Intelligent ship sensing method based on scene complexity
By defining scene complexity levels and integrated learning network models, screening image texture features, and constructing mathematical models, the problem of inaccurate perception of smart ships in complex environments is solved, safety and reliability are improved, and hardware costs and energy consumption are reduced.
Patent Information
- Application Number
- CN202510658460.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-21
- Publication Date
- 2025-09-19
AI Technical Summary
Existing intelligent ship systems have difficulty accurately perceiving and predicting scene complexity in complex navigation environments, leading to erroneous decisions and safety hazards. Multi-sensor fusion technology has high hardware costs, high energy consumption, and limited application in resource-constrained environments. Acoustic perception technology has low resolution and is susceptible to noise interference.
Define the scene complexity level, screen the image texture features, use the gray-level co-occurrence matrix and integrated learning network models (such as XGBoost) to classify the scene complexity, and combine image energy, entropy, contrast, inverse moment and correlation parameters to build a mathematical model for perception.
The safety and reliability of intelligent ships in complex navigation environments are improved. By evaluating the complexity of navigation scenarios and adjusting navigation strategies, hardware costs and energy consumption are reduced, and the accuracy and reliability of perception results are improved.
Smart Images
Figure CN120673350A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of intelligent ship perception, and in particular relates to an intelligent ship perception method based on scene complexity. Background Art
[0002] With the continuous growth of global maritime trade and the deepening development of marine resources, the navigation environment for ships has become increasingly complex and diverse. Traditional ship navigation systems rely primarily on manual control and basic electronic navigation equipment. However, faced with complex navigation environments such as high-traffic waters, adverse weather conditions, narrow waterways, and unknown obstacles, traditional navigation methods are unable to cope. This not only increases the risks of ship operations but also places higher demands on navigation safety and efficiency.
[0003] The concept of smart ships has emerged. By combining advanced sensor technology, artificial intelligence algorithms, data analysis, and automated control systems, they enable autonomous decision-making and intelligent control in a variety of complex navigation scenarios. To effectively cope with the complexity of the navigation environment, smart ships must be able to perceive the complexity of the scene—that is, to monitor and assess the complexity of the current navigation environment in real time and adjust navigation strategies accordingly.
[0004] Currently, existing intelligent ship systems focus primarily on autonomous driving and intelligent navigation technologies, while relatively little research has been conducted on the perception and processing of navigation scene complexity. However, complex navigation environments often carry higher risks. Failure to accurately perceive and predict scene complexity can lead to incorrect decisions and even safety accidents.
[0005] Although smart ship perception technologies have made significant progress in their development, these technologies still face some key challenges and shortcomings.
[0006] Regarding multi-sensor fusion technology, sensor fusion relies on data acquired from different sensors, which may be inconsistent in time and space. Furthermore, fusing data from multiple sensors requires complex computations and high-performance processors, especially in real-time systems. This increases hardware costs and energy consumption, limiting its application in resource-constrained environments. Different sensors also perform differently in different environmental conditions, significantly impacting the accuracy of the fusion results.
[0007] In acoustic sensing technology, while sonar can detect underwater targets, its resolution is typically low, making it difficult to discern small objects or details. Furthermore, sonar's detection range is limited, especially in the deep ocean, where the attenuation and scattering of sound waves significantly reduce detection effectiveness. Furthermore, the ocean environment is plagued by numerous noise sources, such as wind and waves, and sonar signals from other ships. These noises can interfere with acoustic sensing systems, reducing detection effectiveness.
[0008] Therefore, it is necessary to design an intelligent ship perception method based on scene complexity to solve the above problems. Summary of the Invention
[0009] The technical problem to be solved by the present invention is to provide an intelligent ship perception method based on scene complexity, aiming to solve the above technical problems.
[0010] In order to solve the above technical problems, the technical solution adopted by the present invention is: An intelligent ship perception method based on scene complexity, comprising: S1, defines the scene complexity level; S2, filter image texture features; S3, calculate image texture feature parameters; S4, classifies scene complexity using an ensemble learning network model.
[0011] Preferably, in step S1, defining the scene complexity level includes: Complex background: In the shooting scene, the background occupies a certain proportion. The more complex it is, the more complex the pixel values in the corresponding area will change. Complex background not only increases the complexity of the scene, but also interferes with the ship's environmental perception; Camera shake: When a ship is stationary, the effects of wind and waves can cause camera shake, and camera shake is even more common when the ship is in motion. Camera shake can cause the scene image to rotate, tilt, blur, etc., increasing the complexity of the scene. Cast shadows: All kinds of objects in the field of view will cast shadows. Due to cast shadows, bright or dark pixel blocks will appear on the scene image, changing the complexity of the scene; Light changes: Due to changes in natural light and light from other ships, the overall brightness of the navigation scene will change gradually or suddenly, changing the complexity of the scene; Special weather conditions: Ice, snow, fog, haze, heavy rain and other weather conditions will affect the scene complexity of ship navigation. Different weather conditions will lead to different changes in scene complexity. For example, ice and snow will interfere with the perception of other targets, and fog and haze will cause image blur or information loss. Dynamic interference: Dynamic interference such as other ships and floating obstacles will increase the safety risk of the ship during navigation, thereby increasing the complexity of the scene; Preferably, after the uncertainty factors are defined, the scene complexity level is determined according to the number of uncertainty factors included in the navigation scene: C0 (not complex): contains 0 uncertainties; C1 (slightly complex): contains any one uncertain factor; C2 (generally complex): contains any two uncertain factors; C3 (very complex): Contains any 3 or more uncertain factors.
[0012] Preferably, the method for screening image texture features in step S2 is: The texture features in the navigation scene are used to reflect the complexity level of the scene. There are 14 feature quantities that describe the texture complexity of different images: energy, entropy, contrast, inverse moment, correlation, variance, sum average, sum variance, sum entropy, difference variance, difference average, difference entropy, related information measure and maximum correlation coefficient.
[0013] Preferably, in order to avoid the intersection between feature quantities, five texture feature quantities with small correlation and easy calculation are screened out, namely: energy, entropy, contrast, inverse moment and correlation.
[0014] Preferably, the method of step S3 is: Gray co-occurrence matrix of navigation scene image: ; in, , , is the image grayscale, Image size; , is the pixel point on the image.
[0015] Preferably, it also includes: Compute the angular second-order matrix : ; Calculating entropy : ; Calculating contrast : ; Calculate the inverse matrix : ; Calculating correlation : ; By combining the above features, we can get the texture feature vector of any navigation scene image: .
[0016] Preferably, the method in step S4 is to classify the scene complexity of the integrated learning network model, specifically: First, input the feature vector Used to build the first decision tree (CART tree), the goal of this tree is to fit the target value of the input data ,In this process, the first model Model 1 was generated; The second step uses the new eigenvector (Updated residuals based on the previous model). At this point, a second tree is constructed to fit the residuals of the first tree, resulting in the second model, Model 2. Repeat the above steps to generate multiple models; Finally, the output values of all weak models are accumulated to obtain the final result value.
[0017] Furthermore, the XGBoost algorithm is a classic ensemble learning algorithm that continuously improves model performance through iterative operations on sample sets. XGBoost improves the model's predictive capabilities by constructing a series of weak learners (typically decision trees). Furthermore, several optimizations are performed on this basis, making the model more efficient and accurate when handling complex tasks. Intelligent systems based on the XGBoost algorithm can model complex nonlinear processes with multiple inputs and outputs. Within this algorithmic framework, weak learners (decision trees) gradually correct the errors of the previous round using a gradient boosting method to form a powerful predictive model. XGBoost introduces a regularization term to control model complexity, effectively preventing overfitting and further strengthening the integration between the ensemble algorithm framework and weak learners. This XGBoost-based modeling approach can significantly improve the scientificity and effectiveness of the model in complex tasks. It is particularly suitable for intelligent ships to perceive the complexity of the current navigation environment, making their perception results more accurate and reliable.
[0018] Preferably, the formula of step S4 method is expressed as: ; in, is the final predicted value, is the total number of decision trees, is the learning rate, It is A decision tree for a sample The predicted value of .
[0019] Preferably, an intelligent ship perception system includes: A sensor module, used to collect navigation scene images; A processor module configured to execute the intelligent ship perception method based on scene complexity; The control module adjusts the ship navigation strategy according to the output scene complexity level.
[0020] The beneficial effects of the present invention are: The present invention comprehensively considers the uncertain factors in the images of ship navigation scenes, including complex background factors, camera shake factors, cast shadow factors, light change factors, special weather factors and dynamic interference factors. Secondly, the grayscale co-occurrence matrix is used to extract image features, and the XGBoost network model of integrated learning is used to perceive the complexity of the navigation scene. It simulates human visual perception, combines the five parameters of image energy, entropy, contrast, inverse moment and correlation, and constructs a mathematical model that can truly reflect the complexity of the scene; this method provides an important reference for the design and construction of autonomous navigation scenes of intelligent ships, and helps improve the safety and reliability of intelligent ships in actual navigation by evaluating the complexity of virtual test scenes. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 It is a schematic diagram of the process framework of the present invention; Figure 2 This is the flow chart of the XGBoost algorithm in the present invention. DETAILED DESCRIPTION
[0022] Example 1: like Figure 1 As shown, an intelligent ship perception method based on scene complexity includes: S1, defines the scene complexity level; S2, filter image texture features; S3, calculate image texture feature parameters; S4, classifies scene complexity using an ensemble learning network model.
[0023] Preferably, in step S1, the complexity of the navigation scene can reflect the difficulty of constructing the scene itself and the navigation capability of the tested ship. However, the complexity of the ship navigation scene is a relative subjective perception, and different people may have different understandings. Therefore, a unified standard for scene complexity is needed.
[0024] Navigable waters can be categorized as open waters and complex waters. Open waters, for example, are areas of water with a wide field of view, a vast area, and no obvious obstacles, such as the sea, lakes, or wide rivers. These waters typically lack clear navigational restrictions, allowing ships to navigate freely and are suitable for activities such as ocean voyages, fishing, and shipping. Compared to harbors, rivers, or narrow waters, open waters offer greater space and fewer navigation restrictions. Complex waters typically refer to areas with complex navigation conditions and the presence of multiple potential risks or obstacles. Because complex waters harbor many uncertainties, typical complex scenarios often occur in these areas. To categorize these scenarios according to their level of complexity, it is necessary to fully consider the impact of these uncertainties on the scenarios.
[0025] Defining scene complexity levels includes: Complex background: In the shooting scene, the background occupies a certain proportion. The more complex it is, the more complex the pixel values in the corresponding area will change. Complex background not only increases the complexity of the scene, but also interferes with the ship's environmental perception; Camera shake: When a ship is stationary, the effects of wind and waves can cause camera shake, and camera shake is even more common when the ship is in motion. Camera shake can cause the scene image to rotate, tilt, blur, etc., increasing the complexity of the scene. Cast shadows: All kinds of objects in the field of view will cast shadows. Due to cast shadows, bright or dark pixel blocks will appear on the scene image, changing the complexity of the scene; Light changes: Due to changes in natural light and light from other ships, the overall brightness of the navigation scene will change gradually or suddenly, changing the complexity of the scene; Special weather conditions: Ice, snow, fog, haze, heavy rain and other weather conditions will affect the scene complexity of ship navigation. Different weather conditions will lead to different changes in scene complexity. For example, ice and snow will interfere with the perception of other targets, and fog and haze will cause image blur or information loss. Dynamic interference: Dynamic interference such as other ships and floating obstacles will increase the safety risk of the ship during navigation, thereby increasing the complexity of the scene; Preferably, after the uncertainty factors are defined, the scene complexity level is determined according to the number of uncertainty factors included in the navigation scene: C0 (not complex): contains 0 uncertainties; C1 (slightly complex): contains any one uncertain factor; C2 (generally complex): contains any two uncertain factors; C3 (very complex): Contains any 3 or more uncertain factors.
[0026] Preferably, the method for screening image texture features in step S2 is: The texture features in the navigation scene are used to reflect the complexity level of the scene. There are 14 feature quantities that describe the texture complexity of different images: energy, entropy, contrast, inverse moment, correlation, variance, sum average, sum variance, sum entropy, difference variance, difference average, difference entropy, related information measure and maximum correlation coefficient.
[0027] Preferably, in order to avoid the intersection between feature quantities, five texture feature quantities with small correlation and easy calculation are screened out, namely: energy, entropy, contrast, inverse moment and correlation.
[0028] Preferably, the method of step S3 is: Gray co-occurrence matrix of navigation scene image: ; in, , , is the image grayscale, Image size; , is the pixel point on the image.
[0029] Preferably, it also includes: Compute the angular second-order matrix : ; Calculating entropy : ; Calculating contrast : ; Calculate the inverse matrix : ; Calculating correlation : ; By combining the above features, we can get the texture feature vector of any navigation scene image: .
[0030] Example 2: like Figure 2 As shown, preferably, the method in step S4 is to classify the scene complexity of the integrated learning network model, specifically: First, input the feature vector Used to build the first decision tree (CART tree), the goal of this tree is to fit the target value of the input data ,In this process, the first model Model 1 was generated; The second step uses the new eigenvector (Updated residuals based on the previous model). At this point, a second tree is constructed to fit the residuals of the first tree, resulting in the second model, Model 2. Repeat the above steps to generate multiple models; Finally, the output values of all weak models are accumulated to obtain the final result value.
[0031] Furthermore, the XGBoost algorithm is a classic ensemble learning algorithm that continuously improves model performance through iterative operations on sample sets. XGBoost improves the model's predictive capabilities by constructing a series of weak learners (typically decision trees). Furthermore, several optimizations are performed on this basis, making the model more efficient and accurate when handling complex tasks. Intelligent systems based on the XGBoost algorithm can model complex nonlinear processes with multiple inputs and outputs. Within this algorithmic framework, weak learners (decision trees) gradually correct the errors of the previous round using a gradient boosting method to form a powerful predictive model. XGBoost introduces a regularization term to control model complexity, effectively preventing overfitting and further strengthening the integration between the ensemble algorithm framework and weak learners. This XGBoost-based modeling approach can significantly improve the scientificity and effectiveness of the model in complex tasks. It is particularly suitable for intelligent ships to perceive the complexity of the current navigation environment, making their perception results more accurate and reliable.
[0032] Preferably, the formula of step S4 method is expressed as: ; in, is the final predicted value, is the total number of decision trees, is the learning rate, It is A decision tree for a sample The predicted value of .
[0033] Example 2: An intelligent ship perception system includes: A sensor module, used to collect navigation scene images; A processor module configured to execute the intelligent ship perception method based on scene complexity; The control module adjusts the ship navigation strategy according to the output scene complexity level.
[0034] The above embodiments are merely preferred technical solutions of the present invention and should not be construed as limiting the present invention. The embodiments and features in the embodiments of this application may be arbitrarily combined with each other unless they conflict. The scope of protection of the present invention shall be the technical solutions described in the claims, including equivalent alternatives to the technical features of the technical solutions described in the claims. Equivalent alternatives and improvements within this scope are also within the scope of protection of the present invention.
Claims
1. An intelligent ship perception method based on scene complexity, characterized in that: include: S1, defines the scene complexity level; S2, filter image texture features; S3, calculate image texture feature parameters; S4, classifies scene complexity using an ensemble learning network model.
2. The intelligent ship perception method based on scene complexity according to claim 1, characterized in that: In step S1, the scene complexity level is defined to include complex background, camera shake, cast shadows, light changes, special weather and dynamic interference.
3. The intelligent ship perception method based on scene complexity according to claim 2, characterized in that: After defining the uncertainties, the complexity level of the scenario is determined based on the number of uncertainties contained in the navigation scenario: C0 (not complex): contains 0 uncertainties; C1 (slightly complex): contains any one uncertain factor; C2 (generally complex): contains any two uncertain factors; C3 (very complex): Contains any 3 or more uncertain factors.
4. The intelligent ship perception method based on scene complexity according to claim 1, characterized in that: The method for screening image texture features in step S2 is: The texture features in the navigation scene are used to reflect the complexity level of the scene. There are 14 feature quantities that describe the texture complexity of different images: energy, entropy, contrast, inverse moment, correlation, variance, sum average, sum variance, sum entropy, difference variance, difference average, difference entropy, related information measure and maximum correlation coefficient.
5. The intelligent ship perception method based on scene complexity according to claim 4 is characterized in that: In order to avoid the intersection between feature quantities, five texture feature quantities with small correlation and easy to calculate are screened out, namely: energy, entropy, contrast, inverse moment and correlation.
6. The intelligent ship perception method based on scene complexity according to claim 1, characterized in that: The method of step S3 is: Gray co-occurrence matrix of navigation scene image: ； in, , , is the image grayscale, Image size; , is the pixel point on the image.
7. The intelligent ship perception method based on scene complexity according to claim 6 is characterized in that: Also includes: Compute the angular second-order matrix : ； Calculating entropy : ； Calculating contrast : ； Calculate the inverse matrix : ； Calculating correlation : ； By combining the above features, we can get the texture feature vector of any navigation scene image: 。 8. The intelligent ship perception method based on scene complexity according to claim 1 is characterized in that: Step S4 is an ensemble learning network model classification scenario complexity, specifically: Input feature vector Used to build the first decision tree (CART tree), the goal of this tree is to fit the target value of the input data ,In this process, the first model Model 1 was generated; Use the new eigenvectors , that is, the residuals updated based on the previous model, build a second tree to fit the residuals of the first tree, and get the second model Model 2; Repeat the above steps to generate multiple models; Finally, the output values of all weak models are accumulated to obtain the final result value.
9. The intelligent ship perception method based on scene complexity according to claim 5 is characterized in that: The formula of step S4 method is expressed as: ； in, is the final predicted value, is the total number of decision trees, is the learning rate, It is A decision tree for a sample The predicted value of .
10. An intelligent ship perception system, characterized in that: include: A sensor module, used to collect navigation scene images; A processor module configured to execute the method according to any one of claims 1 to 9; The control module adjusts the ship navigation strategy according to the output scene complexity level.