Method and system for visual detection of navigation obstacles for a ship

Abstract

The present application belongs to the field of computer vision and image recognition technology, and relates to a ship navigation obstacle visual detection method and system. The method comprises: converting the initial image of the water area in front of the ship to the LAB color space to separate the brightness and chrominance channels; performing CLAHE processing on the brightness channel, calculating the absolute difference value feature, and combining the pixel value to extract the local brightness distortion degree; extracting the color Euclidean distance and the local brightness distortion degree for fusion to obtain the scene chrominance response feature; based on the theoretical maximum normalization constant, combining the local brightness distortion degree to generate a dynamic gain compensation term to construct a color shift compensation coefficient; and using the compensation coefficient to compensate the chrominance channel, and converting the enhanced brightness image back to RGB to generate a global enhanced image. The present application can effectively separate the brightness and chrominance to avoid mutual influence, scientifically analyze the color protection demand, effectively correct the color shift to maintain the original characteristics, and improve the accuracy and safety of ship navigation obstacle visual detection.

Description

Technical Field

[0001] This invention belongs to the field of computer vision and image recognition technology, and relates to a visual detection method and system for obstacles in ship navigation. Background Technology

[0002] With the development of intelligent water transportation, ships navigating in complex waters need to rely on visual detection systems to identify obstacles such as reefs, navigation marks, and floating objects in real time to ensure navigational safety. In water navigation, many critical obstacles, such as red-on-left and green-on-right channel buoys and warning flags of specific colors, are highly dependent on color features for classification and identification.

[0003] In actual visual inspection, water scenes are often accompanied by strong specular reflections or backlighting, which can cause the images captured by the camera to be locally overexposed or the target to be in shadow, resulting in the loss of the edge and texture features of the obstacle. In order to restore local details, the industry usually introduces the Contrast Limited Adaptive Histogram Equalization (CLAHE) algorithm as a pre-image enhancement method. The CLAHE algorithm can effectively improve the contrast of dark obstacles and suppress the excessive enhancement of strong light on the water surface by calculating the local histogram of the image.

[0004] For example, Chinese patent application CN115457258A discloses a foggy ship detection method based on image enhancement algorithms and an improved YOLOv5, including: preprocessing collected foggy ship image data using MSRCP and CLAHE algorithms; labeling the preprocessed images to construct a dataset; improving the network by using k-means++ clustering to design prior boxes; embedding the CA attention module into the C3 module in the main body of the network; replacing the original MaxPool pooling with SoftPool pooling; training the improved YOLOv5 network using the dataset to obtain a detection model and weight file based on the improved YOLOv5 network; and using the trained network model to detect images in the dataset to obtain prediction results.

[0005] However, existing image enhancement methods, when dealing with ship navigation scenes, often only nonlinearly stretch or compress the image brightness when improving brightness and contrast. This can easily induce color shifts under complex lighting conditions on the water surface. When forcibly stretching the contrast of dark areas or suppressing highlights, the original color space balance is broken, causing the originally bright colors of navigation marks to fade or change. This directly prevents subsequent visual detection models from relying on color features to correctly semantically classify obstacle types, thus affecting the accuracy of visual detection of obstacles in ship navigation. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art, solve the technical problem that the color shift and fading of obstacles caused by the complex illumination enhancement on the water surface, thereby affecting the accuracy of visual detection of obstacles in ship navigation, and provide a method and system for visual detection of obstacles in ship navigation.

[0007] To achieve the above-mentioned objective, the present invention provides a visual detection method for obstacles in ship navigation, comprising the following steps: An initial image of the waters ahead of the ship is acquired and converted from the RGB color space to the LAB color space to separate the initial luminance channel image and the initial chrominance channel image. CLAHE processing is performed on the initial luminance channel image to obtain an enhanced luminance image. The difference between the enhanced luminance image and the initial luminance channel image is calculated, and the local luminance distortion degree, representing the degree of nonlinear luminance deformation, is extracted by combining the pixel values ​​of the initial luminance channel image. The color Euclidean distance based on the initial chrominance channel image is extracted and fused with the local luminance distortion degree to obtain scene chrominance response features representing color protection requirements. Based on the scene chrominance response features and the theoretical maximum normalization constant, combined with the local luminance distortion degree, a dynamic gain compensation term for the chrominance channel is generated to construct a color shift compensation coefficient. Pixel compensation is performed on the initial chrominance channel image using the color shift compensation coefficient to obtain a corrected chrominance channel image. The enhanced luminance image and the corrected chrominance channel image are then inversely converted back to the RGB color space to generate a global enhanced image, enabling obstacle visual detection during ship navigation.

[0008] This invention achieves effective separation of luminance and chromaticity information by converting the initial image to the LAB color space, ensuring that subsequent luminance enhancement processing does not directly affect chromaticity information, thus providing a reliable technical foundation for color protection. By calculating local luminance distortion, it assesses the degree of nonlinear luminance deformation, accurately reflecting the impact of CLAHE processing on luminance distribution, providing a reliable basis for luminance change characteristics in subsequent color protection. By extracting scene chromaticity response features, it analyzes color protection requirements, accurately reflecting the combined impact of color saturation and luminance distortion, providing a reliable basis for generating dynamic gain compensation terms. By constructing color shift compensation coefficients and performing pixel compensation, it effectively corrects color shifts, ensuring that the original color characteristics are maintained while improving luminance contrast, thereby enhancing the accuracy of visual detection of obstacles in ship navigation.

[0009] Furthermore, to separate the initial luminance channel image and the initial chroma channel image, the method further includes: normalizing the pixel values ​​of the initial luminance channel image by dividing by the theoretical maximum luminance value. The initial chroma channel image is decomposed into an initial red-green chroma channel image and an initial yellow-blue chroma channel image. The pixel values ​​of the initial red-green chroma channel image and the initial yellow-blue chroma channel image are then normalized by dividing by the theoretical maximum absolute value of chroma. Interval.

[0010] This invention ensures that pixel values ​​are within a uniform numerical range through normalization, providing a reliable numerical basis for subsequent calculations of absolute difference and color saturation amplitude, and effectively guaranteeing the accuracy of color space conversion.

[0011] Furthermore, the local brightness distortion degree, which characterizes the degree of nonlinear brightness deformation, is extracted, including: In the formula, coordinates Local brightness distortion at the location, coordinates Pixel values ​​in an image with enhanced brightness. coordinates The pixel values ​​in the initial brightness channel image. It is the absolute value symbol.

[0012] This invention achieves a scientific assessment of local brightness distortion by constructing a self-normalized fraction of the absolute difference in brightness and the overall amplitude. It more accurately reflects the degree of local brightness distortion caused by CLAHE processing. By using basic absolute value operations to replace complex transcendental functions, it can smoothly converge to extreme distortions and reduce hardware computing power.

[0013] Furthermore, scene chromaticity response features used to characterize color protection requirements are obtained, including: In the formula, coordinates The scene's color response characteristics, coordinates The color Euclidean distance between the initial red-green channel image and the initial yellow-blue channel image. coordinates Local brightness distortion at a given location.

[0014] This invention constructs a product model of color saturation Euclidean distance and local brightness distortion, which more accurately reflects the combined effects of color and brightness distortion, ensuring that color protection requirements are only activated in areas that simultaneously have high color saturation and experience significant brightness distortion.

[0015] Furthermore, to construct the color shift compensation coefficients, the following are included: In the formula, coordinates Color shift compensation coefficient at that location. coordinates The scene's color response characteristics, coordinates Local brightness distortion at the location, This is the theoretical maximum normalization constant.

[0016] This invention achieves low-latency calculation of color shift compensation coefficients by constructing an adaptive gain model that includes constant division operations and a product term of local brightness distortion. This ensures that the coefficients are always stably greater than or equal to 1, effectively balancing the needs of color protection and brightness enhancement, and overcoming background distortion caused by global uniform compensation.

[0017] Furthermore, the theoretical maximum normalization constant is determined based on the normalization boundary range of the initial chroma channel image.

[0018] Furthermore, pixel compensation is performed on the initial chroma channel image using a color offset compensation coefficient to obtain a corrected chroma channel image. This includes multiplying the color offset compensation coefficient pixel-by-pixel with the initial red-green chroma channel image and the initial yellow-blue chroma channel image respectively to obtain the corrected chroma channel image. The corrected chroma channel image includes the corrected red-green chroma channel image and the corrected yellow-blue chroma channel image.

[0019] Furthermore, the enhanced brightness image and the corrected chroma channel image are jointly inversely converted back to the RGB color space to generate a global enhanced image. This includes: performing boundary value clamping operations on pixel values ​​that exceed the display range after inverse conversion back to the RGB color space to generate an image without color shift; and adjusting the size of the image without color shift using bilinear interpolation to generate a global enhanced image that adapts to the input size of the pre-trained convolutional neural network object detection model.

[0020] Furthermore, to achieve visual obstacle detection for ship navigation, the method includes: inputting a global augmented image into a pre-trained convolutional neural network object detection model, and outputting obstacle category labels and location bounding boxes.

[0021] This invention also provides a visual detection system for obstacles in ship navigation, employing the following technical solution: A visual detection system for ship navigation obstacles includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned visual detection method for ship navigation obstacles is implemented.

[0022] By adopting the above technical solution, the above-mentioned visual detection method for ship navigation obstacles is generated into a computer program and stored in a memory so that it can be loaded and executed by a processor. A terminal device can then be made based on the memory and the processor for convenient use.

[0023] Compared with the prior art, the present invention has at least the following beneficial effects: Firstly, addressing the issue that traditional CLAHE algorithms only nonlinearly adjust brightness, leading to a disruption of color space balance and color fading / variation of key obstacles such as navigation markers, this invention achieves complete separation of the luminance and chrominance channels by converting the initial image to the LAB color space: CLAHE processing is performed only on the luminance channel to restore details in dark areas and suppress strong light on the water surface, while preserving the original color information of the chrominance channel; then, the degree of nonlinear brightness distortion is assessed by calculating the local luminance distortion, and dynamic gain compensation terms and color shift compensation coefficients are generated by combining the initial color saturation amplitude to perform pixel compensation on the chrominance channel. This overcomes the color fading interference caused by traditional nonlinear brightness adjustments, reconstructs and enhances the color saliency of obstacles highly dependent on color features, such as left-red-right-green navigation buoys and warning flags, highlighting the target's category attributes and effectively solving the problem of subsequent classification and recognition failure caused by key color fading under complex lighting conditions.

[0024] Secondly, this invention effectively solves the problem of local overexposure or shadows caused by water surface reflection and backlighting through CLAHE processing of the luminance channel, clearly restoring the edge and texture features of obstacles such as reefs and floating objects. On the other hand, through adaptive chroma channel compensation, it avoids color distortion caused by forcibly stretching the contrast of dark areas or suppressing highlights, ensuring that the enhanced image has both clear dark structural details and easily readable color features. Compared with traditional image enhancement methods that only focus on the naturalness of human visual perception, the globally enhanced image output by this invention is more in line with the feature extraction logic of deep learning models. The generated high color contrast image provides high-quality input data for subsequent convolutional neural network object detection models, effectively reducing the difficulty of the model to extract color features.

[0025] Third, considering the characteristic that key obstacles in water navigation rely on color features for semantic classification, this invention ensures that the subsequent visual detection model can correctly use color features to complete the obstacle type determination through precise color protection: avoiding fatal errors such as red navigation marks being misidentified as green navigation marks and warning flags being misidentified as ordinary floating objects; at the same time, clear edge texture features also effectively reduce the missed detection of obstacles such as reefs and shoals that do not have obvious color features, thereby improving the accuracy and reliability of visual detection of obstacles in ship navigation. Attached Figure Description

[0026] Figure 1 This is a flowchart of the visual detection method for ship navigation obstacles in an embodiment of the present invention.

[0027] Figure 2 This is the initial image of the waterway in front of the ship in the visual detection method for ship navigation obstacles according to an embodiment of the present invention.

[0028] Figure 3 This is an image of the waters ahead of the ship enhanced using the CLAHE algorithm in the visual detection method for ship navigation obstacles according to an embodiment of the present invention.

[0029] Figure 4 This is a global enhanced image of the waters in front of the ship in the visual detection method for ship navigation obstacles according to an embodiment of the present invention. Detailed Implementation

[0030] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings.

[0031] This invention discloses a visual detection method for obstacles in ship navigation, referring to... Figure 1 This includes steps S01-S05: S01: Acquire an initial image of the waters ahead of the ship and perform conversion.

[0032] It should be noted that the traditional RGB color space highly couples brightness and chromaticity, and directly stretching the contrast within this space will inevitably lead to color instability. Therefore, this step requires converting it to a physical space that is decoupled from sensory perception and unifying the dimensions.

[0033] Acquire an initial image of the waters ahead of the ship and convert the initial image from the RGB color space to the LAB color space to separate the initial luminance channel image and the initial chrominance channel image.

[0034] Specifically, to ensure dimensional consistency in subsequent nonlinear operations, normalization is performed on the separated channels. The pixel values ​​of the initial luminance channel image are normalized by dividing by the theoretical maximum luminance value of 100. The initial chroma channel image is decomposed into an initial red-green chroma channel image and an initial yellow-blue chroma channel image, and the pixel values ​​of both are normalized by dividing by the theoretical maximum absolute value of chroma, 128. After normalization, the 0 value in the interval accurately represents a grayscale state without color bias.

[0035] S02: Extract local brightness distortion.

[0036] It should be noted that CLAHE enhances local contrast through block histogram equalization, resulting in differentiated nonlinear adjustments to the brightness of different areas. This brightness distortion is the core cause of subsequent color shifts and needs to be characterized pixel by pixel. Therefore, this step first uses the CLAHE algorithm to restore details in the dark areas of the image and suppress overexposure in the highlight areas of the water surface. Then, by calculating the self-normalized amplitude attenuation relationship based on absolute values, it extracts the local brightness distortion degree, which can accurately reflect the degree of nonlinear brightness distortion of each pixel.

[0037] CLAHE processing is performed on the initial luminance channel image to obtain an enhanced luminance image. The difference between the enhanced luminance image and the initial luminance channel image is calculated. Based on the pixel values ​​of the initial luminance channel image, the local luminance distortion degree, which characterizes the degree of nonlinear luminance deformation, is extracted.

[0038] Specifically, the local brightness distortion degree, which characterizes the degree of nonlinear brightness deformation, is extracted, including: ; In the formula, coordinates Local brightness distortion at the location, coordinates Pixel values ​​in an image with enhanced brightness. coordinates The pixel values ​​in the initial brightness channel image. The sign is absolute. When the acquired image is completely black, there are no pixels with varying brightness. In this case, both the numerator and denominator are 0, which will result in a division by zero error when substituting them into the calculation. Therefore, let... .

[0039] In this case, regardless of whether the brightness is increased or decreased, the absolute difference term is always positive and monotonically increasing, therefore the independent variable is increased brightness. Relative to initial brightness When the absolute change in increases, the local brightness distortion of the dependent variable... The fractional curve increases non-linearly and smoothly; when the pixel does not experience brightness deformation, i.e. hour, A value of zero indicates that when the navigation mark in the backlight is brightened, the distortion in that area increases, meaning there is a higher risk that the original color balance in that area will be broken. However, when the brightness of the flat water surface does not change, the distortion is zero, and subsequent color compensation will not be triggered.

[0040] S03: Obtain scene chromaticity response features.

[0041] It should be noted that the color protection requirements differ significantly across different pixel regions: high-saturation color regions are more sensitive to brightness distortion, and their color fidelity requirements are much higher than those of low-saturation background regions. Therefore, it is necessary to fuse the original color information with brightness distortion information to achieve precise grading of color protection requirements. Thus, this step calculates the color saturation amplitude of the initial chroma channel and combines it with the nonlinear mapping results of local brightness distortion to generate scene chroma response features that can characterize the intensity of color protection requirements pixel by pixel.

[0042] The color Euclidean distance based on the initial chroma channel image is extracted, and the color Euclidean distance is fused with the local luminance distortion to obtain scene chroma response features that characterize color protection requirements.

[0043] Specifically, the scene chromaticity response features used to characterize color protection requirements are obtained, including: ; In the formula, coordinates The scene's color response characteristics, coordinates The color Euclidean distance between the initial red-green channel image and the initial yellow-blue channel image. coordinates Local brightness distortion at a given location.

[0044] Among them, scene chromaticity response features The value is determined by directly multiplying the Euclidean distance of color saturation by the brightness distortion. When both the independent variable, color distance, and the local brightness distortion index increase, the dependent variable, scene chromaticity response characteristics, increase significantly. This indicates that only when a certain pixel region possesses both highly saturated color (such as a red buoy) and has undergone significant brightness changes will the value increase. Only then will it be activated by the highlight, thus filtering out the grayish-white reflective areas on the water surface.

[0045] S04: Construct color shift compensation coefficients.

[0046] It should be noted that using a globally uniform chromaticity gain can lead to color distortion in the background area. Therefore, a pixel-by-pixel dynamic gain strategy is necessary to match the compensation intensity with the color protection requirements and brightness distortion level of the area. Furthermore, to meet the real-time requirements of edge devices, time-consuming operations of globally searching for extrema should be avoided. Therefore, this step first performs global normalization on the scene's chromaticity response characteristics, and then combines this with the local brightness distortion to construct a dynamic color offset compensation coefficient with a base of 1, achieving differentiated chromaticity compensation for different areas.

[0047] Based on the scene's chromaticity response characteristics and the theoretical maximum normalization constant, combined with the local luminance distortion, a dynamic gain compensation term for the chromaticity channel is generated to construct the color shift compensation coefficient.

[0048] Specifically, to construct the color shift compensation coefficient, the following are included: ; In the formula, coordinates Color shift compensation coefficient at that location. coordinates The scene's color response characteristics, coordinates Local brightness distortion at the location, This is the theoretical maximum normalization constant, which is determined based on the normalization boundary range of the initial chroma channel image. Since the initial chroma channels have already been normalized to... Its corresponding theoretical maximum Euclidean distance is Therefore, in this embodiment .

[0049] Among them, the color shift compensation coefficient It is a The dynamic gain as a reference, due to The middle already contains one Multiply again A quadratic suppression effect is formed; as the independent variable scene chromaticity response characteristics and local brightness distortion increase, the dependent variable color shift compensation coefficient increases nonlinearly; in real-world scenarios, for high-saturation key obstacles deemed to have a high risk of color shift, such as the red buoy area, a stability greater than [value missing] is calculated. The dynamic gain compensation coefficient is used to enhance color features; simultaneously, for the large proportion of ordinary water surface areas, due to the quadratic attenuation of extremely small local brightness distortion, the compensation coefficient strongly approaches the value. This avoids the color noise interference introduced by the microwave reflection on the water surface, thus highlighting the color of the obstacle itself and increasing the characteristic distance between the target obstacle and the background environment in the color space.

[0050] S05: Enable visual detection of obstacles for ship navigation.

[0051] It should be noted that the initial chroma channel retains the most original and realistic color information of the image. Gain compensation based on the initial chroma channel can minimize the introduction of additional color deviations. At the same time, the generated enhanced image needs to be compatible with the input specifications of the subsequent deep learning detection model. Therefore, this step first performs pixel-by-pixel gain compensation on the original red-green and yellow-blue chroma channels, then merges them with the enhanced luminance channel and converts them back to the RGB color space, finally generating a globally enhanced image that meets the input requirements of the detection model and completing obstacle recognition.

[0052] The initial chroma channel image is pixel-compensated using a color shift compensation coefficient to obtain a corrected chroma channel image. The enhanced luminance image and the corrected chroma channel image are then converted back to the RGB color space to generate a global enhanced image, thereby enabling visual detection of obstacles during ship navigation.

[0053] Specifically, pixel compensation is performed on the initial chroma channel image using a color offset compensation coefficient. The color offset compensation coefficient is then multiplied pixel by pixel with the initial red-green chroma channel image and the initial yellow-blue chroma channel image to obtain a corrected chroma channel image that includes the corrected red-green chroma channel image and the corrected yellow-blue chroma channel image. The enhanced luminance image and the corrected chroma channel image are inversely converted back to the RGB color space. Boundary value clamping is performed on pixel values ​​that exceed the display range after inverse conversion, setting them to 0 or 255 to prevent color gamut overflow. The image without color shift is resized using bilinear interpolation to generate a globally enhanced image that fits the input size of a pre-trained convolutional neural network object detection model. The globally enhanced image is then input into the convolutional neural network object detection model, which uses the restored true color features to output category labels and location bounding boxes for obstacles such as navigation buoys and warning flags, thus achieving visual detection of obstacles in ship navigation.

[0054] like Figure 2 , Figure 3 as well as Figure 4 As shown, Figure 2 There are issues such as mirror-like reflections on the water surface, low overall brightness, insufficient contrast, low color saturation of key obstacles such as red channel buoys and warning flags, and unclear edge details. Figure 3 While the CLAHE algorithm alone improves overall image brightness and shadow detail, and enhances contrast in reflective areas of the water surface, it does not specifically process the chroma channel. Consequently, the color saturation of the buoys and warning flags is not effectively improved, resulting in limited color differentiation from the background. The output image obtained using the method of this invention... Figure 4 While retaining the brightness and contrast enhancement effects of the CLAHE algorithm, dynamic chromaticity compensation was used to improve the color saturation of the red buoys and warning flags, making their color features more vivid. At the same time, the color distinction between the target obstacle and the background environment was improved, and the edge structure and texture details of the buoys were also clearer, which can provide higher quality input data for subsequent convolutional neural network object detection models.

[0055] This invention also discloses a visual detection system for ship navigation obstacles, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the visual detection method for ship navigation obstacles according to this invention.

[0056] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A visual detection method for obstacles in ship navigation, characterized in that, include: Acquire an initial image of the waters ahead of the ship and convert the initial image from the RGB color space to the LAB color space to separate the initial luminance channel image and the initial chrominance channel image; CLAHE processing is performed on the initial luminance channel image to obtain an enhanced luminance image. The difference between the enhanced luminance image and the initial luminance channel image is calculated. Combined with the pixel values ​​of the initial luminance channel image, the local luminance distortion degree, which characterizes the degree of nonlinear luminance deformation, is extracted. Extract the color Euclidean distance based on the initial chroma channel image, and fuse the color Euclidean distance with the local luminance distortion to obtain scene chroma response features that characterize color protection requirements; Based on the scene chromaticity response characteristics and the theoretical maximum normalization constant, combined with the local luminance distortion, a dynamic gain compensation term for the chromaticity channel is generated to construct the color shift compensation coefficient. The initial chroma channel image is pixel-compensated using a color shift compensation coefficient to obtain a corrected chroma channel image. The enhanced luminance image and the corrected chroma channel image are then converted back to the RGB color space to generate a global enhanced image, thereby enabling visual detection of obstacles during ship navigation. Obtain scene chromaticity response features to characterize color protection requirements, including: ； In the formula, coordinates The scene's color response characteristics, coordinates The color Euclidean distance between the initial red-green channel image and the initial yellow-blue channel image. coordinates Local brightness distortion at a given location.

2. The visual detection method for ship navigation obstacles according to claim 1, characterized in that, To separate the initial luminance channel image and the initial chroma channel image, the method also includes: The pixel values ​​of the initial luminance channel image are normalized by dividing by the theoretical maximum luminance value. interval; The initial chroma channel image is decomposed into an initial red-green chroma channel image and an initial yellow-blue chroma channel image. The pixel values ​​of the initial red-green chroma channel image and the initial yellow-blue chroma channel image are then normalized by dividing by the maximum absolute value of chroma theory. Interval.

3. The visual detection method for ship navigation obstacles according to claim 1, characterized in that, Extracting the local brightness distortion degree, which characterizes the degree of nonlinear brightness deformation, includes: ； In the formula, coordinates Local brightness distortion at the location, coordinates Pixel values ​​in an image with enhanced brightness. coordinates The pixel values ​​in the initial brightness channel image. It is the absolute value symbol.

4. The visual detection method for ship navigation obstacles according to claim 1, characterized in that, To construct color shift compensation coefficients, including: ； In the formula, coordinates Color shift compensation coefficient at that location. coordinates The scene's color response characteristics, coordinates Local brightness distortion at the location, This is the theoretical maximum normalization constant.

5. The visual detection method for ship navigation obstacles according to claim 4, characterized in that, The theoretical maximum normalization constant is determined based on the normalization boundary range of the initial chroma channel image.

6. The visual detection method for ship navigation obstacles according to claim 2, characterized in that, The initial chroma channel image is pixel-compensated using a color shift compensation coefficient to obtain a corrected chroma channel image, including: The color shift compensation coefficient is multiplied pixel-by-pixel by the initial red-green hue channel image and the initial yellow-blue hue channel image to obtain the corrected chroma channel image, which includes the corrected red-green hue channel image and the corrected yellow-blue hue channel image.

7. The visual detection method for ship navigation obstacles according to claim 1, characterized in that, The enhanced luminance image and the corrected chroma channel image are inversely converted back to the RGB color space to generate a globally enhanced image, including: Perform boundary value clamping operation on pixel values ​​that are out of display range after being reverse converted back to RGB color space to generate an image without color offset; The image without color shift is resized using bilinear interpolation to generate a globally enhanced image that fits the input size of a pre-trained convolutional neural network object detection model.

8. The visual detection method for ship navigation obstacles according to claim 1, characterized in that, To achieve visual detection of obstacles for ship navigation, including: The global augmented image is input into a pre-trained convolutional neural network object detection model, which outputs obstacle category labels and location bounding boxes.

9. A visual detection system for obstacles in ship navigation, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement the visual detection method for ship navigation obstacles according to any one of claims 1 to 8.

Images