Water surface floating oil identification sampling ship based on machine vision

Abstract

The invention relates to the technical field of water environment monitoring, in particular to a water surface floating oil recognition sampling ship based on machine vision, which comprises a ship body, a navigation propulsion system, a visual recognition system, a sampling system and a control unit, the visual identification system comprises a double-spectrum visual module, and the double-spectrum visual module comprises a first visual sensor and a second visual sensor. The control unit generates a distribution diagram for representing the natural optical flicker intensity of the water surface based on the second waveband image data; based on the first wave band image data, identifying a suspected floating oil area; performing spatial position matching on the suspected floating oil area and an area with optical flicker intensity obviously lower than that of the surrounding water body in the distribution map, and determining the floating oil area according to a matching result; according to the invention, a dual-spectrum visual module is adopted and water surface flicker characteristic analysis based on a near-infrared band is combined as a criterion of floating oil identification, so that the problem of poor identification reliability of a traditional RGB visual method under complex light and shadow and water surface fluctuation is solved.

Description

Technical Field

[0001] This invention relates to the field of aquatic environmental monitoring technology, and in particular to a sampling vessel for identifying floating oil on the water surface based on machine vision. Background Technology

[0002] Oil slicks on the water surface are one of the important indicators of water pollution. Rapid and accurate identification and sampling of oil slicks are crucial for pollution source tracing, environmental assessment and accident handling. Among existing technologies, machine vision-based unmanned sampling vessels have become a research hotspot due to their advantages such as high efficiency, low cost and no need for human on-site contact with pollution sources.

[0003] Existing machine vision-based oil slick identification technology mainly relies on unmanned vessels equipped with conventional RGB (color) cameras. The workflow is usually as follows: capture water surface images through the camera, use image processing algorithms (such as color segmentation, texture analysis, edge detection, etc.) to segment suspected oil slick areas that are different in color and texture from the surrounding water, and then control the vessel to sail to the area for sampling.

[0004] However, this method has some shortcomings in practical applications, resulting in low recognition accuracy and high false alarm rate, especially poor reliability in complex natural environments. Specifically, specular reflections on the water surface (such as solar flares), cloud shadows, floating algae, or fallen leaves present color and texture features in RGB images that are extremely similar to thin oil slicks. It is difficult to reliably distinguish them based solely on color and texture information, leading to frequent misjudgments by the system, forcing vessels to perform ineffective sampling, wasting energy, and reducing operational efficiency. In windy and wavy conditions, ripples on the water surface constantly change the shape of reflected light, causing algorithms based on static texture feature analysis to fail, resulting in large fluctuations in recognition performance. Furthermore, existing technologies typically only navigate vessels to the geometric center of the recognition area, but due to uneven oil slick distribution, vessel positioning errors, and the influence of water flow, the sampling head may not accurately collect the most representative oil film sample, affecting the accuracy of subsequent analysis.

[0005] Therefore, to overcome the above problems, a machine vision-based oil spill identification sampling vessel is proposed. It adopts a dual-spectrum vision module and combines it with water surface scintillation feature analysis based on the near-infrared band as the criterion for oil spill identification, thus solving the problem of poor reliability of traditional RGB vision methods under complex lighting and water surface fluctuations. Summary of the Invention

[0006] In order to overcome the problems of low accuracy, high false alarm rate and poor reliability of existing oil spill identification technology for sampling vessels, especially in complex natural environments.

[0007] The technical solution of this invention is: a machine vision-based oil slick identification and sampling vessel, comprising:

[0008] The hull, used for navigation on water; A propulsion system, installed on the hull, is used to provide power to the hull and control the direction of navigation; A visual recognition system, installed on the hull, is used to collect optical information from the water surface; A sampling system, installed on the hull, is used to collect water samples when oil slicks are detected; The control unit is communicatively connected to the navigation propulsion system, the vision recognition system, and the sampling system, respectively. The visual recognition system includes a dual-spectrum visual module, which comprises: The first visual sensor is configured to acquire first-band image data of the water surface; The second visual sensor has a different spectral response band than the first visual sensor and is configured to acquire second-band image data of the water surface; the second band includes the near-infrared band, which is sensitive to the optical properties of floating oil and has significant specular reflection of the water body. The control unit is configured as follows: Based on the second band image data, a distribution map is generated to characterize the intensity of natural optical scintillation on the water surface by analyzing the changes in pixel statistical features in local areas of the image. Based on the image data of the first band, a suspected oil slick area was identified; The suspected oil slick area is spatially matched with the area in the distribution map that shows a significantly lower optical scintillation intensity than the surrounding water body, and the oil slicker area is confirmed based on the matching results.

[0009] Preferably, the visual recognition system of the device adopts a dual-spectrum visual module, and its image processing logic performs authenticity identification by analyzing the suppression effect of floating oil on the natural optical scintillation characteristics of the water surface.

[0010] Preferably, the first visual sensor is an RGB color camera, and the second visual sensor is a near-infrared camera equipped with a filter. The filter limits the second band to a narrow band with a center wavelength in the range of 850nm to 950nm. The first visual sensor, being an RGB color camera, is used to acquire color images of the scene and initially identify suspected oil slick areas with abnormal color and texture. The second visual sensor, being a near-infrared camera with a specific wavelength, has a narrow-band filter on its lens, making it sensitive only to near-infrared light with a center wavelength in the range of 850nm to 950nm. This wavelength is selected based on the fact that the water surface still produces strong specular reflection of light in this wavelength range and that most petroleum-based oil slicks have a significant absorption effect on near-infrared light in this wavelength range.

[0011] Preferably, the control unit is configured to generate the distribution map by analyzing the changes in pixel statistical features in local regions of the image, specifically: Calculate the pixel value variance or high-frequency component energy of multiple predetermined local regions in the second band image data; Based on the variance value or energy value, a scintillation intensity distribution map corresponding to the second band image data space is generated, wherein areas with high variance values ​​or energy values ​​are characterized as strong scintillation areas, and areas with low variance values ​​or energy values ​​are characterized as weak scintillation areas. In the oil-free, turbulent water surface, the rapid movement and changes of specular reflection spots caused by waves result in high and drastically changing pixel values ​​in local areas of the image sequence, i.e., high variance, manifested as "strong flicker". In contrast, in areas covered by floating oil, the oil film suppresses capillary waves, smooths the liquid surface, and absorbs incident light, causing the specular reflection spots in the area to weaken or even disappear. The pixel values ​​in this area change gradually in the sequence, i.e., low variance, manifested as "weak flicker" or "flicker suppression". Therefore, this distribution map can intuitively indicate the areas where the natural optical flicker of the water surface is suppressed.

[0012] Preferably, the control unit is configured to identify the oil spill area based on the matching result, and the specific determination logic is as follows: When a suspected oil slick area identified in the first band image data has a spatial overlap with a weak flickering area in the distribution map that exceeds a preset threshold, the area is determined to be a real oil slick area that needs to be sampled. Specifically, when a suspected oil slick area spatially overlaps with a clearly defined "weak flickering area," the area is identified as a genuine oil slick area; thus, interference caused by shadows and specular reflections in RGB images can be effectively eliminated, improving the reliability of identification.

[0013] Preferably, the sampling system includes: A sampling arm is movably mounted on the hull; A sampling head, located at the end of the sampling arm, is used to collect water samples; The control unit is also configured to: after confirming the oil slick area, control the propulsion system to make the hull sail above the oil slick area, and control the sampling arm to extend so that the sampling head reaches the predetermined sampling position.

[0014] Preferably, the sampling head integrates an active illumination source, the spectral emission band of which matches the spectral response band of the second visual sensor; after the sampling head is in place, the control unit activates the light source to project a circular light spot onto the water surface directly below.

[0015] Preferably, the control unit is further configured to perform a precise positioning sampling operation after the sampling head reaches the predetermined sampling position, including: Activate the active lighting source to project light spots onto the water surface below; The second visual sensor acquires an image containing the light spot; The center position of the light spot in the image is identified and located using an image processing algorithm; Adjust the position of the sampling arm and / or the hull so that the center of the light spot coincides with the sampling entrance of the sampling head in the image coordinate system, and then start sampling.

[0016] Preferably, the image processing algorithm identifies and locates the center position of the light spot, specifically using the gray-level centroid method to achieve sub-pixel-level center positioning, and calculating the weighted average coordinates of the gray values ​​of all pixels within the light spot region; the formula can be expressed as: ,

[0017] in, The coordinates of the center of the light spot are For pixels grayscale value, For pixels The coordinates are used to quickly and accurately locate the center of brightness of the light spot.

[0018] After accurately locating the center of the light spot, the control unit fine-tunes the position of the sampling arm or hull to make the center of the light spot coincide with the predetermined position of the sampling inlet of the sampling head in the image coordinate system. Due to the absorption of near-infrared light by the oil film area, the light spot on the oil film will present a different shape or brightness than the surrounding water, ensuring that the sampling head can accurately target the center area of ​​the oil film for sampling and improve the representativeness of the sample.

[0019] Preferably, the control unit is configured to perform spatiotemporal synchronization processing on the image data acquired by the dual-spectrum vision module to ensure that the two images are aligned at the pixel level and timestamp level.

[0020] Preferably, the hull is equipped with an attitude sensor, and the control unit is communicatively connected to the attitude sensor and configured to perform motion compensation on the image data collected by the visual recognition system based on the hull attitude data provided by the attitude sensor. The control unit receives the hull roll and pitch data provided by the control unit and performs motion compensation on the images collected by the visual recognition system. Digital image stabilization technology is used to eliminate the image rotation and translation caused by hull swaying, ensuring the stability and accuracy of image analysis.

[0021] The beneficial effects of this invention are: This invention utilizes a near-infrared camera to capture the natural scintillation patterns of water bodies in specific wavelength bands and generates a scintillation intensity distribution map to locate scintillation anomaly areas. It then correlates and matches the color anomaly areas identified by an RGB camera with this scintillation suppression area, thereby eliminating optical interference such as specular reflection and shadows, achieving high-confidence identification of floating oil. Combined with a near-infrared active illumination source integrated into the sampling head, visual servo control enables sub-pixel-level precise positioning of sampling points. This reduces the false alarm rate during the identification stage and improves accuracy during the sampling stage, solving the problem of poor reliability of traditional RGB vision methods under complex lighting and water surface fluctuations. Attached Figure Description

[0022] Figure 1 The diagram shown is a first three-dimensional structural schematic of the water surface oil spill identification and sampling vessel based on machine vision according to the present invention. Figure 2 The diagram shown is a second three-dimensional structural schematic of the water surface oil spill identification and sampling vessel based on machine vision according to the present invention. Figure 3 The diagram shown is a schematic representation of the third three-dimensional structure of the water surface oil spill identification and sampling vessel based on machine vision according to the present invention. Explanation of reference numerals in the attached drawings: 1-hull, 2-propulsion system, 3-visual recognition system, 4-sampling system, 31-first visual sensor, 32-second visual sensor, 41-sampling arm, 42-sampling head, 43-active illumination source. Detailed Implementation

[0023] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0024] Example 1 Please see Figure 1 , Figure 2 and Figure 3 The present invention provides an embodiment: a machine vision-based oil slick identification and sampling vessel, comprising: Hull 1, used for navigation on the water; The navigation propulsion system 2 is installed on the hull 1 and is used to provide power to the hull 1 and control the navigation direction; Visual recognition system 3, installed on hull 1, is used to collect optical information from the water surface; Sampling system 4, installed on hull 1, is used to collect water samples when oil slicks are detected; The control unit is communicatively connected to the navigation propulsion system 2, the vision recognition system 3, and the sampling system 4, respectively. The visual recognition system 3 includes a dual-spectrum visual module, which includes: The first visual sensor 31 is configured to acquire first-band image data of the water surface; The second visual sensor 32 has a different spectral response band than the first visual sensor 31 and is configured to acquire second-band image data of the water surface; the second band includes the near-infrared band, which is sensitive to the optical properties of floating oil and has significant specular reflection of the water body. The control unit is configured as follows: Based on second-band image data, a distribution map is generated to characterize the intensity of natural optical scintillation on the water surface by analyzing the changes in pixel statistical features in local areas of the image. Based on the first-band image data, suspected oil slick areas were identified; The suspected oil slick area was spatially matched with the area in the distribution map that showed a significantly lower optical scintillation intensity than the surrounding water body, and the oil slick area was confirmed based on the matching results.

[0025] The device's visual recognition system 3 employs a dual-spectrum visual module, whose image processing logic identifies authenticity by analyzing the suppression effect of floating oil on the natural optical scintillation characteristics of the water surface.

[0026] Furthermore, the first visual sensor 31 is an RGB color camera, and the second visual sensor 32 is a near-infrared camera equipped with a filter. The filter limits the second band to a narrow band with a center wavelength in the range of 850nm to 950nm. The first visual sensor 31 is an RGB color camera used to acquire color images of the scene and initially identify suspected oil slick areas with abnormal color and texture. The second visual sensor 32 is a near-infrared camera with a specific band. Its lens is equipped with a narrow-band filter, making it sensitive only to near-infrared light with a center wavelength in the range of 850nm to 950nm. The selection of this band is based on the fact that the water surface still produces strong specular reflection of light in this band and that most petroleum-based oil slicks have a significant absorption effect on near-infrared light in this band.

[0027] Furthermore, the control unit is configured to generate a distribution map by analyzing the changes in pixel statistical features in local regions of the image, specifically: Calculate the pixel value variance or high-frequency component energy of multiple predetermined local regions in the second-band image data; A scintillation intensity distribution map corresponding to the second band image data space is generated based on the variance value or energy value. The area with high variance value or energy value is characterized as a strong scintillation area, and the area with low variance value or energy value is characterized as a weak scintillation area. In the oil-free, turbulent water surface, the rapid movement and changes of specular reflection spots caused by waves result in high and drastically changing pixel values ​​in local areas of the image sequence, i.e., high variance, manifested as "strong flicker". In contrast, in areas covered by floating oil, the oil film suppresses capillary waves, smooths the liquid surface, and absorbs incident light, causing the specular reflection spots in the area to weaken or even disappear. The pixel values ​​in this area change gradually in the sequence, i.e., low variance, manifested as "weak flicker" or "flicker suppression". Therefore, this distribution map can intuitively indicate the areas where the natural optical flicker of the water surface is suppressed.

[0028] Furthermore, the control unit is configured to identify the oil slick area based on the matching results. The specific determination logic is as follows: When a suspected oil slick area identified in the first band image data has a spatial overlap with a weak flickering area in the distribution map that exceeds a preset threshold, the area is determined to be a real oil slick area that needs to be sampled. Specifically, when a suspected oil slick area spatially overlaps with a clearly defined "weak flickering area," the area is identified as a genuine oil slick area; thus, interference caused by shadows and specular reflections in RGB images can be effectively eliminated, improving the reliability of identification.

[0029] Furthermore, sampling system 4 includes: Sampling arm 41 is movably mounted on hull 1; The sampling head 42 is located at the end of the sampling arm 41 and is used to collect water samples; The control unit is also configured to: after confirming the oil spill area, control the navigation propulsion system 2 to make the hull 1 sail above the oil spill area, and control the sampling arm 41 to extend so that the sampling head 42 reaches the predetermined sampling position.

[0030] Furthermore, the sampling head 42 integrates an active illumination source 43, the spectral emission band of which matches the spectral response band of the second vision sensor 32; after the sampling head 42 is in place, the control unit activates the light source to project a circular light spot onto the water surface directly below.

[0031] Furthermore, the control unit is further configured to perform a precise positioning sampling operation after the sampling head 42 reaches the predetermined sampling position, including: Activate the active lighting source 43 to project light spots onto the water surface below; The second visual sensor 32 acquires an image containing light spots; The center position of the light spot in the image is identified and located using image processing algorithms; Adjust the position of sampling arm 41 and / or hull 1 so that the center of the light spot coincides with the sampling entrance of sampling head 42 in the image coordinate system, and then start sampling.

[0032] Furthermore, the image processing algorithm identifies and locates the center position of the light spot, specifically using the gray-level centroid method to achieve sub-pixel-level center positioning, and calculating the weighted average coordinates of the gray values ​​of all pixels within the light spot region; the formula can be expressed as: ,

[0033] in, The coordinates of the center of the light spot are For pixels grayscale value, For pixels The coordinates are used to quickly and accurately locate the center of brightness of the light spot.

[0034] After accurately locating the center of the light spot, the control unit fine-tunes the position of the sampling arm 41 or the hull 1 so that the center of the light spot coincides with the predetermined position of the sampling inlet of the sampling head 42 in the image coordinate system. Due to the absorption of near-infrared light by the oil film area, the light spot on the oil film will present a different shape or brightness than the surrounding water, ensuring that the sampling head 42 can accurately target the center area of ​​the oil film for sampling, thereby improving the representativeness of the sample.

[0035] Furthermore, the control unit is configured to perform spatiotemporal synchronization processing on the image data acquired by the dual-spectrum vision module to ensure that the two images are aligned at the pixel level and in terms of timestamps.

[0036] Furthermore, an attitude sensor is installed on the hull 1. The control unit is connected to the attitude sensor and configured to perform motion compensation on the image data collected by the visual recognition system 3 based on the attitude data of the hull 1 provided by the attitude sensor. The control unit receives the roll and pitch data of the hull 1 provided by the control unit and performs motion compensation on the images collected by the visual recognition system 3. Digital image stabilization technology is used to eliminate the image rotation and translation caused by the swaying of the hull 1, so as to ensure the stability and accuracy of image analysis.

[0037] Through the above steps, a near-infrared camera is used to capture the natural scintillation pattern of water in a specific wavelength band, and a scintillation intensity distribution map is generated to locate scintillation anomaly areas. The color anomaly areas identified by the RGB camera are correlated and matched with this scintillation suppression area, thereby eliminating optical interference such as specular reflection and shadows, and achieving high-confidence identification of floating oil. Combined with the near-infrared active illumination source integrated into the sampling head 42, sub-pixel-level precise positioning of sampling points is achieved through visual servo control. This reduces the false alarm rate in the identification stage and improves the accuracy in the sampling stage, solving the problem of poor reliability of traditional RGB vision methods under complex lighting and water surface fluctuations.

[0038] Example 2 Optionally, the present invention provides another embodiment in which a dual-spectrum vision module is used to identify and confirm oil slicks.

[0039] This embodiment of a machine vision-based oil spill identification and sampling vessel includes a hull 1, a propulsion system 2, a vision recognition system 3, a sampling system 4, and a control unit integrated inside the hull 1.

[0040] The propulsion system 2 adopts a conventional DC brushless motor to drive the propeller, and can be integrated with a servo motor or use dual-thrust differential steering to achieve flexible heading control.

[0041] The visual recognition system 3 includes a dual-spectrum visual module, which is fixed to the front of the hull 1 by a rigid bracket to obtain a wide forward field of view. The dual-spectrum visual module includes a first visual sensor 31 and a second visual sensor 32. In this embodiment, the first visual sensor 31 is an industrial-grade USB global shutter RGB camera to obtain clear color images. The second visual sensor 32 is a CMOS camera of the same specification that is sensitive to near-infrared light, with a narrowband interference filter with a center wavelength of 880nm and a half-width of 20nm installed in front of its lens. The optical axes of the two cameras are parallel as much as possible, and stereo calibration is performed by checkerboard calibration method to obtain the spatial transformation matrix between the two, so as to achieve pixel-level alignment of the images. The control unit ensures at the software level that the two image data acquired at the same time are stamped with the same timestamp when capturing images.

[0042] The hardware of the control unit is an industrial computer embedded in the hull 1, which is connected to the vision recognition system 3 via a USB interface and communicates with the navigation propulsion system 2 and the sampling system 4 via a serial port or CAN bus.

[0043] The image processing and control logic operating within the control unit is as follows: Step S1: Image Acquisition and Preprocessing Simultaneously capture RGB and near-infrared images; perform denoising and distortion correction on the two images, and use the aforementioned calibration matrix to reproject the two images onto the same virtual imaging plane to achieve pixel registration; Step S2: Parallel processing of initial screening of suspected areas and scintillation feature analysis S2a (RGB processing): The RGB image undergoes color space conversion (e.g., from RGB to HSV). Color segmentation is performed by setting threshold ranges on the H (hue) and S (saturation) channels that correspond to the characteristics of the oil film, resulting in a binarized image. Then, morphological operations (e.g., closing operations) are used to eliminate minor noise. After connected component analysis, each connected component is marked as a suspected oil spill region R_oi (i=1,2,...). S2b (Near-infrared processing): Generates a "scintillation intensity distribution map" for the registered near-infrared image I_nir; the specific method is as follows: define a sliding window (e.g., 15×15 pixels in size), and traverse the entire image I_nir with a certain step size (e.g., 5 pixels); for each pixel block P within the window, calculate the variance Var(P) of its pixel values; the formula for calculating the variance is:

[0044] in, This represents the total number of pixels within window P. Let j be the grayscale value of the j-th pixel. I_nir is the grayscale mean of all pixels within window P; the calculated variance is assigned to the center pixel of the window, and finally a grayscale map of the same size as I_nir is generated, namely the scintillation intensity distribution map F_map; in F_map, high variance values ​​(bright areas) represent normal water bodies with strong scintillation, and low variance values ​​(dark areas) represent potential oil slicker areas where scintillation is suppressed. Step S3: Feature matching and oil spill confirmation Each suspected oil slick region R_oi obtained in step S2a is mapped onto the scintillation intensity distribution map F_map generated in step S2b; the average variance of the corresponding pixel in F_map is calculated for each R_oi region; simultaneously, the average variance of the annular region surrounding R_oi (representing the surrounding background water) in F_map is calculated; a confidence index C is defined, where C is the difference between the latter and the former of the two average variance values; if C is greater than a preset positive threshold T_confidence, then the region R_oi is determined to be a real oil slick region. This determination logic is based on the fact that real oil slicks significantly suppress the surface scintillation of the water in its coverage area, resulting in its internal variance being much lower than that of the surrounding water. Step S4: Sampling Trigger Once an area is confirmed as a real oil slick, the control unit generates a sampling command to initiate the subsequent navigation and sampling process.

[0045] Example 3 Optionally, this embodiment further elaborates on the specific structure of the sampling system 4 and the method for achieving precise positioning sampling, based on embodiment 2.

[0046] The sampling system 4 in this embodiment includes a multi-joint sampling arm 41 driven by a stepper motor; a sampling head 42 is fixed to the end of the sampling arm 41 through a quick-connect mechanism; the sampling head 42 integrates a miniature water pump and a sampling tube; a ring-shaped near-infrared LED light panel is also embedded on the outer shell of the sampling head 42 as an active illumination source 43, and its emission band is strictly matched with the response band (880nm) of the second visual sensor 32. After the control unit confirms the oil slick area according to the procedure of Example 2 and controls the hull 1 to navigate above the area, the following precise positioning and sampling procedure is executed: Step S5: Sampling arm 41 extends and active illumination is turned on.

[0047] The control unit controls the sampling arm 41 to extend to a predetermined height above the water surface; then, the near-infrared LED light panel on the sampling head 42 is turned on to project a near-infrared light spot onto the water surface below. Step S6: Spot center positioning The second visual sensor 32 acquires near-infrared images containing the light spot in real time. Due to the absorption of near-infrared light by the floating oil, the light spot will exhibit different grayscale or shape characteristics in the oil film area and the clean water area, but this does not affect the center positioning. The control unit uses the grayscale centroid method to perform sub-pixel-level center positioning of the light spot. The specific steps are as follows: Define a region of interest (ROI) in the image that encompasses the entire light spot; Set a grayscale threshold, binarize the ROI, and initially separate the spot area; Within the binarized region, the precise center coordinates of the light spot are calculated using the gray-level centroid method. The calculation formula is: ,

[0048] Where M is the total number of pixels in the initial spot area. For the first grayscale value of each pixel. For the first The coordinates of each pixel in the image; Step S7: Servo Alignment and Sampling On the sampling head 42, the center of the sampling tube's inlet has a preset reference position in the image coordinate system of the second vision sensor 32. The control unit calculates the center of the light spot. relative to the reference position The deviation is reduced to near zero by controlling the micro-motion mechanism (piezoelectric ceramic or small servo motor) at the end of the sampling arm 41 or by slightly adjusting the navigation propulsion system 2; when the deviation is less than the set tolerance, the alignment is considered complete; the control unit then starts the micro water pump to collect water surface samples into the storage bottle through the sampling tube.

[0049] Example 4 Optionally, this embodiment adds an attitude sensor and motion compensation function to the basis of embodiment 2 or 3 to improve the stability and reliability of the system under wind and wave conditions.

[0050] In this embodiment, a high-precision inertial measurement unit (IMU), such as MPU-9250, is integrated inside the hull 1. The IMU measures the three-axis acceleration and three-axis angular velocity of the hull 1 in real time and sends the data to the control unit through the I2C or SPI interface.

[0051] The motion compensation algorithm running within the control unit is as follows: Attitude calculation: Based on the raw data from the IMU, the roll and pitch angles of hull 1 are calculated in real time using complementary filtering or Kalman filtering algorithms. Image stabilization: In the preprocessing stage of image processing (step S1), the calculated Roll and Pitch angles are used to perform reverse rotation and translation transformation on the RGB image and near-infrared image of the current frame to compensate for the image jitter caused by the swaying of the hull 1. Specifically, affine transformation or perspective transformation can be used to achieve this. This operation stabilizes the dynamically captured image sequence onto a virtual horizontal reference plane, greatly eliminating the interference caused by the movement of the ship 1 on the calculation of the "scintillation intensity distribution map" and the location of suspected areas, thus ensuring the stability of feature analysis. In addition, the control unit manages the image acquisition timing of the dual-spectrum vision module, using hardware triggering or precise software synchronization strategies to ensure high temporal consistency between the two image data streams and avoid matching errors caused by acquisition time differences.

[0052] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A machine vision-based oil slick identification and sampling vessel, characterized in that: include: The hull (1) is used for navigation on the water. A propulsion system (2) is installed on the hull (1) to provide power to the hull (1) and control the direction of navigation; A visual recognition system (3) is installed on the hull (1) to collect optical information of the water surface; A sampling system (4), installed on the hull (1), is used to collect water samples when oil slicks are detected; The control unit is communicatively connected to the navigation propulsion system (2), the vision recognition system (3), and the sampling system (4), respectively. The visual recognition system (3) includes a dual-spectrum visual module, which comprises: The first visual sensor (31) is configured to acquire first-band image data of the water surface; The second visual sensor (32) has a different spectral response band than the first visual sensor (31) and is configured to acquire second-band image data of the water surface; the second band includes the near-infrared band, which is sensitive to the optical properties of floating oil and has significant specular reflection of the water body. The control unit is configured as follows: Based on the second band image data, a distribution map is generated to characterize the intensity of natural optical scintillation on the water surface by analyzing the changes in pixel statistical features in local areas of the image. Based on the image data of the first band, a suspected oil slick area was identified; The suspected oil slick area is spatially matched with the area in the distribution map that shows a significantly lower optical scintillation intensity than the surrounding water body, and the oil slicker area is confirmed based on the matching results.

2. The water surface oil spill identification and sampling vessel based on machine vision according to claim 1, characterized in that: The first visual sensor (31) is an RGB color camera, and the second visual sensor (32) is a near-infrared camera equipped with a filter. The filter limits the second band to a narrow band with a center wavelength in the range of 850nm to 950nm.

3. A machine vision-based oil slick identification and sampling vessel according to claim 1 or 2, characterized in that: The control unit is configured to generate the distribution map by analyzing the changes in pixel statistical features in local regions of the image, specifically: Calculate the pixel value variance or high-frequency component energy of multiple predetermined local regions in the second band image data; Based on the variance or energy value, a scintillation intensity distribution map corresponding to the second band image data space is generated, wherein areas with high variance or energy values ​​are characterized as strong scintillation areas, and areas with low variance or energy values ​​are characterized as weak scintillation areas.

4. The water surface oil spill identification and sampling vessel based on machine vision according to claim 1, characterized in that: The control unit is configured to identify the oil spill area based on the matching result, and the specific determination logic is as follows: When a suspected oil slick area identified in the first band image data has a spatial overlap with a weak flickering area in the distribution map that exceeds a preset threshold, the area is determined to be a real oil slick area that needs to be sampled.

5. A machine vision-based oil slick identification and sampling vessel according to claim 1, characterized in that: The sampling system (4) includes: The sampling arm (41) is movably mounted on the hull (1). A sampling head (42) is located at the end of the sampling arm (41) and is used to collect water samples; The control unit is also configured to: after confirming the oil spill area, control the navigation propulsion system (2) to make the hull (1) sail above the oil spill area, and control the sampling arm (41) to extend so that the sampling head (42) reaches the predetermined sampling position.

6. A machine vision-based oil slick identification and sampling vessel according to claim 5, characterized in that: The sampling head (42) integrates an active illumination source (43), the spectral emission band of which matches the spectral response band of the second visual sensor (32).

7. A machine vision-based oil slick identification and sampling vessel according to claim 6, characterized in that: The control unit is further configured to perform a precise positioning sampling operation after the sampling head (42) reaches the predetermined sampling position, including: The active lighting source (43) is activated to project light spots onto the water surface below; An image containing the light spot is acquired through the second visual sensor (32); The center position of the light spot in the image is identified and located using an image processing algorithm; Adjust the position of the sampling arm (41) and / or the hull (1) so that the center of the light spot coincides with the sampling entrance of the sampling head (42) in the image coordinate system, and then start sampling.

8. A machine vision-based oil slick identification and sampling vessel according to claim 7, characterized in that: The image processing algorithm identifies and locates the center position of the light spot, specifically using the gray-scale centroid method to achieve sub-pixel level center positioning.

9. A machine vision-based oil slick identification and sampling vessel according to claim 1, characterized in that: The control unit is configured to perform spatiotemporal synchronization processing on the image data acquired by the dual-spectrum vision module.

10. A machine vision-based oil slick identification and sampling vessel according to claim 1, characterized in that: An attitude sensor is provided on the hull (1). The control unit is connected to the attitude sensor and is configured to perform motion compensation on the image data collected by the visual recognition system (3) based on the attitude data of the hull (1) provided by the attitude sensor.

Images