Polar region ice layer thickness underwater intelligent measurement method based on optical image and single-point laser
By combining optical image and single-point laser technology in polar ice thickness measurement, neural network model is constructed, which solves the problems of high cost and low accuracy of traditional measurement methods, and achieves rapid and accurate measurement of polar ice thickness, which is suitable for a large range of application.
Patent Information
- Application Number
- CN202410951526.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-16
- Publication Date
- 2024-11-08
AI Technical Summary
The existing polar ice thickness measurement technology is difficult to achieve high-precision, low-cost and high-efficiency measurement in polar environments. Traditional drilling methods are costly and time-consuming. It is impossible to effectively deal with the complex optical phenomena of ice by relying solely on laser measurement methods. This results in large measurement errors.
Using an intelligent underwater measurement method of polar ice thickness based on optical images and single-point lasers, we predict by constructing a data set, calculating the location of the single-point laser action area, and building a fully connected neural network model, and deploying it on an underwater robot. Drilling technology verifies the data set to achieve rapid and accurate measurement of ice thickness.
It realizes rapid and accurate measurement of ice thickness in polar environments, reduces operating steps and costs, improves measurement accuracy, and is suitable for large-scale and rapid measurements.
Smart Images

Figure CN118918068A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to an intelligent underwater measurement method for polar ice thickness, belonging to the field of deep learning technology. Background Technology
[0002] Accurate ice thickness measurement is crucial for various applications in polar research and navigation. First, it is a fundamental task in planning polar routes, ensuring safe passage for ships and avoiding navigational hazards caused by unforeseen thick ice. Second, ice thickness data is essential for environmental monitoring and climate change research; analyzing ice thickness trends can provide important information about global climate change. Furthermore, for underwater vehicles such as submarines, accurate ice thickness is key to surfacing and icebreaking. Only with accurate ice thickness information can appropriate icebreaking locations and methods be selected, ensuring the safety and effectiveness of the icebreaking process. In short, ice thickness measurement not only involves navigation safety and scientific research but also has broad application prospects in military and resource development.
[0003] However, measuring polar ice thickness presents significant challenges. Polar ice layers are typically thick and structurally complex, and traditional measurement methods have numerous limitations in this environment. Drilling is a common method, but it is costly, requires complex equipment and procedures, and is time-consuming, making it unsuitable for rapid and large-scale ice thickness measurements. Furthermore, the anisotropy within the ice layer causes scattering and refraction of light as it propagates, further increasing the difficulty of measurement. Laser measurement methods alone often produce significant measurement errors because they cannot effectively handle these complex optical phenomena, such as optical scattering and refraction. Therefore, existing measurement technologies struggle to achieve high-precision, low-cost, and efficient ice thickness measurements in polar environments. To address this, a new method is needed that combines optical imaging and single-point laser technology. By fusing optical imaging and laser ranging data, measurement accuracy can be improved while simplifying procedures and reducing costs, thereby enabling intelligent and efficient measurement of polar ice thickness. Summary of the Invention
[0004] To address the challenge of achieving high-precision, low-cost, and high-efficiency ice thickness measurement in polar environments using existing measurement technologies, this invention proposes an intelligent underwater measurement method for polar ice thickness based on optical imaging and single-point laser.
[0005] The technical solution adopted by the present invention to solve the above problems is as follows: The steps of the present invention include:
[0006] Step 1: Construct a dataset for measuring the thickness of underwater polar ice layers;
[0007] Step 2: Calculate the pixel coordinates of the single-point laser action area;
[0008] Step 3: Construct a network model for predicting polar ice thickness;
[0009] Step 4: Complete the training of the polar ice thickness prediction network model;
[0010] Step 5: Quantitative deployment of the polar ice thickness prediction network model;
[0011] Step 6: Measure the thickness of the polar ice layer.
[0012] Furthermore, an underwater robot equipped with a single-point laser rangefinder was used to measure the ice thickness at a distance of 0.3 meters from the ice surface. The ice thickness measured by the single-point laser rangefinder was obtained, and an optical camera was used to collect images of the laser beam in the ice. At the same time, drilling technology was used to measure the actual thickness of the ice, and an underwater polar ice thickness measurement dataset was constructed.
[0013] Furthermore, the underwater polar ice thickness measurement dataset includes ice thickness measured by a single-point laser rangefinder, the actual ice thickness measured by drilling technology, and images of the ice layer rendered by a single-point laser.
[0014] Furthermore, step 2 specifically includes:
[0015] Based on the optical images in the underwater polar ice thickness measurement dataset, the region where a single-point laser is imaged in the image is found, and the pixel coordinates of the rectangular bounding box of this region are calculated. This fixed position region is taken as the single-point laser action area.
[0016] Furthermore, step 3 specifically includes:
[0017] A fully connected neural network was used to construct a polar ice thickness prediction network model. The neural network model has 6 layers, with an input dimension of 256, and the dimensions of the intermediate hidden layers are 512, 1024, 512, 256, and 128, respectively. The output dimension of the network is 1. Each neural network layer is followed by a batch normalization layer and a ReLU activation function layer. There are two independent neural network encoding layers at the input of the network model, which encode the ice thickness measurement value of a single laser and the image of the single laser action area into a 128-dimensional vector.
[0018] Furthermore, step 4 specifically includes:
[0019] The ice thickness measured by a single-point laser rangefinder is encoded into a 128-dimensional vector using a fully connected neural network. Simultaneously, based on the bounding box pixel coordinates, an image of the single-point laser's effective area is extracted and flattened, then encoded into a 128-dimensional vector using a fully connected neural network. These two vectors are concatenated and used as input to the polar ice thickness prediction network model. The network outputs an ice thickness vector, which is a 128-dimensional vector. This vector is then decoded into a numerical value using a fully connected neural network, which is the predicted ice thickness.
[0020] Furthermore, step 5 specifically includes:
[0021] The ground truth ice thickness values in the underwater polar ice thickness measurement dataset are encoded using a single-point laser encoding network. Then, the squared difference between the ice thickness vector output by the polar ice thickness prediction network model and the true ice thickness vector is calculated as one loss of the network model. Then, the squared difference between the decoded predicted ice thickness value and the actual ice thickness value is calculated as another loss of the network model. Using these two losses and the underwater polar ice thickness measurement dataset, the polar ice thickness prediction network model is trained. After training, the polar ice thickness prediction network model can achieve fast and accurate polar ice thickness measurement.
[0022] Furthermore, step 6 specifically includes:
[0023] The trained polar ice thickness prediction network model was quantized and accelerated, and then deployed on an underwater robot to measure the polar ice thickness.
[0024] The beneficial effects of this invention are: the underwater intelligent measurement method for polar ice thickness based on optical images and single-point laser designed in this invention can accurately measure the thickness of polar ice. Compared with other methods, it has a faster speed and achieves rapid and accurate polar ice thickness measurement. Attached Figure Description
[0025] Figure 1 is a flow chart of the present invention;
[0026] Figure 2 This is a schematic diagram of the present invention. Detailed Implementation
[0027] Specific implementation method one: as follows Figure 1 and Figure 2 As shown, an intelligent underwater measurement method for polar ice thickness based on optical images and single-point lasers specifically includes:
[0028] Step 1: Construct a dataset for measuring the thickness of underwater polar ice layers;
[0029] Step 2: Calculate the pixel coordinates of the single-point laser action area;
[0030] Step 3: Construct a network model for predicting polar ice thickness;
[0031] Step 4: Complete the training of the polar ice thickness prediction network model;
[0032] Step 5: Quantitative deployment of the polar ice thickness prediction network model;
[0033] Step 6: Measure the thickness of the polar ice layer.
[0034] Specific implementation method two: such as Figure 1 and Figure 2 As shown, step 1 specifically includes:
[0035] The thickness of the ice layer was measured underwater at a distance of 0.3 meters from the surface of the ice layer using an underwater robot equipped with a single-point laser rangefinder. The ice layer thickness measured by the single-point laser rangefinder was obtained, and an image of the laser appearing in the ice layer was captured using an optical camera. At the same time, drilling technology was used to measure the actual thickness of the ice layer, and an underwater polar ice layer thickness measurement dataset was constructed.
[0036] Specific implementation method three: such as Figure 1 and Figure 2 As shown, the underwater polar ice thickness measurement dataset includes ice thickness measured by a single-point laser rangefinder, the actual ice thickness measured by drilling technology, and images of the ice layer rendered by a single-point laser.
[0037] Specific implementation method four: such as Figure 1 and Figure 2 As shown, step 2 specifically includes:
[0038] Based on the optical images in the underwater polar ice thickness measurement dataset, the region where a single-point laser is imaged in the image is found, and the pixel coordinates of the rectangular bounding box of this region are calculated. This fixed position region is taken as the single-point laser action area.
[0039] Since the optical camera and single-point laser rangefinder mounted on the underwater robot are in fixed positions, the area of effect of the single-point laser is also fixed.
[0040] Specific implementation method five: such as Figure 1 and Figure 2 As shown, step 3 specifically includes:
[0041] A fully connected neural network was used to construct a polar ice thickness prediction network model. The neural network model has 6 layers, with an input dimension of 256, and the dimensions of the intermediate hidden layers are 512, 1024, 512, 256, and 128, respectively. The output dimension of the network is 1. Each neural network layer is followed by a batch normalization layer and a ReLU activation function layer. There are two independent neural network encoding layers at the input of the network model, which encode the ice thickness measurement value of a single laser and the image of the single laser action area into a 128-dimensional vector.
[0042] Specific implementation method six: such as Figure 1 and Figure 2 As shown, step 4 specifically includes:
[0043] The ice thickness measured by a single-point laser rangefinder is encoded into a 128-dimensional vector using a fully connected neural network. Simultaneously, based on the bounding box pixel coordinates, an image of the single-point laser's effective area is extracted and flattened, then encoded into a 128-dimensional vector using a fully connected neural network. These two vectors are concatenated and used as input to the polar ice thickness prediction network model. The network outputs an ice thickness vector, which is a 128-dimensional vector. This vector is then decoded into a numerical value using a fully connected neural network, which is the predicted ice thickness.
[0044] Specific implementation method seven: such as Figure 1 and Figure 2 As shown, step 5 specifically includes:
[0045] The ground truth ice thickness values in the underwater polar ice thickness measurement dataset are encoded using a single-point laser encoding network. Then, the squared difference between the ice thickness vector output by the polar ice thickness prediction network model and the true ice thickness vector is calculated as one loss of the network model. Then, the squared difference between the decoded predicted ice thickness value and the actual ice thickness value is calculated as another loss of the network model. Using these two losses and the underwater polar ice thickness measurement dataset, the polar ice thickness prediction network model is trained. After training, the polar ice thickness prediction network model can achieve fast and accurate polar ice thickness measurement.
[0046] Specific implementation method eight: such as Figure 1 and Figure 2 As shown, step 6 specifically includes:
[0047] The trained polar ice thickness prediction network model was quantized and accelerated, and then deployed on an underwater robot to measure the polar ice thickness.
[0048] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent substitutions, and improvements made to the above embodiments without departing from the scope of the present invention, based on the technical essence of the present invention and within the spirit and principles of the present invention, shall still fall within the protection scope of the present invention.
Claims
1. An underwater intelligent measurement method for polar ice thickness based on optical images and single-point laser, characterized in that: Specifically include: Step 1: Construct underwater polar ice thickness measurement dataset; Step 2, calculating the pixel coordinates of the single-point laser action area; Step 3: Construct a polar ice thickness prediction network model; Step 4: Complete the polar ice thickness prediction network model training; Step 5: Quantitative deployment of polar ice thickness prediction network model; Step 6: Measure the thickness of polar ice.
2. According to claim 1, a method for underwater intelligent measurement of polar ice thickness based on optical images and single-point laser is characterized in that: Step 1 specifically includes: The thickness of the ice layer was measured by an underwater robot equipped with a single-point laser rangefinder at a distance of 0.3 meters from the ice surface underwater. The ice thickness measured by the single-point laser rangefinder was obtained, and an optical camera was used to capture the image of the laser in the ice. At the same time, drilling technology was used to measure the actual thickness of the ice layer, and an underwater polar ice thickness measurement data set was constructed.
3. The method for underwater intelligent measurement of polar ice thickness based on optical images and single-point laser according to claim 2 is characterized in that: The underwater polar ice thickness measurement dataset includes ice thickness measured by a single-point laser rangefinder, true ice thickness measured by drilling technology, and images presented by a single-point laser in the ice.
4. The method for underwater intelligent measurement of polar ice thickness based on optical images and single-point laser according to claim 1 is characterized in that: Step 2 specifically includes: According to the optical images in the underwater polar ice thickness measurement dataset, the area where the single-point laser is imaged in the image is found, and the pixel coordinates of the rectangular bounding box of the area are calculated, and the fixed position area is used as the single-point laser action area.
5. The method for underwater intelligent measurement of polar ice thickness based on optical images and single-point laser according to claim 1 is characterized in that: Step 3 specifically includes: A fully connected neural network is used to construct a polar ice thickness prediction network model; the neural network model has a total of 6 layers, with an input dimension of 256, the dimensions of the middle hidden layers are 512, 1024, 512, 256, and 128 respectively, and the final network output dimension is 1; each layer of the neural network is followed by a BN batch normalization layer and a ReLU activation function layer, and there are two independent neural network encoding layers at the input end of the network model, which encode the ice thickness measurement value of the single-point laser and the image of the single-point laser action area into a 128-dimensional vector.
6. The method for underwater intelligent measurement of polar ice thickness based on optical images and single-point laser according to claim 1 is characterized in that: Step 4 specifically includes: The ice thickness value measured by the single-point laser rangefinder is encoded into a 128-dimensional vector using a layer of fully connected neural network. At the same time, according to the pixel coordinates of the bounding box, the image of the single-point laser action area is intercepted and flattened and encoded into a 128-dimensional vector using a layer of fully connected neural network. These two vectors are concatenated as the input of the polar ice thickness prediction network model. The network outputs the ice thickness vector, which is a 128-dimensional vector. Then, a layer of fully connected neural network is used to decode the vector into a numerical value, which is the predicted ice thickness.
7. The method for underwater intelligent measurement of polar ice thickness based on optical images and single-point laser according to claim 1 is characterized in that: Step 5 specifically includes: The true value of ice thickness in the underwater polar ice thickness measurement dataset is encoded using a single-point laser encoding network, and then the square difference between the ice thickness vector output by the polar ice thickness prediction network model and the true ice thickness vector is calculated as a loss of the network model. Then the square difference between the decoded predicted ice thickness value and the actual ice thickness value is calculated as another loss of the network model. These two losses and the underwater polar ice thickness measurement dataset are used to complete the training of the polar ice thickness prediction network model. After the training is completed, the polar ice thickness prediction network model can realize fast and accurate polar ice thickness measurement.
8. The method for underwater intelligent measurement of polar ice thickness based on optical images and single-point laser according to claim 1 is characterized in that: Step 6 specifically includes: The trained polar ice thickness prediction network model is quantized and accelerated and deployed on an underwater robot to measure the polar ice thickness.
