Intelligent identification method and system for benthic animals and storage medium

Abstract

This invention relates to the field of image recognition technology, and more particularly to an intelligent method, system, and storage medium for identifying benthic animals. The method includes the following steps: acquiring underwater image sets at different water depths in a target area using a pre-set underwater camera; determining the light attenuation at different water depths based on pixel grayscale variations in the underwater image sets at different depths, and correcting the underwater image sets at the target water depth based on the light attenuation at different depths to obtain a target underwater image set; performing minimum observation unit segmentation on the target underwater image set, and dividing the target underwater image set into candidate image regions for benthic animals based on the segmentation results; identifying target image regions for benthic animals based on edge feature variations in the candidate image regions, and binding the target image regions for benthic animals with the target water depth, thus generating a benthic animal water depth distribution report. This invention effectively reduces the impact of light, noise, and background interference, helping to improve the recognition accuracy and environmental adaptability of benthic animals.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method, system and storage medium for intelligent identification of benthic animals. Background Technology

[0002] Traditional benthic animal identification methods mainly rely on manual sampling and morphological analysis under a microscope. Although this method has a certain degree of reliability in terms of species classification accuracy, it suffers from problems such as long operation cycle, high labor intensity, low data processing efficiency, and subjective judgment error, making it difficult to meet the needs of large-scale ecological monitoring and real-time environmental assessment.

[0003] In recent years, the development of information technology and artificial intelligence has provided new technological approaches for biometrics. Some existing technologies attempt to automatically classify aquatic organisms using image recognition, pattern recognition, and machine learning methods, but several shortcomings remain in the field of benthic animal identification. On the one hand, existing methods are often limited by image quality, shooting angle, and lighting conditions when collecting data, resulting in limited ability to extract subtle morphological features of benthic animals. On the other hand, existing models generally rely on a large number of labeled samples for training, making it difficult to maintain high-accuracy identification in aquatic environments with scarce or complex samples. Summary of the Invention

[0004] Therefore, it is necessary for the present invention to provide a method, system and storage medium for intelligent identification of benthic animals in order to solve at least one of the above-mentioned technical problems.

[0005] To achieve the above objectives, a method for intelligent identification of benthic animals includes the following steps:

[0006] Step S1: Use a pre-set underwater camera to collect underwater image sets at different water depths in the target area;

[0007] Step S2: Determine the light attenuation at different water depths based on the pixel grayscale changes of underwater image sets at different water depths, and correct the underwater image sets at the target water depth based on the light attenuation at different water depths to obtain the target underwater image set.

[0008] Step S3: Perform minimum observation unit segmentation on the target underwater image set, and divide the benthic animal candidate image regions according to the segmentation results;

[0009] Step S4: Based on the changes in edge features of the candidate image regions of benthic animals, identify the target image regions of benthic animals, bind the target image regions of benthic animals with the target water depth, and generate a report on the water depth distribution of benthic animals.

[0010] The proposed intelligent benthic animal identification method achieves high-precision automatic identification of benthic animals in complex aquatic environments by introducing underwater multi-depth illumination attenuation correction, minimum observation unit segmentation, and target recognition technology based on edge structure features. This method dynamically compensates for the illumination attenuation of images at different depths, significantly improving the clarity and texture discernibility of underwater images. By constructing multi-scale observation units based on edge connectivity and gray-level gradient direction, it achieves adaptive extraction of benthic animal contour features. Furthermore, by combining a matching and recognition mechanism based on biological tissue and morphological statistical features in benthic animal growth rules, it effectively reduces the impact of illumination, noise, and background interference. Compared with traditional manual identification and sample-dependent deep learning methods, this method offers higher identification accuracy and stronger environmental adaptability, supporting real-time monitoring and automatic ecological status assessment of benthic communities.

[0011] Optionally, this application also provides a benthic animal intelligent identification system for performing the benthic animal intelligent identification method described above, the benthic animal intelligent identification system comprising:

[0012] The image acquisition module is used to acquire underwater image sets at different water depths in the target area using a preset underwater camera;

[0013] The image preprocessing module is used to determine the light attenuation at different water depths based on the pixel grayscale changes of underwater image sets at different water depths, and to correct the underwater image set at the target water depth based on the light attenuation at different water depths, so as to obtain the target underwater image set.

[0014] The observation unit segmentation module is used to perform minimum observation unit segmentation on the target underwater image set and divide the benthic animal candidate image regions according to the segmentation results;

[0015] The benthic animal identification module is used to identify the target image region of benthic animals based on the changes in the edge features of the candidate image region of benthic animals, and to bind the target image region of benthic animals with the target water depth, and to generate a benthic animal water depth distribution report.

[0016] The present invention discloses an intelligent benthic animal identification system capable of implementing any of the intelligent benthic animal identification methods of the present invention. This system serves as a medium for coordinating the operations and signal transmission between various modules to complete the intelligent benthic animal identification method. The internal modules collaborate with each other, thereby effectively reducing the impact of light, noise, and background interference. Compared with traditional manual identification and sample-dependent deep learning methods, this method offers higher identification accuracy and stronger environmental adaptability, supporting real-time monitoring and automatic ecological status assessment of benthic communities.

[0017] Optionally, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the benthic animal intelligent identification method as described above. Attached Figure Description

[0018] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0019] Figure 1 This is a schematic diagram of the steps of the intelligent identification method for benthic animals of the present invention;

[0020] Figure 2 These are partial images from the original underwater image set in this embodiment of the invention;

[0021] Figure 3 These are partial images from the corrected underwater image set in this embodiment of the invention;

[0022] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0023] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0024] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0025] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0026] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a method for intelligent identification of benthic animals, the method comprising the following steps:

[0027] Step S1: Use a pre-set underwater camera to collect underwater image sets at different water depths in the target area;

[0028] In this embodiment, if the target area is a nearshore shallow sea area, underwater images are simultaneously acquired using an array of underwater cameras deployed at different water depths (0.5m, 1m, 5m, 10m, 20m) under calm conditions. Each camera uses a wide dynamic range image sensor with infrared compensation, with an exposure time set to 0.03s and a frame rate of 20fps to ensure real-time recording of changes in illumination. During the acquisition process, the control unit automatically marks the shooting depth based on the depth calibration sensor unit and performs time synchronization. The acquired multi-depth images are uniformly formatted to generate an initial underwater image set, providing depth stratification for subsequent illumination correction.

[0029] Step S2: Determine the light attenuation at different water depths based on the pixel grayscale changes of underwater image sets at different water depths, and correct the underwater image sets at the target water depth based on the light attenuation at different water depths to obtain the target underwater image set.

[0030] In a further embodiment, the difference in mean grayscale values ​​between adjacent water depths at the same sampling time is statistically analyzed to obtain a grayscale attenuation value sequence. Regression correction is performed on the grayscale attenuation values ​​based on turbidity, suspended particle concentration, and temperature data collected by a water quality sensor, and a second-order polynomial is used to fit the light attenuation curve. Using the mean grayscale value at the water surface as the baseline grayscale value, environmental parameter compensation corrections are applied to abrupt change points to obtain the light attenuation at different water depths. Then, combined with suitable spring water depth parameters from a benthic animal characteristic database, the light attenuation is converted into a light compensation coefficient, and pixel-by-pixel grayscale values ​​of the target water depth image are corrected to generate a target underwater image set.

[0031] Step S3: Perform minimum observation unit segmentation on the target underwater image set, and divide the benthic animal candidate image regions according to the segmentation results;

[0032] In a further embodiment, Canny edge detection is performed on the image to obtain the edge pixel distribution. Closed contour regions are determined based on the connectivity of edge pixels, and the minimum contour pixel area is calculated. When the minimum area is less than 0.2% of the total pixels in the image, it is used as the segmentation step size. Region segmentation is performed according to this segmentation step size to obtain several minimum observation units. The edge pixel density of each minimum observation unit is detected, and units with a density higher than 0.35 are selected as candidate units. Then, they are aggregated based on the connectivity between units to generate a preliminary candidate region set. The correlation of local texture orientation within the candidate regions is statistically analyzed. When the correlation is >0.75 and the gray-level mean square error is within the benthic animal tissue characteristic range, it is marked as a benthic animal candidate image region.

[0033] Step S4: Based on the changes in edge features of the candidate image regions of benthic animals, identify the target image regions of benthic animals, bind the target image regions of benthic animals with the target water depth, and generate a report on the water depth distribution of benthic animals.

[0034] In a further embodiment, the grayscale gradient direction angle of the edge pixels of the candidate benthic animal image region is identified, and the gradient direction concentration is statistically analyzed within a 5×5 pixel neighborhood. When the concentration exhibits a unimodal or multimodal distribution (selected according to the characteristics of different benthic animals), the neighborhood is determined to be a continuous edge region. Subsequently, the gradient intensity change rate curve of the continuous edge region is calculated, adjacent intensity reversal points are identified, and connected to form an edge closure structure. The system matches the closure structure with the shape statistical features (including roundness 0.7-0.9 and aspect ratio 1.2-1.6) in the benthic animal growth rules. When the edge direction change is smooth and the roundness is within a similar range, the closure structure is determined to be the target image region of the benthic animal, and it is bound to the corresponding water depth label to generate a benthic animal water depth distribution report.

[0035] Optionally, determining the light attenuation at different water depths in step S2 includes:

[0036] Based on underwater image sets at different water depths, the difference in the mean gray value of pixels at adjacent water depths within the same sampling time window is statistically analyzed.

[0037] In this embodiment, if multiple sets of underwater image data at different depths are detected within the same time window, the image sampling depth is divided into depth intervals from 0.5m to 3.0m using a depth calibration sensing unit. First, the original color image is converted to grayscale. Then, pixel grayscale statistics are performed on the image at each depth interval, calculating the difference in the mean grayscale value of pixels at the same spatial location between adjacent depths. Each grayscale mean is taken as the local average value of a 5×5 pixel neighborhood.

[0038] The grayscale attenuation value for different water depths is determined based on the difference in the average grayscale value of pixels between adjacent water depths. Then, environmental impact correction is performed on the grayscale attenuation value based on the pre-acquired environmental sensing data to obtain the illumination attenuation amount for different water depths.

[0039] In a further embodiment, by comparison The trend of grayscale variation with depth was observed, resulting in a grayscale attenuation curve. This attenuation curve preliminarily characterizes the depth-dependent loss of light intensity. Based on this, and using pre-acquired environmental sensor data including water turbidity, suspended particle concentration, and water dispersion coefficient, a secondary correction was performed on the grayscale attenuation value. For example, if the turbidity is higher than 10 NTU, the grayscale attenuation is increased by a factor of 1.2; if the turbidity is lower than 5 NTU, the correction is applied by a factor of 0.9.

[0040] It is worth noting that the pre-acquired environmental sensing data is collected by a multi-parameter water quality environmental sensing unit deployed in the water body of the sampling area. This sensing unit is typically deployed at sampling depth points synchronized with the underwater camera and includes sensors such as light intensity sensors, turbidity sensors, temperature sensors, dissolved oxygen probes, and chlorophyll fluorescence sensors to monitor physical and chemical parameters related to underwater imaging quality in real time. The environmental sensing unit records each sensing parameter at a sampling frequency of 0.5Hz and performs time synchronization calibration before and after each image acquisition to ensure consistency between image data and environmental data on the time axis. Subsequently, the continuously sampled multi-dimensional environmental data undergoes time-series interpolation and noise filtering to form an environmental sensing dataset corresponding one-to-one with the underwater image acquisition depth.

[0041] Figure 2 and Figure 3 These are the original underwater image set and the corrected underwater image set in the embodiments of the present invention; the images contain three different water depths, namely 5m, 10m and 20m. Figure 2 and Figure 3 In order to reflect samples collected at different locations or repeatedly at the same water depth, two different image sets were selected at the same water depth (5m). The number of image sets at each water depth can be adjusted according to different application scenarios, and this application does not impose any restrictions on this.

[0042] Optionally, implementing environmental impact corrections includes:

[0043] The underwater image set at different water depths, which falls within the range of underwater acquisition depth from zero to near the surface, is used as the reference water surface image set, and the average gray value of the pixels in this reference water surface image set is used as the gray value reference.

[0044] In this embodiment, if the underwater camera is detected to have completed continuous acquisition within a depth range of 0m to 3m, the group of image frames acquired at depths of 0m to 0.3m is designated as the reference surface image set. The mean grayscale value is calculated by performing global grayscale statistics on the pixels of each frame within this image set. , as the grayscale reference value.

[0045] Crucially, the zero point refers to the reference depth at the water surface, specifically 0m, where the water just touches the air. This is where illumination is strongest and can be considered the initial reference point for calculating underwater image illumination attenuation. The near-surface layer comprises shallow water areas close to the surface where illumination has not yet significantly attenuated, typically between 0m and 1m or 0m and 2m (the specific range depends on water transparency and sampling resolution). Within this range, water turbidity, suspended particle concentration, and temperature variations are minimal, and illumination intensity is almost identical to that of incident light at the water surface, thus serving as a stable reference layer for illumination attenuation correction.

[0046] Based on the grayscale reference value, abrupt decay points in the grayscale decay values ​​of different water depths are screened, and environmental sensing parameters of the water depth adjacent to the abrupt decay point are extracted from the environmental sensing data.

[0047] In a further embodiment, based on the grayscale reference value, a layer-by-layer differential analysis is performed on the grayscale attenuation sequence of each water depth image. If the grayscale attenuation rate of any depth layer increases by more than 30% compared to the previous layer, it is determined to be a sudden attenuation point. Subsequently, turbidity, dispersion coefficient, and water temperature parameters corresponding to the depths adjacent to the sudden attenuation point are extracted from the environmental sensing data.

[0048] The grayscale attenuation values ​​adjacent to the abrupt attenuation point are regressed and correlated with the environmental sensing parameters of the adjacent water depth to determine the associated environmental sensing parameters.

[0049] The grayscale attenuation value at the abrupt attenuation point is corrected based on the associated environmental sensing parameters, and the grayscale attenuation value correction result is converted into light attenuation amount based on the regression relationship between grayscale attenuation value and light intensity in the regression correlation results, thereby obtaining the light attenuation amount at different water depths.

[0050] In a further embodiment, to more accurately determine the environmental correlation of abrupt decay points, a multiple linear regression model is constructed based on the extracted grayscale decay sequence and environmental sensor data. This model uses the grayscale decay values ​​of adjacent layers... The corresponding turbidity parameter T, suspended particle concentration C, and water dispersion coefficient S are used as input variables. The least squares method is used to calculate the regression coefficients of each environmental parameter, forming a regression equation. This regression equation is then cross-validated. If the coefficient of determination of the model after cross-validation is... Furthermore, if the residual standard deviation is less than 0.05, the regression relationship between grayscale attenuation and environmental parameters is considered stable. Based on this, environmental parameters at abrupt attenuation points are input into the model to calculate the theoretical predicted grayscale attenuation value. When the measured grayscale attenuation value... Compared with the predicted value The difference exceeds the preset deviation threshold When the deviation is 0.1, the mutation point is considered to be affected by local environmental disturbances. In this case, a reverse correction operation is performed: adjustments are made based on the direction of the deviation. The correction amount is ,in A correction factor (ranging from 0.6 to 0.8) is used to balance the difference in environmental impact between measured and predicted values. After correction, based on the established empirical regression relationship between grayscale attenuation values ​​and illumination intensity (obtained by fitting measured illumination values ​​from the baseline water surface image set with corrected grayscale values ​​at each depth layer), ≥0.92), the corrected grayscale attenuation value The light attenuation is converted to L_d. This results in light attenuation at different water depths.

[0051] Optionally, the underwater image set for correcting the target water depth in step S2 includes:

[0052] By comparing environmental sensor data with a pre-set benthic animal characteristic database, the suitable water depth for benthic animals in the current season is determined, and this suitable water depth is used as the target water depth.

[0053] In this embodiment, the suitable water depth range, seasonal growth environment parameters, and benthic animal growth rules of each species in the benthic animal characteristic database are indexed. Then, based on the currently collected environmental sensor data such as temperature, dissolved oxygen, turbidity, and light intensity, the environmental parameters of each water depth are matched with the benthic animal growth rules layer by layer. If the multidimensional environmental parameters of a certain water depth all fall within the suitable range specified by the benthic animal characteristic database, then the water depth is determined to be the suitable water depth for the target species in the current season, and the water depth is recorded as the target water depth.

[0054] Extract the underwater image set for the target water depth from the underwater image set at different water depths, and extract the grayscale distribution of the pixels in the underwater image set for the target water depth;

[0055] In a further embodiment, an underwater image set (in this case, a color image) of the target water depth is extracted from an underwater image set of different water depths. Then, the gray value distribution of each pixel in the image is statistically analyzed, the mean, variance and gray value histogram are calculated, and the gray value distribution of the target water depth is generated.

[0056] The grayscale reference value is converted into the light intensity reference value, and the light compensation coefficient of the target water depth is calculated based on the light intensity reference value and the light attenuation of the target water depth.

[0057] In a further embodiment, a linear mapping relationship is then established based on the grayscale mean of the shallow reference water surface image set and the corresponding water surface illumination intensity obtained through the sensing unit: ;in Represents the grayscale value of any pixel. This represents the mapped light intensity value. This is the gain coefficient. This is the bias term. Gain coefficient. and bias The light intensity baseline was determined by fitting the correspondence between the mean grayscale value of the shallow reference image and the measured light intensity using the least squares method. The grayscale values ​​were then linearly mapped to the light intensity values ​​to form the light intensity baseline. Subsequently, based on the light attenuation at the target water depth, the light compensation coefficient was calculated using a linear light attenuation model. : ,in As a reference light intensity, The light attenuation at the target water depth.

[0058] By using the illumination compensation coefficient, the grayscale values ​​of each pixel in the underwater image set of the target water depth are corrected to obtain the target underwater image set.

[0059] In a further embodiment, for each pixel grayscale Perform linear correction: The corrected pixel values ​​are then truncated to grayscale (0–255) to maintain the display range while preserving the original texture features of the image, resulting in a set of target underwater images with balanced illumination that can be used for subsequent identification of benthic animals.

[0060] Alternatively, methods for obtaining a benthic animal characteristic database include:

[0061] To obtain regional sampling data under different seasons in order to determine the environmental characteristic parameters of regional water bodies in each season;

[0062] In this embodiment, multiple sampling stations are set up in different seasons in the target water area, and water quality sensing units are used to collect parameters such as water temperature, dissolved oxygen, pH value, salinity, flow rate and turbidity. The data collection cycle of each sampling point is 30 minutes, and continuous sampling is carried out for 72 hours to obtain a stable seasonal environmental characteristic parameter matrix.

[0063] The environmental characteristic parameters of each season are compared with the preset benthic animal growth rules. If the environmental characteristic parameters of any season are within the suitable growth range of the benthic animal growth rules, then the environmental characteristic parameters of that season are determined to be the suitable growth environment for benthic animals. The suitable growth environments of benthic animals in each season are summarized to construct a benthic animal characteristic database.

[0064] In a further embodiment, the collected environmental characteristic parameters for each season are compared with preset benthic animal growth rules. These rules include parameters such as suitable water temperature ranges, lower limits of dissolved oxygen concentration, salinity ranges, and flow velocity limitations for different species. By matching each parameter individually, it is determined whether the environmental parameters at each sampling point fall within a suitable growth range. If the matching ratio exceeds 80%, the seasonal environment corresponding to that sampling point is considered a suitable growth environment for benthic animals. Finally, the suitable environmental parameters for all seasons and each sampling point are compiled to form a benthic animal characteristic database. Each record includes a season identifier, sampling water depth, geographical coordinates, and the matched environmental parameter range, and is stored as a structured database.

[0065] Of particular importance is that the growth rules of benthic animals include the suitable growth range of each benthic animal, the range of biological tissue characteristics at different growth stages, and the morphological statistical characteristics at different growth stages.

[0066] In this embodiment, environmental adaptation characteristics of different species at different growth stages are obtained through field surveys and laboratory culture. For example, suitable ranges for environmental parameters such as water temperature, dissolved oxygen, salinity, and flow velocity are recorded for juveniles, subadults, and adults, as well as physiological characteristic data such as the density, thickness, and flexibility of biological tissues at each stage. These data are standardized and statistically processed to obtain biological tissue characteristic ranges. Secondly, morphological data of benthic animals at different growth stages are obtained through image analysis or 3D scanning, including length, width, thickness, and the roundness, flatness, and symmetry of the shell or body outline. Statistical methods are used to calculate shape statistical characteristics such as mean and variance, forming morphological characteristic models for different growth stages. Subsequently, the suitable growth range, biological tissue characteristic ranges, and shape statistical characteristics are integrated to establish a benthic animal growth rule library, which is used to compare with sampled environmental parameters or match with image features, thereby assisting in determining whether the target water depth or candidate area meets the growth conditions of benthic animals.

[0067] Optionally, the minimum observation unit segmentation in step S3 includes:

[0068] Edge detection is performed on the target underwater image set, and the contour closure region is determined based on the connectivity of edge pixels in the edge detection results;

[0069] In this embodiment, the original color image is converted to a grayscale image to remove the influence of color information on edge detection. Then, the Canny operator is used to extract edges, with a low threshold of 0.05 and a high threshold of 0.15, and the image is Gaussian smoothed. Subsequently, the gradient magnitude of each pixel is calculated, and non-maximum suppression is applied to the gradient magnitude. Then, a double-threshold algorithm is used to connect the edges in the high-threshold image to form a contour. When reaching the endpoint of the contour, a connectable edge is found in the low-threshold image to determine the true edge pixels. For the extracted edge pixels, 8-neighborhood connectivity is used to determine whether adjacent pixels are continuous, forming a preliminary contour closure region, and isolated pixels are removed.

[0070] Calculate the minimum pixel area of ​​the contour-closed region, and adjust the scale of the minimum pixel area according to the resolution of the target underwater image set to obtain the minimum observation unit segmentation step size;

[0071] In a further embodiment, when calculating the minimum pixel area for a closed region, the total number of pixels within each closed region can be counted. Assuming a resolution of 1024×1024 pixels, the minimum pixel area is set to 25. The area of ​​the smallest pixel is scaled proportionally to the image resolution (e.g., 1024×1024 pixels) to ensure that the smallest observation unit maintains a scale of approximately 5×5 pixels in high-resolution images, thus preserving detail information. This value also serves as the segmentation step size for the smallest observation unit, controlling the accuracy of subsequent image segmentation.

[0072] Image segmentation is performed on the underwater image set of the target using the minimum observation unit segmentation step size, resulting in several minimum observation units.

[0073] In a further embodiment, the image is divided into grids using the obtained minimum observation unit segmentation step size, cutting the entire image into several square grids, each grid corresponding to a minimum observation unit, and the step size is set to 5~10 pixels to balance resolution and detail preservation.

[0074] Optionally, step S3, which involves segmenting candidate image regions for benthic animals, includes:

[0075] The smallest observation unit whose edge pixel density is higher than a preset density threshold is selected as the candidate observation unit.

[0076] In this embodiment, the proportion of edge pixels to total pixels within each smallest observation unit is calculated. If this proportion is higher than a preset density threshold of 0.35, the unit is marked as a candidate observation unit. Edge pixels are derived from pixels with gradient magnitudes greater than 0.1 extracted by the aforementioned Canny operator, while retaining gradient direction information for subsequent analysis.

[0077] Based on the connectivity of each smallest observation unit, candidate observation units are aggregated to generate a preliminary candidate region set;

[0078] In a further embodiment, merging is performed based on the 8-neighborhood connectivity between each unit. If the connectivity of edge pixels between adjacent units is greater than 0.5, they are aggregated into the same region. Each aggregated region stores the edge pixel matrix, grayscale matrix, and unit index information. At the same time, the minimum size of the aggregated region can be limited to no less than 3×3 minimum observation units to avoid generating candidate regions that are too small.

[0079] The local texture direction correlation and gray-level mean square difference of adjacent units in the preliminary candidate region set are statistically analyzed. If the local texture direction correlation exceeds the preset correlation threshold and the gray-level mean square difference is within the biological tissue feature range in the growth rules of benthic animals, then the adjacent unit is marked as a candidate image region of benthic animals.

[0080] In a further embodiment, the local texture direction correlation and gray-level mean square error are statistically analyzed for each adjacent smallest observation unit within the preliminary candidate region. The local texture direction correlation is obtained by calculating the cosine similarity of the gradient directions within the unit. If the correlation exceeds 0.85 and the gray-level mean square error is within the biological tissue characteristic range of 0.02 to 0.08, then the unit is marked as a candidate image region for benthic animals.

[0081] Of particular importance are the methods for obtaining the connectivity of each smallest observation unit, including:

[0082] The connectivity of adjacent edge pixels within the smallest observation unit is determined based on the connectivity of edge pixels.

[0083] In this embodiment, an 8-neighborhood scan is performed on the edge pixels within a cell. If two edge pixels are connected within their 8-neighborhoods, they are considered connected. To improve robustness, a gradient magnitude threshold of 0.1 is set, and only pixels with a magnitude greater than this threshold participate in the connectivity determination. This method generates connected subsets within the cell, and the connectivity of each subset is calculated, which is the ratio of the number of connected pixels to the total number of edge pixels in the cell.

[0084] The connectivity of boundary pixels between adjacent minimum observation units is determined by using the connectivity of adjacent edge pixels within the minimum observation unit, thus obtaining the connectivity of adjacent minimum observation units.

[0085] In a further embodiment, the boundary region of pixels on the boundary lines of two adjacent units is selected, and the proportion of connected subset pixels in this region to the total number of boundary pixels is calculated. If the proportion is greater than 0.5, the boundaries of the two smallest observation units are determined to be connected, thereby establishing a connectivity relationship between adjacent smallest observation units.

[0086] Optionally, identifying the benthic animal target image region in step S4 includes:

[0087] Identify the gray-level gradient direction angle of each edge pixel within the candidate image region of benthic animals, and calculate the gradient direction concentration of the gray-level gradient direction angle within the same local neighborhood.

[0088] In this embodiment, the Sobel or Prewitt operator is applied to the candidate image region of benthic animals to calculate the gradient components of each edge pixel in the X and Y directions. The pixel grayscale gradient direction angle is then calculated from the gradient components. The orientation angles are discretized into 12 orientation intervals, each covering 30°. Subsequently, the distribution of gray-level gradient orientation angles within each local neighborhood (e.g., a 5×5 pixel block) is statistically analyzed, and an orientation concentration index (with a concentration threshold of 0.7) is calculated to determine whether the orientations are highly concentrated.

[0089] When the gradient direction concentration of the gray-level gradient direction angle in any local neighborhood exhibits a single-peak or double-peak characteristic, the local neighborhood is identified as a continuous edge region.

[0090] In a further embodiment, when the concentration of gray-level gradient direction angle within any local neighborhood is greater than 0.7 and the distribution exhibits a unimodal or bimodal pattern, the local neighborhood is marked as a continuous edge region. This marking is used for subsequent edge closure structure identification, ensuring that the edge region has directional continuity and structural integrity.

[0091] Calculate the rate of change curve of edge intensity along the gray-scale gradient direction angle within a continuous edge region, identify the intensity reversal point of the rate of change curve, and connect adjacent intensity reversal points to obtain the edge closed structure;

[0092] In a further embodiment, when calculating the rate of change curve of edge intensity along the gray-scale gradient direction angle within a continuous edge region, the local pixel intensity difference method is used to calculate the intensity change rate of each pixel along the gradient direction. Points with an absolute value of the rate of change greater than 0.05 are selected as intensity reversal points, and adjacent reversal points are connected sequentially to form a closed edge structure outline.

[0093] The closed edge structure is matched with the shape statistical features of benthic animals in the growth rules of benthic animals. If the edge direction of the closed edge structure changes smoothly and the roundness is within the similar range of the shape statistical features, then the closed edge structure is taken as the target image region of the benthic animal.

[0094] In a further embodiment, when matching closed edge structures with the statistical features of benthic animal shapes, the roundness of the contour (0.7~1.0 is the roundness similarity range), the smoothness of the edge direction (change angle ≤15° / pixel) and the area ratio are extracted and compared with the statistical features of different growth stages in the growth rules of benthic animals. If all meet the similarity conditions, the closed structure is determined to be the target image region of the benthic animal, and its spatial location and target water depth information are recorded.

[0095] Optionally, this application also provides a benthic animal intelligent identification system for performing the benthic animal intelligent identification method described above, the benthic animal intelligent identification system comprising:

[0096] The image acquisition module is used to acquire underwater image sets at different water depths in the target area using a preset underwater camera;

[0097] The image preprocessing module is used to determine the light attenuation at different water depths based on the pixel grayscale changes of underwater image sets at different water depths, and to correct the underwater image set at the target water depth based on the light attenuation at different water depths, so as to obtain the target underwater image set.

[0098] The observation unit segmentation module is used to perform minimum observation unit segmentation on the target underwater image set and divide the benthic animal candidate image regions according to the segmentation results;

[0099] The benthic animal identification module is used to identify the target image region of benthic animals based on the changes in the edge features of the candidate image region of benthic animals, and to bind the target image region of benthic animals with the target water depth, and to generate a benthic animal water depth distribution report.

[0100] Optionally, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the benthic animal intelligent identification method as described above.

[0101] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0102] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for intelligent identification of benthic animals, characterized in that, Includes the following steps: Step S1: Use a pre-set underwater camera to collect underwater image sets at different water depths in the target area; Step S2: Determine the light attenuation at different water depths based on the pixel grayscale changes of underwater image sets at different water depths, and correct the underwater image sets at the target water depth based on the light attenuation at different water depths to obtain the target underwater image set. Step S2, determining the light attenuation at different water depths, includes: Based on underwater image sets at different water depths, the difference in the mean gray value of pixels at adjacent water depths within the same sampling time window is statistically analyzed. The grayscale attenuation value for different water depths is determined based on the difference in the average grayscale value of pixels between adjacent water depths, and environmental impact correction is performed on the grayscale attenuation value based on the pre-acquired environmental sensing data to obtain the light attenuation amount for different water depths. The implementation of environmental impact modifications includes: The underwater image set at different water depths, which falls within the range of the underwater acquisition depth from zero to the near-surface layer, is used as the reference water surface image set, and the average gray value of the pixels in this reference water surface image set is used as the gray level reference value. Based on the grayscale reference value, abrupt decay points in the grayscale decay values ​​of different water depths are screened, and environmental sensing parameters of the water depth adjacent to the abrupt decay point are extracted from the environmental sensing data. The grayscale attenuation values ​​adjacent to the abrupt attenuation point are regressed and correlated with the environmental sensing parameters of the adjacent water depth to determine the associated environmental sensing parameters. The grayscale attenuation value at the abrupt attenuation point is corrected based on the associated environmental sensing parameters, and the grayscale attenuation value correction result is converted into light attenuation amount based on the regression relationship between grayscale attenuation value and light intensity in the regression correlation results, thereby obtaining the light attenuation amount at different water depths. Step S3: Perform minimum observation unit segmentation on the target underwater image set, and divide the benthic animal candidate image regions according to the segmentation results; Step S4: Based on the changes in edge features of the candidate image regions of benthic animals, identify the target image regions of benthic animals, bind the target image regions of benthic animals with the target water depth, and generate a report on the water depth distribution of benthic animals.

2. The intelligent identification method for benthic animals according to claim 1, characterized in that, The underwater image set for correcting the target water depth in step S2 includes: By comparing environmental sensor data with a pre-set benthic animal characteristic database, the suitable water depth for benthic animals in the current season is determined, and this suitable water depth is used as the target water depth. Extract the underwater image set for the target water depth from the underwater image set at different water depths, and extract the grayscale distribution of the pixels in the underwater image set for the target water depth; The grayscale reference value is converted into the light intensity reference value, and the light compensation coefficient of the target water depth is calculated based on the light intensity reference value and the light attenuation of the target water depth. By using the illumination compensation coefficient, the grayscale values ​​of each pixel in the underwater image set of the target water depth are corrected to obtain the target underwater image set.

3. The intelligent identification method for benthic animals according to claim 1, characterized in that, Methods for obtaining benthic animal characteristic databases include: To obtain regional sampling data under different seasons in order to determine the environmental characteristic parameters of regional water bodies in each season; The environmental characteristic parameters of each season are compared with the preset benthic animal growth rules. If the environmental characteristic parameters of any season are within the suitable growth range of the benthic animal growth rules, then the environmental characteristic parameters of that season are determined to be the suitable growth environment for benthic animals. The suitable growth environments of benthic animals in each season are summarized to construct a benthic animal characteristic database.

4. The intelligent identification method for benthic animals according to claim 1, characterized in that, The smallest observation unit segmentation in step S3 includes: Edge detection is performed on the target underwater image set, and the contour closure region is determined based on the connectivity of edge pixels in the edge detection results; Calculate the minimum pixel area of ​​the contour-closed region, and adjust the scale of the minimum pixel area according to the resolution of the target underwater image set to obtain the minimum observation unit segmentation step size; Image segmentation is performed on the underwater image set of the target using the minimum observation unit segmentation step size, resulting in several minimum observation units.

5. The intelligent identification method for benthic animals according to claim 1, characterized in that, Step S3, which involves segmenting candidate image regions for benthic animals, includes: The smallest observation unit whose edge pixel density is higher than a preset density threshold is selected as the candidate observation unit. Based on the connectivity of each smallest observation unit, candidate observation units are aggregated to generate a preliminary candidate region set; The local texture direction correlation and gray-level mean square difference of adjacent units in the preliminary candidate region set are statistically analyzed. If the local texture direction correlation exceeds the preset correlation threshold and the gray-level mean square difference is within the biological tissue feature range in the growth rules of benthic animals, then the adjacent unit is marked as a candidate image region of benthic animals.

6. The intelligent identification method for benthic animals according to claim 1, characterized in that, Step S4 involves identifying the target image region for benthic animals, including: Identify the gray-level gradient direction angle of each edge pixel within the candidate image region of benthic animals, and calculate the gradient direction concentration of the gray-level gradient direction angle within the same local neighborhood. When the gradient direction concentration of the gray-level gradient direction angle in any local neighborhood exhibits a single-peak or double-peak characteristic, the local neighborhood is identified as a continuous edge region. Calculate the rate of change curve of edge intensity along the gray-scale gradient direction angle within a continuous edge region, identify the intensity reversal point of the rate of change curve, and connect adjacent intensity reversal points to obtain the edge closed structure; The closed edge structure is matched with the shape statistical features of benthic animals in the growth rules of benthic animals. If the edge direction of the closed edge structure changes smoothly and the roundness is within the similar range of the shape statistical features, then the closed edge structure is taken as the target image region of the benthic animal.

7. A benthic animal intelligent identification system, characterized in that, For performing the benthic animal intelligent identification method as described in claim 1, the benthic animal intelligent identification system comprises: The image acquisition module is used to acquire underwater image sets at different water depths in the target area using a preset underwater camera; The image preprocessing module is used to determine the light attenuation at different water depths based on the pixel grayscale changes of underwater image sets at different water depths, and to correct the underwater image sets at the target water depth based on the light attenuation at different water depths, so as to obtain the target underwater image set. The observation unit segmentation module is used to perform minimum observation unit segmentation on the target underwater image set and divide the benthic animal candidate image regions according to the segmentation results; The benthic animal identification module is used to identify the target image region of benthic animals based on the changes in the edge features of the candidate image region of benthic animals, and to bind the target image region of benthic animals with the target water depth, and to generate a benthic animal water depth distribution report.

8. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed, implements the benthic animal intelligent identification method as described in any one of claims 1-6.

Images