Underwater laser echo signal processing model training method and water depth measuring method

Abstract

The invention discloses an underwater laser echo signal processing model training method and a water depth measurement method, and belongs to the field of laser radar water depth detection.The method comprises the steps that multiple sets of underwater laser echo signals are obtained and preprocessed, and an underwater laser echo signal data set is obtained; the preprocessing comprises the following steps: after normalizing each group of underwater laser echo signals, marking the type of a data point located on a water surface waveform rising edge as a water surface point, marking data located on a water bottom waveform falling edge as a water bottom point, and marking other data points as noise points; and building a classification model for predicting the type of each data point in the underwater laser echo signal, and training the classification model by using the underwater laser echo signal data set to obtain an underwater laser echo signal processing model. According to the method, the data points located at the water surface waveform rising edge and the water bottom waveform falling edge can be accurately identified in a complex environment, and a reliable basis is provided for calculation of the time difference between the water bottom point and the water surface point, so that the measurement precision of the water body depth is improved.

Description

Technical Field

[0001] The present invention belongs to the field of lidar bathymetry, and more specifically, relates to a method for training an underwater lidar echo signal processing model and a method for measuring water depth. Background Art

[0002] The area of the ocean is extremely vast, accounting for about 71% of the entire Earth's surface area. The ocean not only provides the necessary physical environment for the sustainability of the ecosystem and biodiversity, but also provides important guarantees for human food, clothing, housing and transportation. Bathymetry is the basic work of ocean resource exploration. High-resolution bathymetric data not only helps the research of the ecological environment and climate change, but also plays an important role in environmental monitoring and protection, as well as the rational utilization of resources. Traditional ocean bathymetry methods mainly include acoustic sounding and electromagnetic sounding, etc. Acoustic sounding measures the water depth by emitting acoustic waves and receiving the reflected signals, while electromagnetic sounding uses the reflection information of electromagnetic waves. Although these methods can obtain accurate seabed terrain data, they have disadvantages such as limited measurement range, slow response time, and being greatly affected by environmental factors.

[0003] In contrast, lidar bathymetry has been widely used in the field of ocean exploration due to its fast response ability and good adaptability to various environments. People can obtain complete waveform information through the lidar system and extract ocean depth information based on the echo signal, so as to achieve high-precision and large-scale three-dimensional seabed terrain mapping. However, the laser pulse energy decays exponentially with the transmission depth, and the attenuation rate is positively correlated with the turbidity of the water column. This results in that the lidar system can only obtain clear sea surface and seabed echo signals in areas with appropriate depth and relatively high seawater clarity, while in areas with deeper water or more turbid water, only seabed echo signals with a low signal-to-noise ratio can be obtained. Traditional lidar echo bathymetry technology cannot well detect such weak seabed echo signals because more information about the signal waveform data is needed for seabed data detection. At the same time, in areas with relatively shallow water depth, the sea surface and seabed echoes are prone to overlap due to the short time interval between them. The overlap of waveforms increases the difficulty of separating and extracting the water surface and seabed waveforms, and reduces the accuracy and reliability of water depth measurement. How to solve the problems of low signal-to-noise ratio and waveform overlap has become the key issue in waveform processing.

[0004] Facing the problems encountered in the waveform extraction process, researchers studied the characteristics of laser echoes and water body characteristics, and proposed various waveform processing methods. For example, Saylam et al. used a moving average filtering algorithm to smooth the laser waveform; Wang et al. believed through experimental comparison that the Richardson-Lucy deconvolution (RLD) method could better solve the waveform extraction of weak echo signals, while the average variance function (ASDF) could better handle noise. Pan et al. proposed a continuous wavelet transform (CWT) algorithm to solve the problem of the overlap of sea surface and seabed echoes in shallow water areas. It can be seen that these methods have given good solutions to different problems in the waveform extraction process. However, this has also led to the lack of unity in all laser echo processing methods. For the echo signal in a certain area, it is necessary to manually classify the echo signal, and then select different algorithms for depth extraction, which increases the time cost and human input. Compared with traditional methods, the neural network in deep learning methods can automatically obtain the feature information of the waveform and the correlation information between adjacent waveforms, showing great potential in laser echo signal processing. Hu et al. constructed a multi-layer fully connected neural network FCN and a one-dimensional convolutional neural network CNN to classify the echo signal into two categories: land and sea, with an accuracy of 95.6%. Xu et al. classified the laser echo signal into 5 categories through deep learning methods, and then processed the waveform signal according to the category. However, such methods focus on using deep learning models to achieve the classification of the overall waveform of laser echo signals, and have limited adaptability to complex water environments, and the accuracy of water depth measurement cannot be guaranteed. Summary of the Invention

[0005] In view of the defects and improvement requirements of the prior art, the present invention provides a method for training an underwater laser echo signal processing model and a method for measuring water depth. The purpose is to propose a new deep learning model for processing underwater laser echo signals based on the characteristic that water depth measurement depends on the time difference between the bottom point and the water surface point, so as to accurately identify the data points on the rising edge of the water surface waveform and the data points on the falling edge of the bottom waveform in the underwater laser echo signal in a complex environment, and provide a reliable basis for calculating the time difference between the bottom point and the water surface point, thereby improving the measurement accuracy of water depth.

[0006] To achieve the above object, according to one aspect of the present invention, there is provided a method for training an underwater laser echo signal processing model, including:

[0007] Obtain multiple groups of underwater laser echo signals and perform preprocessing to obtain an underwater laser echo signal dataset; each group of underwater laser echo signals includes multiple data points, and each data point includes time and echo intensity; the preprocessing includes: after normalizing each group of underwater laser echo signals, label the data point types on the rising edge of the water surface waveform in each group of underwater laser echo data as water surface points, label the data on the falling edge of the bottom waveform as bottom points, and label the remaining data points as noise points;

[0008] Build a classification model based on a deep learning model to predict the type of each data point in the underwater laser echo signal, and use the underwater laser echo signal dataset to train the classification model. After the training is completed, an underwater laser echo signal processing model is obtained.

[0009] Furthermore, the classification model is a CNN-LSTM deep learning model, which includes a convolutional neural network, a long short-term memory network, a fully connected network, and a Softmax function layer connected in sequence;

[0010] The convolutional neural network is used to extract the local features of each data point in the underwater laser echo signal;

[0011] The long short-term memory network is used to capture the correlation between data points in the underwater laser echo signal and obtain the semantic features of each data point;

[0012] The fully connected network is used to map the semantic features of each data point in the underwater laser echo signal to the output category features of each data point;

[0013] The Softmax function layer is used to convert the output category features of each data point into the probability that the data point belongs to each category.

[0014] Furthermore, the convolutional neural network includes five one-dimensional convolutional layers, and each convolutional layer consists of a convolutional operation, batch normalization, and an activation function ReLU;

[0015] The long short-term memory network includes an LSTM layer, a batch normalization layer, and an activation function ReLU.

[0016] According to another aspect of the present invention, there is provided a product of an underwater laser echo signal processing model, including the underwater laser echo signal processing model; the underwater laser echo signal processing model is trained by the above-mentioned underwater laser echo signal processing model training method provided by the present invention.

[0017] According to another aspect of the present invention, there is provided a method for measuring the water depth based on the underwater laser echo signal, including:

[0018] Obtain the underwater laser echo signal of the water body to be measured, normalize it and input it into the underwater laser echo signal processing model to obtain the type of each data point;

[0019] Obtain the peak echo intensity of the water surface points and the corresponding time T respectively 1 , and the peak echo intensity of the water bottom points and the corresponding time T 2 , and calculate the depth H of the water body to be measured according to H = v*(T 2 -T 1 );

[0020] Among them, v represents the propagation speed of the laser in the water body to be measured; the underwater laser echo signal processing model is trained by the above-mentioned underwater laser echo signal processing model training method provided by the present invention.

[0021] According to another aspect of the present invention, there is provided a computer program product, including a computer program; when the computer program is executed by a processor, it implements the above-mentioned underwater laser echo signal processing model training method provided by the present invention, and / or, the above-mentioned water body depth measurement method based on the underwater laser echo signal provided by the present invention.

[0022] According to another aspect of the present invention, there is provided a computer-readable storage medium, including a stored computer program; when the computer program is executed by a processor, it controls the device where the computer-readable storage medium is located to execute the above-mentioned underwater echo signal processing model training method provided by the present invention, and / or, the above-mentioned water body depth measurement method based on the underwater echo signal provided by the present invention.

[0023] According to another aspect of the present invention, there is provided an electronic device, including:

[0024] A computer-readable storage medium for storing a computer program;

[0025] And a processor for reading the computer program in the computer-readable storage medium and executing the above-mentioned underwater laser echo signal processing model training method provided by the present invention, and / or, the above-mentioned water body depth measurement method based on the underwater laser echo signal provided by the present invention.

[0026] According to another aspect of the present invention, there is provided a water body depth measurement system, including: a lidar and a measurement module;

[0027] The lidar is used to obtain the underwater laser echo signal of the water body to be measured;

[0028] The measurement module is used to normalize the underwater laser echo signal and then input it into the underwater laser echo signal processing model to obtain the types of each data point; obtain the peak echo intensity of the water surface points and the corresponding time T respectively 1 , and the peak echo intensity of the water bottom points and the corresponding time T 2 , and calculate according to H = v*(T 2 -T1 ) Calculate the depth H of the water body to be measured;

[0029] Among them, v represents the propagation speed of the laser in the water body to be measured; the underwater laser echo signal processing model is trained by the above-mentioned underwater laser echo signal processing model training method provided by the present invention.

[0030] Generally speaking, through the above technical solutions conceived by the present invention, the following beneficial effects can be achieved:

[0031] (1) The present invention makes full use of the advantages of the deep learning method, classifies each data point in the underwater laser echo data by using a deep learning model, and in the dataset used for model training, only the data points on the rising edge of the water surface waveform of the laser echo data are labeled as water surface points, and the data points on the falling edge of the bottom waveform are labeled as bottom points, and the remaining data points are labeled as noise points. Therefore, when the model processes the underwater laser, it can accurately identify the data points on the rising edge of the water surface waveform and the data points on the falling edge of the bottom waveform, and can accurately determine the time corresponding to the bottom point and the water surface point even in complex environments such as turbid water bodies, overlapping water surface echoes and bottom echoes, thereby improving the measurement accuracy of the water body depth.

[0032] (2) The present invention uses the CNN-LSTM model as the classification model, which can extract the feature information of the data points and can also capture the correlation information between the data points, effectively improving the accuracy of data point classification. Description of the Drawings

[0033] Figure 1 It is a schematic diagram of existing underwater laser echo signal data;

[0034] Figure 2 It is a flow chart of the underwater laser echo signal processing model training method provided by the embodiment of the present invention;

[0035] Figure 3 It is a schematic diagram of the classification model provided by the embodiment of the present invention;

[0036] Figure 4 It is a schematic diagram of the underwater laser echo signal processing result provided by the embodiment of the present invention; among them, (a), (b) and (c) respectively represent the processing results of the underwater laser echo signal in three different scenarios;

[0037] Figure 5 It is a schematic diagram of the water body depth measurement method based on the underwater laser echo signal provided by the embodiment of the present invention. Detailed Embodiments

[0038] To make the objectives, technical solutions, and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0039] In the present invention, terms such as "first" and "second" in the present invention and the accompanying drawings (if any) are used to distinguish similar objects and do not necessarily need to describe a specific order or sequence.

[0040] By irradiating the water surface with a lidar or the like and collecting the corresponding echo signals, the underwater laser echo signals can be obtained. Clear underwater laser echo signals can be obtained in areas with appropriate depths and relatively high seawater clarity. Figure 1 As shown, it is a waveform schematic diagram of an underwater laser echo signal. The water surface waveform and the water bottom waveform can be separated from it. Among these two waveforms, the data points corresponding to the peak values of the echo intensity respectively correspond to the data points where the laser reaches the water surface and the data points where the laser reaches the water bottom. The time difference between these two data points is the propagation time of the laser in the water. After obtaining this time difference and combining it with the propagation speed of the laser in the water, the water depth can be calculated. However, in areas with relatively deep water or relatively turbid water, only seafloor echo signals with a low signal-to-noise ratio can be obtained. In areas with relatively shallow water, the sea surface and seafloor echoes are prone to overlap due to the short time interval between them. These complex situations will affect the accuracy and reliability of the water bottom depth measurement.

[0041] To address the above problems, the present invention provides an underwater laser echo signal processing model training method and a water depth measurement method. The overall concept is to propose a new deep learning model for underwater laser echo signal processing based on the characteristic that water depth measurement depends on the time difference between the water bottom point and the water surface point, enabling it to accurately identify the data points on the rising edge of the water surface waveform and the data points on the falling edge of the water bottom waveform in the underwater laser echo signal in a complex environment, providing a reliable basis for calculating the time difference between the water bottom point and the water surface point, and thus improving the measurement accuracy of the water depth.

[0042] The following are embodiments.

[0043] Embodiment 1:

[0044] An underwater laser echo signal processing model training method, as Figure 2 shown, includes steps S1 - S2. The specific steps are as follows:

[0045] S1: Obtain multiple groups of underwater laser echo signals and perform preprocessing to obtain an underwater laser echo signal dataset;

[0046] It is easy to understand that during the scanning process of lidar, the space is divided into multiple small regions, and each region corresponds to a bin value. These bin values contain information such as the distance and intensity measured by the lidar in that region. In this embodiment, the underwater laser echo signal is composed of a number of bin values, each bin value corresponding to a data point, and each set of underwater laser echo signals can be regarded as a time series;

[0047] Considering that the data environments sampled at different times are not consistent, resulting in a large difference in signal intensity due to complex noise distributions, in this embodiment, the preprocessing of the underwater laser echo signal includes: performing normalization processing on the data points that make up the underwater laser echo signal, and processing its data range to the range of 0 to 1 to prepare for subsequent training;

[0048] To address the problem of the water surface waveform and the bottom waveform overlapping, in this embodiment, the preprocessing further includes: labeling the data point types of the data points on the rising edge of the water surface waveform in each set of underwater laser echo data as water surface points, labeling the data on the falling edge of the bottom waveform as bottom points, and labeling the remaining data points as noise points; based on this data point type labeling method, the trained model can accurately identify the data points on the rising edge of the water surface waveform and the data points on the falling edge of the bottom waveform in the underwater laser echo signal;

[0049] S2: Build a classification model based on a deep learning model to predict the types of each data point in the underwater laser echo signal, and use the underwater laser echo signal data set to train the classification model. After the training is completed, an underwater laser echo signal processing model is obtained.

[0050] Since the underwater laser echo signal has certain temporal information, in order to improve the accuracy of data point classification, as a preferred implementation manner, in this embodiment, the classification model is specifically a CNN-LSTM deep learning model, which includes a convolutional neural network, a long short-term memory network, a fully connected network, and a Softmax function layer connected in sequence.

[0051] The convolutional neural network CNN is a deep learning model specifically used to process data with a grid structure (such as images). Through special layer structures such as convolutional layers and pooling layers, CNN can effectively capture local features in images and achieve high-level representation and classification of images.

[0052] Based on using CNN to capture the spatial features of data, LSTM can further capture the temporal dependence features of data. CNN performs well in processing spatio-temporal data. Through operations such as convolutional layers and pooling layers, it can effectively extract local features and has translational invariance, which can adapt to features at different positions. While LSTM can capture long-term dependence relationships in data through recurrent connections and gating mechanisms and can handle the temporal information of sequential data. Secondly, the CNN-LSTM model can learn multi-level abstract feature information of data. In the waveform dataset, the value of each point contains information about noise points, water surface points, and water bottom points, and this information can often be accurately distinguished and classified only through feature representations at different levels. The multi-layer structure of the CNN-LSTM model can gradually extract higher-level features, from local features to global features, enabling the model to better understand the internal structure and semantic information of the data. In addition, the CNN-LSTM model also has the advantages of parameter sharing and parallel computing. Parameter sharing means that in the convolutional layer of CNN, the same convolutional kernel can extract similar features at different positions, thus reducing the number of model parameters and improving the efficiency of the model. Parallel computing can make full use of the parallel computing capabilities of modern computing devices to accelerate the model training and inference processes and meet the requirements of large-scale data and real-time applications. Finally, the CNN-LSTM model has achieved remarkable results in many fields of sequential data, such as speech recognition, natural language processing, and action recognition. The data in these fields has temporal and spatial correlations, so the CNN-LSTM model has prior advantages and good performance in processing similar data. The main reason for selecting the CNN-LSTM model as the backbone structure in this embodiment is that it can effectively process sequential data with spatio-temporal relationships and can learn multi-level abstract feature representations. By combining the advantages of CNN and LSTM, the model can capture the spatial features and temporal dependence relationships of data, adapt to the characteristics of the dataset, and thus achieve accurate classification tasks.

[0053] As Figure 3 shown, in this embodiment, five one-dimensional convolutional layers are used to achieve feature extraction of the input data; in order to represent the local semantic information of the data after convolutional processing, the feature sequence after passing through the convolutional neural network (CNN) is input into the LSTM, and the correlation between each data point and other data points is obtained through the LSTM layer, so as to better achieve the prediction of the data point category; the features output by the LSTM are mapped to the final output category through a fully connected layer, and the classification probability is calculated through the Softmax function; finally, each data point will contain 3 feature information, respectively representing the probabilities of this data point being a water surface point, a water bottom point, and a noise point in 3 categories, and the category corresponding to the highest probability is the category of the data point, thus realizing the classification of laser echo data points.

[0054] That is, the main structures of the deep learning model are the Convolutional Neural Network (CNN) and the Long Short-Term Memory Network (LSTM). Among them, the Convolutional Neural Network uses five one-dimensional convolutional layers, and each convolutional layer consists of a convolution operation, batch normalization, and the activation function ReLU. The role of batch normalization is to normalize the data of each batch, which can help accelerate the training process of the network and has a regularization effect, making preparations for improving the classification performance of the model subsequently. In addition, the model uses the ReLU activation function after each convolutional layer. The characteristic of the ReLU function is to set the value to zero in the negative part and keep the positive part unchanged. Its role is to introduce non-linearity and increase the expression ability of the model. The number of input channels of the Convolutional Neural Network (CNN) is 1, and the number of output channels is 32. After the input data passes through the Convolutional Neural Network, the original one-dimensional intensity information becomes 32-dimensional feature information, thus realizing the feature extraction of the input data.

[0055] In order to represent the local semantic information of the data after convolution processing, the feature sequence after convolution and batch normalization is input into the LSTM. LSTM is a recurrent neural network suitable for processing sequence data. It has memory units and gating mechanisms, which can effectively capture and utilize the long-term dependencies in the sequence data. The relevance between each data point and other data points is obtained through the LSTM layer, so as to better realize the prediction of the data point category. The number of input channels of the LSTM input data is 32, and the number of output channels is 64. Each data point in the output data contains not only its own feature information but also the correlation information with other data points. Then the data will pass through a fully connected layer to map the features output by the LSTM to the final output category. The role of the fully connected layer is to perform a linear mapping and transformation on the high-dimensional features to obtain the final prediction result. The number of input channels of the fully connected layer input data is 64, and the number of output channels is 3. The 3 pieces of information contained in each data point in the output data represent the weights occupied in the three categories of water surface points, water bottom points, and noise points. After the fully connected layer, the Softmax function is used to calculate the classification probability. The Softmax function converts the output of the fully connected layer into a probability distribution, representing the prediction probability of each category. The 3 pieces of information contained in each data point respectively represent the probabilities of this data point being a water surface point, a water bottom point, and a noise point. The category corresponding to the highest probability is the category of the data point, thus realizing the classification of underwater laser echo data points.

[0056] In step S2 of this embodiment, the classification model is trained using the underwater laser echo signal dataset, which specifically includes:

[0057] The underwater laser echo signal dataset is divided into a training set, a test set, and a validation set;

[0058] First, the CNN-LSTM model is trained using the training set. For each set of data trained, the loss between the output data and the labels is calculated through the loss function. After obtaining the loss between the two, backpropagation is performed through the loss to update the parameters inside the model once, making the loss smaller.

[0059] For each round of data trained, the updated deep learning model is tested on the test set to monitor the changes in the model evaluation metrics. According to the accuracy on the test set, the batch range of the data set for the next model training is changed. Optionally, when the accuracy rate does not exceed a certain threshold (e.g., 90%), the entire training set is used for the next round of training. When the accuracy rate exceeds this threshold, the range batch of the training set data is changed accordingly, and the training model also becomes to perform a set of tests on the test set for each set of training, so as to observe the change in the accuracy rate of the classification model on the test set and thus find the optimal solution.

[0060] Finally, a set of CNN-LSTM deep learning models with the best test results is selected as the echo data classification model for this time and verified through the validation set.

[0061] Figure 4 Shows the processing results of the underwater laser echo signal processing model obtained by training in this embodiment for underwater laser echo signals in different scenarios. (a), (b), and (c) respectively correspond to three different scenarios. The left half of the figure represents the original underwater laser echo signal, and the right half represents the corresponding processing results. According to Figure 4 As can be seen from the shown processing results, the underwater laser echo signal processing model obtained by training in this embodiment can accurately identify the data points located at the rising edge of the water surface waveform and the data points located at the falling edge of the bottom echo in the underwater laser echo signal. Especially according to Figure 4 in (b) and (c), it can be known that in the case where the water surface waveform and the bottom waveform overlap, this embodiment can also accurately complete the classification of the data points in the underwater laser echo signal. The experimental results show that the accuracy rate of classifying the data points in this embodiment reaches 97.62%. Compared with the real water depth data, the root mean square error of the water depth value obtained by the CNN-LSTM model is 0.46m, achieving a good water body sounding effect.

[0062] Generally speaking, this embodiment can accurately identify the data points located at the rising edge of the water surface waveform and the data points located at the falling edge of the bottom echo in the underwater laser echo signal in a complex water body environment, respectively find the peak echo intensities in the water surface points and the bottom points, and respectively use them as the water surface peak point and the bottom peak point. According to the time difference between the water surface peak point and the bottom peak point, the water body depth information can be obtained, thereby accurately improving the water body depth measurement accuracy in a complex water body environment.

[0063] Embodiment 2:

[0064] An underwater laser echo signal processing model product, comprising an underwater laser echo signal processing model; the underwater laser echo signal processing model is obtained by training with the underwater laser echo signal processing model training method provided in the above-mentioned Embodiment 1.

[0065] Embodiment 3:

[0066] An underwater water depth measurement method based on an underwater laser echo signal, as Figure 5 , comprising:

[0067] Obtain the underwater laser echo signal of the water body to be measured, normalize it and input it into the underwater laser echo signal processing model to obtain the types of each data point;

[0068] Respectively obtain the peak echo intensity of the water surface point and the corresponding time T 1 , and the peak echo intensity of the water bottom point and the corresponding time T 2 , and calculate the depth H of the water body to be measured according to H = v * (T 2 -T 1 );

[0069] Wherein, v represents the propagation speed of the laser in the water body to be measured; the underwater laser echo signal processing model is obtained by training with the underwater laser echo signal processing model training method provided in the above-mentioned Embodiment 1.

[0070] Embodiment 4:

[0071] A computer program product, comprising a computer program; when the computer program is executed by a processor, it implements the underwater laser echo signal processing model training method provided in the above-mentioned Embodiment 1, and / or, the underwater water depth measurement method based on the underwater laser echo signal provided in the above-mentioned Embodiment 3.

[0072] Embodiment 5:

[0073] A computer-readable storage medium, comprising a stored computer program; when the computer program is executed by a processor, it controls the device where the computer-readable storage medium is located to execute the underwater echo signal processing model training method provided in the above-mentioned Embodiment 1, and / or, the underwater water depth measurement method based on the underwater echo signal provided in the above-mentioned Embodiment 3.

[0074] Embodiment 6:

[0075] An electronic device, comprising:

[0076] A computer-readable storage medium for storing a computer program;

[0077] And a processor, configured to read a computer program in a computer-readable storage medium and execute the underwater laser echo signal processing model training method provided in Embodiment 1 above, and / or the water depth measurement method based on the underwater laser echo signal provided in Embodiment 3 above.

[0078] Embodiment 7:

[0079] A water depth measurement system, comprising: a lidar and a measurement module;

[0080] The lidar is configured to obtain an underwater laser echo signal of the water body to be measured;

[0081] The measurement module is configured to normalize the underwater laser echo signal and then input it into the underwater laser echo signal processing model to obtain the types of each data point; respectively obtain the peak echo intensity and the corresponding time T of the water surface point 1 , and the peak echo intensity and the corresponding time T of the bottom point 2 , and calculate the depth H of the water body to be measured according to H = v * (T 2 - T 1 );

[0082] Wherein, v represents the propagation speed of the laser in the water body to be measured; the underwater laser echo signal processing model is trained by the underwater laser echo signal processing model training method provided in Embodiment 1 above.

[0083] Those skilled in the art can easily understand that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for training an underwater laser echo signal processing model, characterized in that: include: Acquire multiple groups of underwater laser echo signals and perform preprocessing to obtain underwater laser echo signal data sets; Each group of underwater laser echo signals includes multiple data points, each of which includes time and echo intensity; the preprocessing includes: after normalizing each group of underwater laser echo signals, marking the data points located on the rising edge of the water surface waveform in each group of underwater laser echo data as surface points, marking the data points located on the falling edge of the bottom waveform as bottom points, and marking the remaining data points as noise points; A classification model is built based on the deep learning model to predict the type of each data point in the underwater laser echo signal, and the classification model is trained using the underwater laser echo signal data set. After the training, the underwater laser echo signal processing model is obtained.

2. The underwater laser echo signal processing model training method according to claim 1, characterized in that: The classification model is a CNN-LSTM deep learning model, which includes a convolutional neural network, a long short-term memory network, a fully connected network and a Softmax function layer connected in sequence; The convolutional neural network is used to extract local features of each data point in the underwater laser echo signal; The long short-term memory network is used to capture the correlation between data points in the underwater laser echo signal to obtain the semantic features of each data point; The fully connected network is used to map the semantic features of each data point in the underwater laser echo signal into output category features of each data point; The Softmax function layer is used to convert the output category feature of each data point into the probability that the data point belongs to each category.

3. The underwater laser echo signal processing model training method according to claim 2, characterized in that: The convolutional neural network includes five one-dimensional convolutional layers, each of which consists of a convolution operation, batch normalization, and an activation function ReLU; The long short-term memory network includes an LSTM layer, a batch normalization layer and an activation function ReLU.

4. An underwater laser echo signal processing model product, characterized in that: It comprises an underwater laser echo signal processing model; the underwater laser echo signal processing model is trained by the underwater laser echo signal processing model training method according to any one of claims 1 to 3.

5. A method for measuring water depth based on underwater laser echo signals, characterized in that: include: The underwater laser echo signal of the water body to be measured is obtained, and the signal is input into the underwater laser echo signal processing model after normalization to obtain the type of each data point; Obtain the peak value of the echo intensity of the water surface point and the corresponding time T1, as well as the peak value of the echo intensity of the water bottom point and the corresponding time T2, and calculate the depth H of the water body to be measured according to H=v*(T2-T1); Wherein, v represents the propagation speed of the laser in the water body to be measured; the underwater laser echo signal processing model is trained by the underwater laser echo signal processing model training method described in any one of claims 1-3.

6. A computer program product, characterized in that It comprises a computer program; when the computer program is executed by a processor, it implements the underwater laser echo signal processing model training method described in any one of claims 1 to 3, and / or the water depth measurement method based on the underwater laser echo signal described in claim 5.

7. A computer-readable storage medium, characterized in that: Comprising a stored computer program; when the computer program is executed by a processor, it controls the device where the computer-readable storage medium is located to execute the underwater echo signal processing model training method described in any one of claims 1 to 3, and / or the water depth measurement method based on the underwater echo signal described in claim 5.

8. An electronic device, characterized in that: include: A computer-readable storage medium for storing a computer program; And a processor, used to read the computer program in the computer-readable storage medium, and execute the underwater laser echo signal processing model training method described in any one of claims 1 to 3, and / or the water depth measurement method based on the underwater laser echo signal described in claim 5.

9. A water depth measurement system, characterized in that: Includes: LiDAR and measurement modules; The laser radar is used to obtain underwater laser echo signals of the water body to be measured; The measurement module is used to normalize the underwater laser echo signal and input it into the underwater laser echo signal processing model to obtain the type of each data point; respectively obtain the echo intensity peak value and the corresponding time T1 of the water surface point, and the echo intensity peak value and the corresponding time T2 of the water bottom point, and calculate the depth H of the water body to be measured according to H=v*(T2-T1); Wherein, v represents the propagation speed of the laser in the water body to be measured; the underwater laser echo signal processing model is trained by the underwater laser echo signal processing model training method described in any one of claims 1-3.

Images