Polar region ship channel polarized light semantic segmentation method

By constructing a polarization-generating attention network module and a deep learning algorithm, the resolution and noise interference problems of traditional remote sensing technology in ship navigation monitoring in polar ice areas were solved, achieving high-precision ice condition monitoring and navigation planning.

CN118941790APending Publication Date: 2024-11-12HARBIN ENGINEERING UNIVERSITY SANYA NANHAI INNOVATION & DEVELOPMENT BASE +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410965874.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

Traditional remote sensing technology has limited resolution, is severely affected by noise interference and weather conditions, and is difficult to provide high-precision ice condition monitoring and waterway planning in polar ice areas.

Method used

A polarization-generating attention network module was constructed. By capturing images of the sea surface in polar ice areas and labeling them with category masks, a dataset was built. Using a polarization-based semantic segmentation framework and deep learning algorithms, a semantic segmentation network model was trained to generate high-precision polar ship channel segmentation results.

Benefits of technology

It improves the segmentation accuracy and robustness of shipping channels in polar ice areas, effectively identifies ice blocks and open waters, and provides stable ice condition monitoring and channel planning support.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118941790A_ABST
    Figure CN118941790A_ABST
Patent Text Reader

Abstract

The invention discloses a polar region ship channel polarized light semantic segmentation method, and relates to a polarized light semantic segmentation method. The invention aims to solve the problems of limited resolution, noise interference and weather condition influence in practical application although the traditional remote sensing technology, such as optical and radar imaging, can provide polar ice region ship channel information to a certain extent. According to the method, a sea surface image of a polar region ice region is shot, category mask labeling is carried out on floating ice and a passable channel region of the sea surface, a sea surface target detection data set is constructed, and a polar region ship channel polarized light semantic segmentation data set is constructed; constructing a polarization generation attention network module; constructing a polarized light semantic segmentation framework; semantic segmentation of a data set is carried out by using the constructed polar region ship channel polarized light; and deploying the trained polar region ship channel polarized light semantic segmentation network to realize high-precision polar region ship channel segmentation. The invention belongs to the technical field of deep learning.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a polarized light semantic segmentation method, and belongs to the technical field of deep learning. Background Art

[0002] In recent years, ships sailing in polar regions have faced complex ice conditions and harsh climate environments. Accurate ice monitoring and route planning are essential to ensure navigation safety and efficiency. Although traditional remote sensing technologies, such as optical and radar imaging, can provide information on ship routes in polar ice regions to a certain extent, they still have problems such as limited resolution, noise interference, and the influence of weather conditions in practical applications. Polarized light technology has received widespread attention in polar monitoring in recent years because it can provide more physical information than ordinary optical imaging. Polarized light refers to light with a specific arrangement of the vibration direction of the light wave. The polarization characteristics can be used to more effectively distinguish the surface properties of different substances, such as the difference in surface reflection characteristics of ice and water.

[0003] Polarized light semantic segmentation is a method of image semantic segmentation that combines polarized light image data sources. By fusing polarized light modal information, the accuracy and robustness of the segmentation results can be improved. In polar ship channel planning, polarized light semantic segmentation can provide more comprehensive information on polar ice ship channels and help identify ice, floating ice and open water in the channel. The multimodal semantic segmentation method based on polarized light improves the accuracy and reliability of ice monitoring through data collection, preprocessing, feature extraction, fusion and segmentation, as well as result verification and optimization steps, using polarized light cameras and deep learning algorithms. This method not only excels in accurately distinguishing the boundaries between ice and water, but also provides stable segmentation results in complex environments, providing a reliable means of ice monitoring and channel planning for polar navigation. Summary of the invention

[0004] The present invention aims to solve the problem that although traditional remote sensing technologies, such as optical and radar imaging, can provide information on ship channels in polar ice areas to a certain extent, there are still problems such as limited resolution, noise interference and influence of weather conditions in practical applications. A polar ship channel polarization light semantic segmentation method is proposed in the present invention.

[0005] The technical solution adopted by the present invention to solve the above-mentioned problem is: the steps of the present invention include:

[0006] Step 1: By taking images of the sea surface in the polar ice area and labeling the floating ice and navigable waterway areas on the sea surface with category masks, a sea surface target detection dataset and a polar ship channel polarized light semantic segmentation dataset are constructed;

[0007] Step 2: Construct a polarization generation attention network module;

[0008] Step 3: Build a polarized light semantic segmentation framework;

[0009] Step 4: Use the constructed polar ship channel polarized light semantic segmentation dataset to train the polar ship channel polarized light semantic segmentation network model through the semantic segmentation loss function;

[0010] Step 5: Deploy the trained polar ship channel polarization light semantic segmentation network to achieve high-precision polar ship channel segmentation; the segmentation result is a mask image of the same size as the input image, in which the pixel area with a pixel value of 1 is the navigable channel for polar ships.

[0011] Furthermore, the attention network module extracts four angle polarization images I0, I 45 ,I 90 ,I 135 The features are then stacked in the channel dimension, and features worthy of attention are further obtained by dilated convolution and channel attention. The stacked features are added to the attention features through residual connection, and finally the polarization mode representation image I is obtained through dilated convolution and PReLU activation. P .

[0012] Furthermore, the construction of the polarized light semantic segmentation framework specifically includes:

[0013] The 4-angle polarization images I0,I 45 ,I 90 ,I 135 Input to the polarization generation attention network module to obtain the polarization mode representation image I P , then I P The mean of the four-angle polarization images is input into the multimodal feature fusion network to obtain the multimodal fusion features. Finally, the multimodal fusion features are input into the semantic segmentation head to obtain the polar ship channel segmentation results.

[0014] Furthermore, the semantic segmentation loss in step 4 adopts cross entropy loss, and the specific loss function is:

[0015]

[0016] In the above formula, y c The hot encoding that represents the true pixel category, p c Represents the predicted probability of the pixel category by the semantic segmentation model.

[0017] The beneficial effects of the present invention are as follows: the polarization generation attention network module constructed by the present invention can generate polarization modal representation images with rich polarization information, which is conducive to the semantic segmentation model to fully understand and learn the polarization information characteristics of the segmentation target. By utilizing the different polarization characteristics of different material surfaces, the segmentation network model can understand the difference in surface reflection characteristics of ice and water, thereby achieving high-precision polar ship channel segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flow chart of the present invention;

[0019] Figure 2 is a schematic diagram of the polarization generation attention network module;

[0020] Figure 3 It is a schematic diagram of the network model for polarized light semantic segmentation of polar ship channels. DETAILED DESCRIPTION

[0021] Specific implementation method 1: Figures 1 to 3 As shown, a polar ship channel polarized light semantic segmentation method specifically includes:

[0022] Step 1: By taking images of the sea surface in the polar ice area and labeling the floating ice and navigable waterway areas on the sea surface with category masks, a sea surface target detection dataset and a polar ship channel polarized light semantic segmentation dataset are constructed;

[0023] Step 2: Construct a polarization generation attention network module;

[0024] Step 3: Build a polarized light semantic segmentation framework;

[0025] Step 4: Use the constructed polar ship channel polarized light semantic segmentation dataset to train the polar ship channel polarized light semantic segmentation network model through the semantic segmentation loss function;

[0026] Step 5: Deploy the trained polar ship channel polarization light semantic segmentation network to achieve high-precision polar ship channel segmentation; the segmentation result is a mask image of the same size as the input image, in which the pixel area with a pixel value of 1 is the navigable channel for polar ships.

[0027] Specific implementation method 2: Figures 1 to 3 As shown, the attention network module extracts four angle polarization images I0, I1 through four separate 3x3 convolutions. 45 ,I 90 ,I 135The features are then stacked in the channel dimension, and features worthy of attention are further obtained by dilated convolution and channel attention. The stacked features are added to the attention features through residual connection, and finally the polarization mode representation image I is obtained through dilated convolution and PReLU activation. P .

[0028] Polarization images at four angles are used to generate polarization mode representation images with rich polarization information to enhance the encoder's learning and understanding of polarization information.

[0029] Specific implementation method three: Figures 1 to 3 As shown in the figure, the construction of polarized light semantic segmentation framework specifically includes:

[0030] The 4-angle polarization images I0,I 45 ,I 90 ,I 135 Input to the polarization generation attention network module to obtain the polarization mode representation image I P , then I P The mean of the four-angle polarization images is input into the multimodal feature fusion network to obtain the multimodal fusion features. Finally, the multimodal fusion features are input into the semantic segmentation head to obtain the polar ship channel segmentation results.

[0031] Specific implementation method four: Figures 1 to 3 As shown, the semantic segmentation loss in step 4 adopts cross entropy loss, and the specific loss function is:

[0032]

[0033] In the above formula, y c The hot encoding that represents the true pixel category, p c Represents the predicted probability of the pixel category by the semantic segmentation model.

[0034] The above is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technician familiar with this profession can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent replacement and improvement made to the above embodiments without departing from the content of the technical solution of the present invention, based on the technical essence of the present invention, within the spirit and principles of the present invention, still fall within the protection scope of the technical solution of the present invention.

Claims

1. A polar ship channel polarized light semantic segmentation method, characterized in that: Specifically include: Step 1: By taking images of the sea surface in the polar ice area and labeling the floating ice and navigable waterway areas on the sea surface with category masks, a sea surface target detection dataset and a polar ship channel polarized light semantic segmentation dataset are constructed; Step 2: Construct a polarization generation attention network module; Step 3: Build a polarized light semantic segmentation framework; Step 4: Use the constructed polar ship channel polarized light semantic segmentation dataset to train the polar ship channel polarized light semantic segmentation network model through the semantic segmentation loss function; Step 5: Deploy the trained polar ship channel polarization light semantic segmentation network to achieve high-precision polar ship channel segmentation; the segmentation result is a mask image of the same size as the input image, in which the pixel area with a pixel value of 1 is the navigable channel for polar ships.

2. According to claim 1, a polar ship channel polarized light semantic segmentation method is characterized in that: The attention network module extracts four angle polarization images I0, I 45 ,I 90 ,I 135 The features are then stacked in the channel dimension, and features worthy of attention are further obtained by dilated convolution and channel attention. The stacked features are added to the attention features through residual connection, and finally the polarization mode representation image I is obtained through dilated convolution and PReLU activation. P .

3. The polar ship channel polarized light semantic segmentation method according to claim 1, characterized in that: Building a polarized light semantic segmentation framework specifically includes: The 4-angle polarization images I0,I 45 ,I 90 ,I 135 Input to the polarization generation attention network module to obtain the polarization mode representation image I P , then I P The mean of the four-angle polarization images is input into the multimodal feature fusion network to obtain the multimodal fusion features. Finally, the multimodal fusion features are input into the semantic segmentation head to obtain the polar ship channel segmentation results.

4. The polar ship channel polarized light semantic segmentation method according to claim 1, characterized in that: In step 4, the semantic segmentation loss adopts cross entropy loss, and the specific loss function is: In the above formula, y c The hot encoding that represents the true pixel category, p c Represents the predicted probability of the pixel category by the semantic segmentation model.