinear calculation based on the identified network model and the current node feature to obtain a first parameter; performing a calculation based on the identified network model and the neighbor node feature set to obtain a second parameter;

[0089] Step S503: Update the current node feature according to the first parameter and the second parameter to obtain a node feature set.

[0090] In step S501 of some embodiments, each node feature in the dynamic graph data includes the seabed environmental parameters and spectral parameters collected each time. The LSTM model captures the hourly changes of node features based on the attention mechanism, and the correlation between each node also affects the update of node features; therefore, the update of node features needs to consider the feature interaction between different nodes. In this embodiment, each node feature includes several different seabed environmental parameters and several different spectral parameters. There is also correlation between the parameters within the feature points, which also affects the update of node features.

[0091] In step S502 of some embodiments, the model updates the node features of the dynamic graph based on the attention mechanism to capture the local feature interactions of the dynamic graph. Through continuous feature updates, the model gradually spreads to the global scale, captures global features, and analyzes the feature interactions between multiple nodes. In this embodiment, the model updates the node features of the dynamic graph based on the following formula to capture the feature interactions of different node features and the interactions of parameters within each node feature:

[0092] ,

[0093] in, is the feature of the i-th node at time step t+1, is the activation function, is the learnable weight matrix of the model for the i-th node at time step t, is the feature of the i-th node at time step t, N is the number of neighboring nodes connected to the i-th node at time step t, is the attention coefficient between nodes i and j connected at time step t, representing the influence of node j on node i, is the learnable weight matrix of the model for the j-th connected neighbor node at time step t, is the feature of the j-th connected neighbor node at time step t.

[0094] In step S503 of some embodiments, the model introduces nonlinearity through an activation function to learn more complex feature representations in node features; and calculates the influence of all neighbor node features connected to the current node on the current node features, and updates the current node features; the model updates all node features of the dynamic graph through the node feature update process mentioned above.

[0095] See also Figure 6 In some embodiments, step S103 may include but is not limited to steps S601 to S604:

[0096] Step S601, determining the current node feature and the neighbor node feature set respectively according to the node feature set; wherein the neighbor node feature set includes feature vectors of several nodes connected to the current node;

[0097] Step S602, calculating according to the identified network model and the current node feature to obtain a first vector; calculating according to the identified network model and the neighbor node feature set to obtain a second vector set;

[0098] Step S603, sequentially connecting the second vector in the second vector set with the first vector to obtain a third vector set; performing a second nonlinear calculation based on the third vector set, the attention parameter vector and the recognition network model to obtain a non-normalized attention coefficient set;

[0099] Step S604, normalize the non-normalized attention coefficient set according to a preset function to obtain a normalized coefficient set.

[0100] In step S601 of some embodiments, the model sets an attention mechanism to capture the complex feature representation of each node in the dynamic graph, and sets an attention coefficient between each pair of connected nodes to represent the influence of the connected neighboring nodes on the current node. The model is more directional and selective when updating node features based on the attention coefficient; after the node features of the dynamic graph are updated, the connection relationship between the nodes changes, and the influence between the nodes also changes; therefore, the attention coefficients between the nodes need to be updated.

[0101] In step S602 of some embodiments, the model determines the node features and the connected neighbor node features according to the updated dynamic graph data. The model calculates the attention coefficient between each pair of connected nodes through the activation function and the learnable parameter vector of the attention mechanism. The specific calculation formula is:

[0102] ,

[0103] in, is the unnormalized attention coefficient of the ith node connected to the jth neighbor node at time step t, is the activation function, is the transposed vector of the learnable parameter vector of the attention mechanism, and are the node features of the i-th node connected to the j-th neighbor node at time step t, is the learnable weight matrix of the model at time step t, Vector concatenation operation.

[0104] In step S603 of some embodiments, according to the mentioned calculation formula, the model updates the features of the current node and the connected neighbor nodes respectively, performs vector connection operation on the updated features, and performs nonlinear processing on the connected features through an activation function. At the same time, through feature connection, the model considers the features of the two nodes at the same time when calculating the attention coefficient, and obtains the attention coefficient between the two connected nodes; the model calculates the attention coefficients of all nodes in the dynamic graph to obtain a set of attention coefficients.

[0105] In step S604 of some embodiments, after the model calculates the attention coefficients of all nodes in the dynamic graph, all attention coefficients are normalized to ensure that the sum of the influences of all connected neighbor nodes of the current node is 1, thereby improving the efficiency of model feature updating; in this embodiment, the attention coefficients of the connected neighbor nodes are normalized by the following calculation formula:

[0106] ,

[0107] in, is the normalized value, is the attention coefficient of the i-th node to the j-th connected neighbor node at time step t, is the sum of the attention coefficients of the ith node and all connected neighbor nodes at time step t.

[0108] In some embodiments, after the system updates the node features of the dynamic graph, the system updates the adjacency relationship according to the similarity changes between the updated nodes, that is, updates the adjacency matrix; the system calculates the weight adjustment value of the connection relationship according to the updated node features, and adjusts the weight of the connection relationship between the nodes according to the calculated weight adjustment value; in this embodiment, the weight adjustment value is calculated by the cosine similarity function, and the specific calculation formula is as follows:

[0109] ,

[0110] in, Adjust the values ​​for the matrix parameters, is the feature vector of node i at time t, is the feature vector of node j at time t, is the norm of the feature vector of node i at time t, is the norm of the eigenvector of node j at time t; in order to dynamically combine the similarity of ocean environment parameters and spectral parameters in each node at different time steps, the weight parameter The similarity of mixed ocean environment parameters and spectral parameters, the change of the overall dynamic graph adjacency matrix over time can be expressed by the following formula:

[0111] ,

[0112] in, is the adjacency matrix of the entire dynamic graph, is the weight parameter, is the adjacency matrix of marine environment parameters, is the adjacency matrix of spectral parameters; as time step t goes by, As it gets bigger, the model gradually shifts from relying on the similarity of environmental parameters to relying on spectral parameters, and a dynamic graph structure is obtained.

[0113] See also Figure 7 In some embodiments, step S103 may include but is not limited to steps 701 to S703:

[0114] Step S701, comparing the maximum probability value in the probability value set with a preset threshold;

[0115] Step S702: if the maximum probability value is less than a preset threshold, the area to be detected is marked as an area without cultural relics;

[0116] Step S703: If the maximum probability value is greater than a preset threshold, the area to be detected is marked as an area with cultural relics.

[0117] In step S701 of some embodiments, the recognition network model processes the input seabed environmental data and spectral data, and outputs a probability value of the detection result of the existence of cultural relics in the area to be detected. The largest probability value is used as the predicted probability of the area to be detected, and the predicted probability of the area to be detected is compared with a preset threshold to determine the detection result of the area to be detected.

[0118] In step S702 of some embodiments, if the predicted probability of the area to be detected is less than a set threshold value, it is determined that there are no cultural relics in the area to be detected, and the system marks the area to be detected as an area without cultural relics; the system can also upload the detection results to a storage database for user viewing and analysis.

[0119] In step S703 of some embodiments, if the predicted probability of the area to be detected is greater than a set threshold, it is determined that there are cultural relics in the area to be detected, and the type of seabed sediment covering the cultural relics is further determined based on the detection results corresponding to the predicted probability value. The technicians can formulate a detailed plan to excavate the cultural relics based on the detection results.

[0120] See also Figure 8 In some embodiments, before detecting submarine cultural relics or processing submarine spectral data, it is necessary to train a recognition network model, extract features from the submarine spectral data through the recognition network model, and make a probability prediction of the existence of cultural relics under the submarine sediments in the detection area based on the extracted features, and output a predicted probability value; the training process of the recognition network model may include but is not limited to including Figure 8 Steps S801 to S806 shown:

[0121] Step S801, obtaining a public data set, dividing the public data set into a training set and a validation set; and determining dynamic graphs according to the training set;

[0122] Step S802, updating the dynamic graph according to the created first model to obtain a first dynamic graph; comparing the current update number with the preset number;

[0123] Step S803, if the current update number is less than the preset number, return to execute updating the dynamic graph according to the created first model to obtain the first dynamic graph, and update the current update number until the current update number is greater than or equal to the preset number;

[0124] Step S804: if the current update number is greater than or equal to the preset number, perform full connection processing according to the first model and the first dynamic graph to obtain a predicted probability value set; perform calculation according to the predicted probability value set and a preset loss function to obtain a model loss value; and compare the model loss value with the preset value;

[0125] Step S805, if the model loss value is greater than the preset value, update the parameters of the first model according to the model loss value and the preset function, and return to execute the update of the node features of the dynamic graph according to the created first model until the model loss value is less than or equal to the preset value;

[0126] Step S806: If the model loss value is less than or equal to the preset value, calculate the model performance value set according to the validation set and the first model, compare the model performance value set with the preset threshold, and if the model performance value meets the preset threshold, use the first model as the recognition network model; otherwise, optimize the parameters of the first model, and return to execute the update of the node features of the dynamic graph according to the created first model until the model performance value meets the preset threshold, and use the first model as the recognition network model.

[0127] In step S801 of some embodiments, a recognition network model is obtained by training a long short-term memory neural network (LSTM) to extract features of sensor feedback data in practical applications, and perform probability prediction based on the extracted features; open source data of discovered submarine cultural relics areas are collected through the Internet and other means, including submarine environmental parameters, spectral parameters and cultural relic detection results of the corresponding areas, the collected data are classified according to the submarine cultural relic areas, and an n-row matrix is ​​constructed based on the classified data. , construct an n-dimensional column vector based on the cultural relic detection results corresponding to the submarine cultural relic area , according to the constructed matrix and column vector Constructing the Matrix ; Randomly extract the matrix The m rows in the new matrix As a training set, the remaining data forms a new matrix As a validation set, in this embodiment, the data volume ratio of the training set to the validation set is 7 to 3; a dynamic graph is constructed based on the sample data in the training set, each sample data is used as a node of the dynamic graph, and the similarity between the sample data is used as the edge of the dynamic graph.

[0128] In step S802 of some embodiments, the constructed dynamic graph is input into the established LSTM network for training. The LSTM network updates the node features of the dynamic graph, the attention coefficients between nodes, and the adjacency matrix of the dynamic graph, and records the current number of updates through a counter. It is determined whether to end the training based on the current number of updates.

[0129] In step S803 of some embodiments, when the number of updates of the model does not reach a preset number, the model repeats the update operation on the dynamic graph and updates the value recorded by the counter until the value recorded by the counter reaches the preset number.

[0130] In step S804 of some embodiments, after the model updates the dynamic graph a preset number of times, the dynamic graph obtained by the last update is output and input into the fully connected layer set in the model. The fully connected layer passes The activation function processes the input dynamic graph and calculates the probability value of each cultural relic detection. The specific calculation formula is as follows:

[0131] ,

[0132] in, is the probability of detecting the i-th cultural relic, is the activation function, is the weight matrix of the fully connected layer, The dynamic graph node features obtained by the last update of the model; the probability value output by the model is the predicted value, according to the column vector in the training set The loss value predicted by the model is calculated based on the data and predicted values ​​in the calculation. The specific calculation formula is as follows:

[0133] ,

[0134] in, is the model loss, is the number of nodes in the dynamic graph, is the probability reference value in the training set, It is the probability prediction value output by the model.

[0135] In step S705 of some embodiments, the model determines whether the currently obtained model loss value meets the requirements. If it does not meet the requirements, the network parameters are optimized by back propagation and gradient descent algorithms. The optimizer adjusts the model's weight matrix to minimize the loss function:

[0136] ,

[0137] in, is the model parameter, including the weight matrix when the node feature is transformed and , and the final weight matrix of the model output layer , η is the learning rate of the model, The loss function parameter The model with adjusted parameters updates the node features of the dynamic graph, predicts the probability of cultural relics based on the updated dynamic graph, and calculates the loss value of the model until the model loss value meets the requirements.

[0138] In step S806 of some embodiments, after the model is trained and the calculated model loss value meets certain requirements, the model performance is evaluated through the validation set to check the accuracy, universality and stability of the model; in this embodiment, the accuracy, universality and stability of the model are calculated through the validation set. and ; If the performance of the model does not meet the preset requirements, the model parameters are tuned, including the model's hyperparameters, such as learning rate, batch size, network depth, etc.; If the performance of the model meets the preset requirements, a trained recognition network model is obtained.

[0139] The following is a detailed description and explanation of the solution of the embodiment of the present invention in conjunction with a specific application example:

[0140] See also Fig. 9 , underwater cultural relics detection is performed in sequence according to step S901, step S902, step S903, step S904 and step S905, the underwater hyperspectral device collects seabed environmental data and spectral data according to the set 1m*1m interval, and feeds the collected data back to the system, the system classifies the fed-back data according to the collected area, and constructs dynamic graph data according to the classified data, and inputs the constructed dynamic graph data into the recognition network model for processing; the recognition network model performs feature learning on the dynamic graph data, and updates the node features of the dynamic graph data to extract key features therein; finally, the model inputs the extracted dynamic graph node features into the final output layer, the output layer calculates the predicted probability value set of different cultural relics detection results in the area according to the input node features, the system uses the cultural relics detection result corresponding to the maximum probability value in the predicted probability value set output by the model as the detection result of the area, the system compares the areas predicted to have cultural relics according to the set threshold, uses the Gauss Kriging method to interpolate the areas with probability values ​​greater than the preset threshold, and draws the boundaries of the seabed area predicted to have cultural relics according to the interpolation results.

[0141] The implementation of the embodiment of the present invention includes the following beneficial effects: the present embodiment provides a method, system, electronic device and storage medium for detecting underwater cultural relics based on hyperspectral. The scheme acquires underwater environmental data and enhanced hyperspectral data with a detection area, constructs an original data set according to the underwater environmental data and enhanced hyperspectral data, calculates according to the original data set and the trained recognition network model, obtains a set of predicted probability values ​​for the area to be detected; sorts the calculated set of predicted probability values ​​to determine the target probability value; marks the area to be detected with a target probability value greater than or equal to a preset threshold as an area with cultural relics, and marks the area to be detected with a target probability value less than a preset threshold as an area without cultural relics. By reflecting the spectrum of the seabed material and taking into account the influence of the seabed environment on the spectrum transmission, noise-free pollution detection of the seabed is achieved.

[0142] like Fig.10 As shown, an embodiment of the present invention further provides a seabed cultural relic detection system based on hyperspectral, comprising:

[0143] The first module is used to obtain seabed environmental data and enhanced hyperspectral data of the area to be detected;

[0144] The second module is used to construct an original data set according to the seabed environment data and the enhanced hyperspectral data, and to perform calculations according to the original data set and the trained recognition network model to obtain a probability value set;

[0145] The third module is used to determine the area with cultural relics according to the probability value set and a preset threshold.

[0146] It can be seen that the contents of the above method embodiments are all applicable to the present system embodiments, the functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0147] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned method for detecting underwater cultural relics based on hyperspectral when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0148] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0149] See also Fig.11 , Fig.11 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0150] The processor 1101 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0151] The memory 1102 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1102 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1102, and the processor 1101 calls and executes a method for detecting underwater cultural relics based on hyperspectral in the embodiment of this application;

[0152] Input / output interface 1103, used to implement information input and output;

[0153] The communication interface 1104 is used to realize the communication interaction between the device and other devices. The communication can be realized through a wired manner (such as USB, network cable, etc.) or a wireless manner (such as mobile network, WIFI, Bluetooth, etc.);

[0154] A bus 1105 that transmits information between various components of the device (e.g., the processor 1101, the memory 1102, the input / output interface 1103, and the communication interface 1104);

[0155] The processor 1101 , the memory 1102 , the input / output interface 1103 and the communication interface 1104 are connected to each other in communication within the device via the bus 1105 .

[0156] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned method for detecting underwater cultural relics based on hyperspectral.

[0157] It can be understood that the contents of the above method embodiments are all applicable to the present storage medium embodiments, the functions specifically implemented by the present storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0158] Among them, the memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. The memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a remote memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0159] In addition, the embodiment of the present application also discloses a computer program product or a computer program, and the computer program product or the computer program is stored in a computer-readable storage medium. The processor of the computer device can read the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device performs the above method. Similarly, the contents in the above method embodiment are all applicable to the storage medium embodiment, and the functions specifically implemented by the storage medium embodiment are the same as those in the above method embodiment, and the beneficial effects achieved are also the same as those achieved by the above method embodiment.

[0160] The embodiments of the present invention provide a method for detecting underwater cultural relics based on hyperspectrum, an underwater cultural relics detection system based on hyperspectrum, an electronic device and a storage medium. The method acquires underwater environmental data and enhanced hyperspectral data with a detection area, constructs an original data set according to the underwater environmental data and the enhanced hyperspectral data, performs calculations according to the original data set and a trained recognition network model, obtains a predicted probability value set of the area to be detected; sorts the calculated predicted probability value set to determine the target probability value; marks the area to be detected with a target probability value greater than or equal to a preset threshold as an area with cultural relics, and marks the area to be detected with a target probability value less than the preset threshold as an area without cultural relics. By reflecting the spectrum of the seabed material and taking into account the influence of the seabed environment on the spectrum transmission, noise-free pollution detection of the seabed is achieved.

[0161] It is understood that all or some of the steps and systems in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0162] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A method for detecting underwater cultural relics based on hyperspectral, characterized in that: The method comprises: Acquire seabed environmental data and enhanced hyperspectral data of the area to be detected; Constructing an original data set according to the seabed environment data and the enhanced hyperspectral data, and constructing first dynamic graph data according to the original data set; wherein the first dynamic graph data includes node features and a first adjacency matrix; The node features of the first dynamic graph data are updated according to the first preset formula to obtain a node feature set; a non-normalized attention coefficient set is calculated according to the second preset formula and the node feature set; a normalized coefficient set is calculated according to the non-normalized attention coefficient set and the third preset formula; a matrix parameter adjustment value set is determined according to the node feature set and the first preset function; the first adjacency matrix is ​​updated according to the matrix parameter adjustment value set to obtain a second adjacency matrix; wherein the first preset function is, in, is the feature of the i-th node at time step t+1, is the activation function, is the learnable weight matrix of the model for the i-th node at time step t, is the feature of the i-th node at time step t, N is the number of neighboring nodes connected to the i-th node at time step t, is the attention coefficient between nodes i and j connected at time step t, representing the influence of node j on node i, is the learnable weight matrix of the model for the j-th connected neighbor node at time step t, is the feature of the j-th connected neighbor node at time step t; the second preset formula is, in, is the unnormalized attention coefficient of the ith node connected to the jth neighbor node at time step t, is the activation function, is the transposed vector of the learnable parameter vector of the attention mechanism, and are the node features of the i-th node connected to the j-th neighbor node at time step t, is the learnable weight matrix of the model at time step t, It is a vector concatenation operation; The third preset formula is: in, is the normalized value, is the attention coefficient of the i-th node to the j-th connected neighbor node at time step t, is the sum of the attention coefficients of the ith node and all connected neighboring nodes at time step t; the first preset function is, in, Adjust the values ​​for the matrix parameters, is the feature vector of node i at time t, is the feature vector of node j at time t, is the norm of the feature vector of node i at time t, is the norm of the feature vector of node j at time t; updating the first dynamic graph data according to the node feature set, the normalization coefficient set and the second adjacency matrix, and comparing the current update number with a preset iteration number; If the current update number is less than the preset iteration number, return to execute the node feature update of the first dynamic graph data according to the first preset formula to obtain a node feature set until the current update number is equal to the preset iteration number; If the current update number is equal to the preset iteration number, a probability value set is obtained by calculating according to the fourth preset formula and the first dynamic graph data updated last time; wherein the fourth preset formula is, in, is the probability of detecting the i-th cultural relic, is the activation function, is the weight matrix of the fully connected layer, The dynamic graph node features obtained from the last update of the model; The area with cultural relics is determined according to the probability value set and a preset threshold.

2. The method according to claim 1, characterized in that The enhanced hyperspectral data is obtained by processing a dynamically updated filter, and the filter is dynamically updated by the following method: Acquire current sensor data according to a preset cycle; wherein the current sensor data includes any one or more of light intensity data, water depth data, suspended particle concentration data, temperature data, salinity data or pH data; Calculating according to a preset linear model and the current sensor data to obtain target parameters of the filter; wherein the target parameters include the bandwidth and center frequency of the filter; The filter is updated according to the target parameter.

3. The method according to claim 1, characterized in that The determining of the area with cultural relics according to the probability value set and the preset threshold specifically includes: Comparing the maximum probability value in the probability value set with the preset threshold; If the maximum probability value is less than the preset threshold, the area to be detected is marked as an area without cultural relics; If the maximum probability value is greater than the preset threshold, the area to be detected is marked as the area with cultural relics.

4. A seabed cultural relic detection system based on hyperspectral, characterized in that: include: The first module is used to obtain seabed environmental data and enhanced hyperspectral data of the area to be detected; A second module is used to construct an original data set according to the seabed environment data and the enhanced hyperspectral data, and to construct first dynamic graph data according to the original data set; wherein the first dynamic graph data includes node features and a first adjacency matrix; The node features of the first dynamic graph data are updated according to the first preset formula to obtain a node feature set; a non-normalized attention coefficient set is calculated according to the second preset formula and the node feature set; a normalized coefficient set is calculated according to the non-normalized attention coefficient set and the third preset formula; a matrix parameter adjustment value set is determined according to the node feature set and the first preset function; the first adjacency matrix is ​​updated according to the matrix parameter adjustment value set to obtain a second adjacency matrix; wherein the first preset function is, in, is the feature of the i-th node at time step t+1, is the activation function, is the learnable weight matrix of the model for the i-th node at time step t, is the feature of the i-th node at time step t, N is the number of neighboring nodes connected to the i-th node at time step t, is the attention coefficient between nodes i and j connected at time step t, representing the influence of node j on node i, is the learnable weight matrix of the model for the j-th connected neighbor node at time step t, is the feature of the j-th connected neighbor node at time step t; the second preset formula is, in, is the unnormalized attention coefficient of the ith node at time step t to the jth connected neighbor node, is the activation function, is the transposed vector of the learnable parameter vector of the attention mechanism, and are the node features of the i-th node connected to the j-th neighbor node at time step t, is the learnable weight matrix of the model at time step t, It is a vector concatenation operation; The third preset formula is: in, is the normalized value, is the attention coefficient of the i-th node to the j-th connected neighbor node at time step t, is the sum of the attention coefficients of the ith node and all connected neighboring nodes at time step t; the first preset function is, in, Adjust the values ​​for the matrix parameters, is the feature vector of node i at time t, is the feature vector of node j at time t, is the norm of the feature vector of node i at time t, is the norm of the feature vector of node j at time t; updating the first dynamic graph data according to the node feature set, the normalization coefficient set and the second adjacency matrix, and comparing the current update number with a preset iteration number; If the current update number is less than the preset iteration number, return to execute the node feature update of the first dynamic graph data according to the first preset formula to obtain a node feature set until the current update number is equal to the preset iteration number; If the current update number is equal to the preset iteration number, a probability value set is obtained by calculating according to the fourth preset formula and the first dynamic graph data updated last time; wherein the fourth preset formula is, in, is the probability of detecting the i-th type of cultural relic, is the activation function, is the weight matrix of the fully connected layer, The dynamic graph node features obtained from the last update of the model; The third module is used to determine the area with cultural relics according to the probability value set and a preset threshold.

5. An electronic device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 3.

6. A computer-readable storage medium storing a program executable by a processor, characterized in that: The processor-executable program is used to perform the method according to any one of claims 1 to 3 when executed by the processor.

Images