Intelligent benthonic animal identification method and system and storage medium

Abstract

The invention relates to the technical field of image recognition, in particular to an intelligent recognition method and system for benthonic animals and a storage medium. The method comprises the following steps: acquiring underwater image sets of different water depths of a target area by using a preset underwater camera; determining the illumination attenuation of different water depths according to the pixel gray level change conditions of the underwater image sets of different water depths, and correcting the underwater image set of the target water depth according to the illumination attenuation of different water depths to obtain a target underwater image set; performing minimum observation unit segmentation on the target underwater image set, and dividing benthonic animal candidate image areas according to a segmentation result; and according to the edge feature change condition of the benthonic animal candidate image region, identifying a benthonic animal target image region, binding the benthonic animal target image region with a target water depth, and summarizing to generate a benthonic animal water depth distribution report. According to the method, the influence of illumination, noise and background interference is effectively reduced, and the recognition precision and environmental adaptability of the benthonic animals are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image recognition, and in particular to a benthic animal intelligent recognition method and system and a storage medium. BACKGROUND

[0002] Traditional benthic animal recognition methods mainly rely on manual sampling and morphological analysis under a microscope. Although these methods have certain reliability in species classification accuracy, they have problems such as long operation period, high labor intensity, low data processing efficiency, and subjective determination errors, and are 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 technology has provided new technical approaches for biological recognition. Some existing technologies attempt to use image recognition, pattern recognition, and machine learning methods to automatically classify aquatic organisms, but there are still some deficiencies in the field of benthic animal recognition. On the one hand, existing methods are often limited by image quality, shooting angle, and lighting conditions when collecting data, and have limited ability to extract the small morphological features of benthic animals. On the other hand, existing models generally rely on a large number of labeled samples for training, and it is difficult to maintain high-precision recognition in water environments with scarce samples or complex diversity. SUMMARY

[0004] Therefore, it is necessary to provide a benthic animal intelligent recognition method, system, and storage medium to solve at least one of the above technical problems.

[0005] To achieve the above-mentioned purpose, a benthic animal intelligent recognition method includes the following steps: Step S1: using a pre-set underwater camera to collect an underwater image set of different water depths in a target area; Step S2: determining the light attenuation amount of different water depths according to the pixel gray level change of the underwater image set of different water depths, and correcting the underwater image set of the target water depth according to the light attenuation amount of different water depths to obtain a target underwater image set; Step S3: performing minimum observation unit segmentation on the target underwater image set, and dividing the benthic animal candidate image region according to the segmentation result; Step S4: identifying the benthic animal target image region according to the edge feature change of the benthic animal candidate image region, and binding the benthic animal target image region with the target water depth to generate a benthic animal water depth distribution report.

[0006] The intelligent benthic animal recognition method provided in the application realizes high-precision automatic recognition of benthic animals in a complex water environment by introducing underwater multi-depth light attenuation correction, minimum observation unit segmentation and target recognition technology based on edge structure features. The method uses the light attenuation amount of images at different water depths to dynamically compensate the target image, significantly improving the definition and texture distinguishability of underwater images. By constructing a multi-scale observation unit based on edge connectivity and gray gradient direction, the adaptive extraction of the contour features of benthic animals is realized. In combination with the matching recognition mechanism of biological tissue and morphological statistical features in the growth rule of benthic animals, the influence of light, noise and background interference is effectively reduced. Compared with traditional manual recognition and sample-dependent deep learning methods, the method has higher recognition accuracy and stronger environmental adaptability, and can support real-time monitoring and automatic ecological state evaluation of benthic communities.

[0007] Optionally, the application also provides an intelligent benthic animal recognition system for executing the intelligent benthic animal recognition method as described above, which comprises: an image acquisition module for acquiring a set of underwater images of a target region at different water depths by using a preset underwater camera; an image preprocessing module for determining the light attenuation amount at different water depths according to the pixel gray level variation of the set of underwater images at different water depths, and correcting the set of underwater images at the target water depth according to the light attenuation amount at different water depths to obtain a set of target underwater images; an observation unit segmentation module for performing minimum observation unit segmentation on the set of target underwater images and dividing a benthic animal candidate image region according to the segmentation result; a benthic animal recognition module for identifying a benthic animal target image region according to the edge feature variation of the benthic animal candidate image region, and binding the benthic animal target image region with the target water depth to generate a benthic animal water depth distribution report.

[0008] The intelligent benthic animal recognition system of the application can implement any one of the intelligent benthic animal recognition methods of the application, and is used as a medium for joint operation and signal transmission between modules to complete the intelligent benthic animal recognition method. The modules in the system cooperate with each other, thereby effectively reducing the influence of light, noise and background interference. Compared with traditional manual recognition and sample-dependent deep learning methods, the method has higher recognition accuracy and stronger environmental adaptability, and can support real-time monitoring and automatic ecological state evaluation of benthic communities.

[0009] Optionally, the application also provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed to implement the intelligent benthic animal recognition method as described above. BRIEF DESCRIPTION OF DRAWINGS

[0010] Other features, objects, and advantages of the application will become more apparent from the following detailed description when read in connection with the following drawings: Fig. 1 A flow chart of the steps of the intelligent benthic animal recognition method of the present application; Fig. 2 Part of the original underwater image set in the embodiment of the present application; Fig. 3 Part of the corrected underwater image set in the embodiment of the present application; The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0011] The technical method of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0012] In addition, the accompanying drawings are only schematic illustrations of the present application, and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and thus repeated description thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities, which do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software, or in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0013] It should be understood that although the terms “first”, “second” and the like can be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, without departing from the scope of the exemplary embodiments, a first element can be called a second element, and similarly a second element can be called a first element. The term “and / or” as used herein includes any and all combinations of one or more of the associated listed items.

[0014] To achieve the above-mentioned object, please refer to Figs. 1 to 3 The present application provides an intelligent benthic animal recognition method, which comprises the following steps: Step S1: using a preset underwater camera to collect an underwater image set of different water depths in a target area; In this embodiment, if the target area is the near-shore shallow sea area, the underwater camera array arranged at different water depths (0.5 m, 1 m, 5 m, 10 m, 20 m) is used to synchronously collect underwater images under the condition of no wind and wave. Each camera uses a wide dynamic image sensor with infrared compensation, the exposure time is set to 0.03 s, and the frame rate is 20 fps to ensure that the illumination change can be recorded in real time. During the collection process, the control unit automatically marks the shooting depth according to the depth calibration sensor unit and performs time synchronization. After the collected multi-water-depth images are formatted, the initial underwater image set is generated, providing depth layering basis for subsequent illumination correction.

[0015] Step S2: determining the illumination attenuation at different water depths according to the pixel gray level change of the underwater image set at different water depths, and correcting the underwater image set at the target water depth according to the illumination attenuation at different water depths to obtain the target underwater image set; In a further embodiment, the average pixel gray level difference between adjacent water depths at the same sampling time is calculated to obtain a gray level attenuation value sequence. According to the turbidity, suspended particle concentration and temperature data collected by the water quality sensor, the gray level attenuation value is corrected by regression, and a second-order polynomial fitting is used to obtain an illumination attenuation curve. Taking the average gray level of the water surface as the reference gray level, the environmental parameter compensation correction of the mutation point is performed to obtain the illumination attenuation at different water depths. Combined with the suitable water depth parameters in the benthic animal characteristic library in spring, the illumination attenuation is converted into an illumination compensation coefficient, and the target water depth image is corrected pixel by pixel to generate the target underwater image set.

[0016] Step S3: performing minimum observation unit segmentation on the target underwater image set, and dividing the benthic animal candidate image region according to the segmentation result; In a further embodiment, Canny edge detection is performed on the image to obtain the edge pixel distribution. According to the connectivity of the edge pixels, a closed contour region is determined, and the minimum contour pixel area is calculated. When the minimum area is less than 0.2% of the total image pixels, it is taken as the segmentation step. According to this segmentation step, region segmentation is performed to obtain a plurality of minimum observation units. The edge pixel density of each minimum observation unit is detected, and the units with a density higher than 0.35 are selected as candidate units. Then, a preliminary candidate region set is generated according to the connectivity between the units. The local texture direction correlation in the candidate region is counted, and when the correlation is greater than 0.75 and the gray level mean square deviation is within the benthic animal tissue characteristic interval, it is marked as a benthic animal candidate image region.

[0017] Step S4: identifying the benthic animal target image region according to the edge feature change of the benthic animal candidate image region, and binding the benthic animal target image region with the target water depth to generate a benthic animal water depth distribution report.

[0018] In a further embodiment, the gradient direction angle of the edge pixel of the benthic animal candidate image region is identified, and the concentration of the gradient direction in a 5x5 pixel neighborhood is counted. When the concentration is unimodal or multimodal distribution (selected according to the characteristics of different benthic animals), the neighborhood is determined as a continuous edge region. Then the gradient intensity variation rate curve of the continuous edge region is calculated, the adjacent intensity inversion points are identified and connected to form an edge closed structure. The system matches the closed structure with the shape statistical characteristics (including roundness 0.7-0.9, aspect ratio 1.2-1.6) in the benthic animal growth rule, when the edge direction changes smoothly and the roundness is in the similar interval, the closed structure is determined as the benthic animal target image region, and it is bound with the corresponding water depth label to generate the benthic animal water depth distribution report.

[0019] Optionally, the step S2 of determining the light attenuation amount of different water depths comprises: According to the set of underwater images of different water depths, the difference between the average gray values of the pixel points between adjacent water depths in the same sampling time window is counted. In this embodiment, it is detected that a plurality of sets of underwater image data of different water depths have been acquired in the same time window, and the image sampling depth is divided into depth intervals of 0.5m to 3.0m by the depth calibration sensing unit. First, the color original image is converted into a gray image, and then the pixel gray value of each depth interval image is counted, and the difference between the average gray values of the pixels at the same spatial position between adjacent depths is calculated . Wherein, each average gray value is the local average value of a 5x5 pixel neighborhood.

[0020] The gray attenuation value of different water depths is determined based on the difference between the average gray values of the pixel points between adjacent water depths, and the gray attenuation value is corrected according to the pre-acquired environmental sensing data to obtain the light attenuation amount of different water depths.

[0021] In a further embodiment, by comparing The attenuation curve of the gray value with the water depth is obtained by comparing the trend with the depth. The attenuation curve preliminarily characterizes the depth loss relationship of the light intensity. On this basis, according to the pre-acquired environmental sensing data including water turbidity, suspended particle concentration and water dispersion coefficient, the gray attenuation value is corrected again. For example, if the turbidity is higher than 10 NTU, the gray attenuation amount is increased by 1.2 times in proportion; if the turbidity is lower than 5 NTU, it is corrected by 0.9 times.

[0022] It is worth noting that the pre-acquired environmental sensing data is obtained by a multi-parameter water quality environmental sensing unit deployed in the water body of the sampling area. The sensing unit is usually arranged at the sampling depth point synchronized with the underwater camera, which includes an illumination intensity sensor, a turbidity sensor, a temperature sensor, a dissolved oxygen probe, and a chlorophyll fluorescence sensor, etc., for real-time monitoring of physical and chemical parameters related to underwater imaging quality. The environmental sensing unit records each sensing parameter at a sampling frequency of 0.5Hz, and performs time synchronization calibration once before and after each image acquisition, respectively, to ensure the consistency of image data and environmental data on the time axis. Subsequently, the multi-dimensional environmental data obtained by continuous sampling is subjected to time series interpolation and noise filtering to form an environmental sensing data set corresponding to the underwater image acquisition depth.

[0023] Fig. 2 With Fig. 3 are the original underwater image set and the corrected underwater image set in the embodiment of the application, respectively. The figure includes three different water depths, i.e. 5m, 10m and 20m. Fig. 2 With Fig. 3 In order to reflect the same water depth but different image sets at the same water depth (5m) for samples at different positions or repeated collection, the number of image sets for each water depth can be adjusted according to different application scenarios, which is not limited in the application.

[0024] Optionally, performing environmental influence correction comprises: The underwater image set in the underwater image set at different water depths within the range of zero-point to near-surface water collection depth is taken as the reference water surface image set, and the average gray value of the pixels in the reference water surface image set is taken as the gray reference value. In the embodiment, if it is detected that the underwater camera completes continuous collection within the depth range of 0m to 3m, the image frame group with a collection depth of 0m to 0.3m is set as the reference water surface image set. By performing global gray statistics on the pixels of each image in the image set, the average gray value is calculated as the gray reference value.

[0025] Especially important is that the zero-point refers to the reference depth of the water surface, i.e. the water depth of 0m, which is the position where the water surface just contacts the air. The illumination here is the strongest, which can be considered as the initial reference point for the calculation of underwater image illumination attenuation. The near-surface range is the shallow water area close to the water surface, where the illumination has not been significantly attenuated, usually between 0m to 1m or 0m to 2m (the specific range depends on the water transparency and sampling resolution). In this range, the turbidity, suspended particle concentration and temperature variation of the water body are small, and the illumination intensity is almost close to the incident light on the water surface, so it can be used as a stable reference layer for illumination attenuation correction.

[0026] Based on the gray reference value, the mutation attenuation point in the gray attenuation value of different water depths is screened, and the environmental sensing parameters of the adjacent water depth of the mutation attenuation point are extracted from the environmental sensing data; In further embodiments, based on the gray reference value, the gray attenuation sequence of each water depth image is analyzed layer by layer, and if the gray attenuation rate of any depth layer increases by more than 30% compared to the previous layer, it is determined to be a mutation attenuation point. Subsequently, the turbidity, dispersion coefficient and water temperature parameters corresponding to the adjacent depth of the mutation point are extracted from the environmental sensing data.

[0027] The gray attenuation value adjacent to the mutation attenuation point is regressed and associated with the environmental sensing parameters of the adjacent water depth to determine the associated environmental sensing parameters. The gray attenuation value of the mutation attenuation point is corrected according to the associated environmental sensing parameters, and the gray attenuation value correction result is converted to the illumination attenuation amount based on the regression relationship between the gray attenuation value and the illumination intensity in the regression association result, thereby obtaining the illumination attenuation amount at different water depths.

[0028] In further embodiments, to more accurately determine the environmental correlation of the mutation attenuation point, a multiple linear regression model is constructed based on the extracted gray attenuation sequence and environmental sensing data. The model takes the gray attenuation values of the adjacent two layers , the corresponding turbidity parameter T, the suspended particle concentration C and the water dispersion coefficient S as input variables, and uses the least squares method to calculate the regression coefficients of each environmental parameter to form a regression equation. The regression equation is cross-validated, and if the determination coefficient of the model after cross-validation is greater than 0.95 and the residual standard deviation is less than 0.05, it is considered that the regression relationship between the gray attenuation and the environmental parameters is stable. On this basis, the environmental parameters at the mutation attenuation point are input into the model to calculate the theoretical gray attenuation prediction value. When the difference between the measured gray attenuation value and the prediction value exceeds the preset deviation threshold =0.1, it is considered that the mutation point is affected by local environmental disturbance. At this time, the reverse correction operation is performed: adjusting according to the deviation direction, and the correction amount is , where is the correction coefficient (0.6-0.8), to balance the environmental influence difference between the measured and predicted values. After the correction is completed, based on the established empirical regression relationship between the gray attenuation value and the illumination intensity (fitted by the measured illumination value of the reference water surface image set and the corrected gray value of each depth layer, ≥0.92), the corrected gray attenuation value is converted to the illumination attenuation amount L_d. The illumination attenuation amount at different water depths is formed.

[0029] Optionally, the underwater image set of the target water depth corrected in step S2 comprises: The environmental sensing data is compared with the preset benthic animal characteristic library to determine the suitable water depth of the benthic animal in the current seasonal environment, and the suitable water depth is taken as the target water depth; In this embodiment, the suitable water depth range of each species in the benthic animal characteristic library, the seasonal growth environment parameters and the benthic animal growth rules are indexed; then, according to the current collected environmental sensing data such as temperature, dissolved oxygen, turbidity and light intensity, the environmental parameters and the benthic animal growth rules of each water depth are matched layer by layer, if the multi-dimensional environmental parameters of a water depth all fall within the suitable range specified by the benthic animal characteristic library, it is determined that the water depth is the suitable water depth of the target species in the current season, and the water depth is recorded as the target water depth.

[0030] The underwater image set of the target water depth is extracted from the underwater image set of different water depths, and the gray scale distribution of the pixel points in the underwater image set of the target water depth is extracted; In a further embodiment, the underwater image set of the target water depth is extracted from the underwater image set of different water depths (at this time it is a color image), then the gray scale value distribution of each pixel point in the image is counted, the mean, variance and gray scale histogram are calculated, and the gray scale distribution of the target water depth is generated.

[0031] The gray scale reference value is converted into a light intensity reference value, and the light compensation coefficient of the target water depth is calculated according to the light intensity reference value and the light attenuation amount of the target water depth; In a further embodiment, according to the gray scale mean of the shallow reference water surface image set and the corresponding water surface light intensity obtained by the sensing unit, a linear mapping relationship is then established: ; wherein represents the gray scale value of any pixel, is the mapped light intensity value, is the gain coefficient, is the bias term. The gain coefficient and the bias are determined by fitting the corresponding relationship between the gray scale mean of the shallow reference image and the measured light intensity by the least square method, and then the gray scale value is linearly mapped to the light intensity value to form the light reference value. Then, according to the light attenuation amount of the target water depth, the light compensation coefficient is calculated using the linear light attenuation model: , wherein is the reference light intensity, is the light attenuation amount of the target water depth.

[0032] The light compensation coefficient is used to correct the gray scale value of each pixel point in the underwater image set of the target water depth to obtain the target underwater image set.

[0033] In a further embodiment, linear correction is performed on each pixel gray scale : And perform gray truncation (0-255) on the corrected pixel value to maintain the display range while preserving the original texture features of the image, finally obtaining a target underwater image set with balanced illumination that can be used for subsequent benthic animal identification.

[0034] Optionally, the method for obtaining the benthic animal characteristic library comprises: Obtaining regional sampling data in different seasons to determine the environmental characteristic parameters of the regional water body in each season; In this embodiment, by arranging multiple sampling stations in different seasons in the target water area, the water quality sensing unit is used to collect parameters such as water temperature, dissolved oxygen, pH value, salinity, flow rate and turbidity. The data collection period of each sampling point is 30 minutes, and continuous sampling is performed for 72 hours to obtain a stable seasonal environmental characteristic parameter matrix.

[0035] Compare the environmental characteristic parameters of each season with the preset benthic animal growth rule. If the environmental characteristic parameters of any season are within the suitable growth range of the benthic animal growth rule, the environmental characteristic parameters of that season are determined as the suitable growth environment of the benthic animal, and the suitable growth environments of the benthic animal under the environments of each season are collected to construct the benthic animal characteristic library.

[0036] In a further embodiment, the collected seasonal environmental characteristic parameters are compared with the preset benthic animal growth rule. The benthic animal growth rule includes parameters such as suitable water temperature range, lower limit of dissolved oxygen concentration, salinity range and flow rate limit of different species, and whether the sampling point environmental parameters fall within the suitable growth interval is judged by parameter-by-parameter matching. If the matching proportion exceeds 80%, it is determined that the seasonal environment corresponding to the sampling point is the suitable growth environment of the benthic animal. Finally, the suitable environmental parameters of all seasons and each sampling point are collected to form a benthic animal characteristic library, each record including season identification, sampling water depth, geographic coordinates and matched environmental parameter interval, and stored as a structured database.

[0037] Especially important is that the benthic animal growth rule includes the suitable growth range of each benthic animal, the biological tissue characteristic interval of different growth stages, and the shape statistical characteristics of different growth stages.

[0038] In this embodiment, the environmental adaptation characteristics of different species at different growth stages are obtained through field investigation and laboratory cultivation. For example, the suitable intervals of environmental parameters such as water temperature, dissolved oxygen, salinity, and flow rate for larvae, subadults, and adults are recorded, as well as the physiological characteristic data of biological tissues at each stage, such as density, thickness, and flexibility. These data are obtained through standardized measurement and statistical processing to obtain the biological tissue characteristic interval. Secondly, morphological data of benthic animals at different growth stages are obtained through image analysis or three-dimensional scanning, including length, width, thickness, roundness, flatness, and symmetry of shell or body contour, and statistical methods are used to calculate mean, variance, and other shape statistical characteristics to form a morphological characteristic model at different growth stages. Subsequently, the suitable growth range, biological tissue characteristic interval, and shape statistical characteristics are integrated to establish a benthic animal growth rule library, which is used for comparison with the sampling environmental parameters or matching with the image features, thereby assisting in determining whether the target water depth or candidate area meets the growth conditions of benthic animals.

[0039] Optionally, the minimum observation unit segmentation in step S3 includes: Edge detection is performed on the target underwater image set, and the contour closed region is determined according to the connectivity of the edge pixels in the edge detection result; In this embodiment, the color original image is converted to a grayscale image to remove the influence of color information on edge determination. Then the Canny operator is used to extract the edge, the low threshold is set to 0.05, the high threshold is set to 0.15, and the image is subjected to Gaussian smoothing processing. Then the gradient amplitude of each pixel is calculated, the non-maximum suppression is performed on the gradient amplitude, and then the double-threshold algorithm is used to connect the edges in the high-threshold image into a contour. When the end point of the contour is reached during the connection, the edge that can be connected is found on the low-threshold image to determine the real edge pixels. For the extracted edge pixels, the 8-neighbor connectivity is used to judge whether the adjacent pixels are continuous to form a preliminary contour closed region, and the isolated pixel points are removed.

[0040] The minimum pixel area of the contour closed region is calculated, and the scale of the minimum pixel area is adjusted according to the resolution of the target underwater image set to obtain the minimum observation unit segmentation step; In a further embodiment, when calculating the minimum pixel area of the contour closed region, the total number of pixels in each closed region can be counted. Assuming that the resolution is 1024x1024 pixels, the minimum pixel area is set to 25 The minimum pixel area is scaled in proportion to the image resolution (e.g. 1024x1024 pixels) to keep the minimum observation unit at a scale of about 5x5 pixels in high-resolution images, ensuring the preservation of detailed information. This value is also used as the minimum observation unit segmentation step to control the subsequent image division accuracy.

[0041] The image segmentation is performed on the target underwater image set by using a minimum observation unit segmentation step to obtain a plurality of minimum observation units.

[0042] In a further embodiment, the image is grid divided by using the obtained minimum observation unit segmentation step to cut the whole image into a plurality of square grids, each of which corresponds to a minimum observation unit, and the step is set to 5-10 pixels to balance the resolution and detail retention.

[0043] Optionally, the step S3 of dividing the benthic animal candidate image region comprises: The minimum observation unit with an edge pixel density higher than a preset density threshold is taken as a candidate observation unit; In the embodiment, the proportion of the edge pixels in each minimum observation unit to the total pixels is counted, and if the proportion is higher than the preset density threshold 0.35, the unit is marked as a candidate observation unit. The edge pixels are derived from the pixels with a gradient amplitude greater than 0.1 extracted by the aforementioned Canny operator, and the gradient direction information is retained for subsequent analysis.

[0044] According to the connectivity of each minimum observation unit, the candidate observation units are aggregated to generate a preliminary candidate region set; In a further embodiment, the merging is performed according to the 8-neighbor connectivity between the units. If the connectivity of the edge pixels between the adjacent units is greater than 0.5, the units are aggregated into the same region. Each aggregated region saves the edge pixel matrix, the gray matrix and the unit index information, and the aggregated region size can be limited to be not less than 3*3 minimum observation units to avoid generating too small candidate regions.

[0045] The local texture direction correlation and the gray mean square deviation of the adjacent units in the preliminary candidate region set are counted, if the local texture direction correlation exceeds a preset correlation threshold, and the gray mean square deviation is in the biological tissue feature interval in the benthic animal growth rule, the adjacent unit is marked as a benthic animal candidate image region.

[0046] In a further embodiment, the local texture direction correlation and the gray mean square deviation of each adjacent minimum observation unit in the preliminary candidate region are counted. The local texture direction correlation is obtained by calculating the cosine similarity of the gradient directions inside the unit, if the correlation exceeds 0.85 and the gray mean square deviation is between 0.02 and 0.08, the unit is marked as a benthic animal candidate image region.

[0047] Especially importantly, the method for obtaining the connectivity of each minimum observation unit comprises: The connectivity of the adjacent edge pixels in the minimum observation unit is determined according to the connectivity of the edge pixels; In this embodiment, the edge pixels in a unit are scanned in 8-neighborhood, and if there is a connection between two edge pixels in 8-neighborhood, it is determined that the two edge pixels are connected. In order to improve robustness, a gradient amplitude threshold of 0.1 is set, and only the pixels with an amplitude greater than the threshold are involved in the connectivity determination. Through this method, the connected subsets in the unit are generated, and the connectivity of each subset is calculated, i.e. the ratio of the number of connected pixels to the total number of edge pixels in the unit.

[0048] The connectivity of the boundary pixels between adjacent minimum observation units is determined by the connectivity of the adjacent edge pixels in the minimum observation unit, so as to obtain the connectivity of the adjacent minimum observation units.

[0049] In a further embodiment, an intersection region of the pixels on the boundary lines of two adjacent units is selected, and the proportion of the connected subset pixels in the total number of boundary pixels in the region is counted. If the proportion is greater than 0.5, it is determined that the boundary of the two minimum observation units is connected, and the connectivity relationship between the adjacent minimum observation units is established.

[0050] Optionally, the identifying the benthic animal target image region in step S4 comprises: identifying the gray level gradient direction angle of each edge pixel in the benthic animal candidate image region, and counting the gradient direction concentration of the gray level gradient direction angles in the same local neighborhood; In this embodiment, the Sobel operator or the Prewitt operator is executed on the benthic animal candidate image region, and the gradient components of each edge pixel in the X and Y directions are calculated. The gray level gradient direction angle of the pixel is calculated from the gradient components , and the direction angle is discretized into 12 direction intervals, each interval covering 30°. Then, the distribution of the gray level gradient direction angles in each local neighborhood (such as a 5x5 pixel block) is counted, and the direction concentration index (the concentration threshold can be set to 0.7) is calculated, which is used to determine whether the direction is highly concentrated.

[0051] When the gradient direction concentration of the gray level gradient direction angles in any local neighborhood shows unimodal or bimodal characteristics, the local neighborhood is identified as a continuous edge region; In a further embodiment, when the concentration of the gray level gradient direction angles in any local neighborhood is greater than 0.7 and the distribution shows unimodal or bimodal characteristics, the local neighborhood is marked as a continuous edge region. This marking is used for subsequent edge closed structure identification, ensuring that the edge region has direction continuity and structural integrity.

[0052] The rate of change curve of the edge strength along the gray level gradient direction angle in the continuous edge region is calculated, the strength reversal points of the rate of change curve are identified, and the adjacent strength reversal points are connected to obtain the edge closed structure; In a further embodiment, when calculating the rate curve of the edge intensity in the continuous edge region along the gradient direction angle, the local pixel intensity difference method is used to calculate the rate of the intensity of each pixel along the gradient direction, the points with the absolute value of the rate greater than 0.05 are selected as the intensity reversal points, and the adjacent reversal points are sequentially connected to form a closed edge structure contour.

[0053] The edge closed structure is matched with the shape statistical characteristics of the benthic animals in the benthic animal growth rule, if the edge direction change of the edge closed structure is smooth and the roundness is in the similar interval of the shape statistical characteristics, the edge closed structure is taken as the benthic animal target image region.

[0054] In a further embodiment, when matching the closed edge structure with the benthic animal shape statistical characteristics, the roundness (0.7-1.0 is the roundness similar interval), the edge direction smoothness (change angle ≤15° / pixel) and the area ratio of the contour are extracted, and are compared with the statistical characteristics of different growth stages in the benthic animal growth rule, if all satisfy the similarity condition, the closed structure is determined as the benthic animal target image region, and the spatial position and the target water depth information are recorded.

[0055] Optionally, the present application also provides a benthic animal intelligent identification system for executing the benthic animal intelligent identification method as described above, the benthic animal intelligent identification system comprises: An image acquisition module is configured to acquire a set of underwater images of different water depths in a target region by using a preset underwater camera; An image preprocessing module is configured to determine the light attenuation amount of different water depths according to the pixel gray scale variation of the set of underwater images of different water depths, and correct the set of underwater images of the target water depth according to the light attenuation amount of different water depths to obtain a set of target underwater images; An observation unit segmentation module is configured to perform minimum observation unit segmentation on the set of target underwater images, and divide benthic animal candidate image regions according to the segmentation result; A benthic animal identification module is configured to identify a benthic animal target image region according to the edge feature variation of the benthic animal candidate image region, bind the benthic animal target image region with the target water depth, and generate a benthic animal water depth distribution report.

[0056] Optionally, the present application also provides a computer readable storage medium having a computer program stored thereon, the computer program is executed to implement the benthic animal intelligent identification method as described above.

[0057] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, the scope of the present application is defined by the appended claims rather than the above description, and therefore all changes falling within the meaning and scope of the equivalent elements of the application file are intended to be included in the present application.

[0058] The foregoing is considered as illustrative only of the principles of the application. Numerous modifications and changes will readily occur to those skilled in the art, and the generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the application. Therefore, the scope of the application is indicated by the appended claims rather than by the foregoing description, and all changes that come within the meaning and range of equivalents are intended to be embraced therein.

Claims

1. A benthic animal intelligent recognition method, characterized in that, The method comprises the following steps: Step S1: collecting underwater image sets of different water depths in a target area by using a preset underwater camera; Step S2: determining the light attenuation amount of different water depths according to the pixel gray level change of the underwater image sets of different water depths, and correcting the underwater image set of the target water depth according to the light attenuation amount of different water depths to obtain the target underwater image set; Step S3: performing minimum observation unit segmentation on the target underwater image set, and dividing the benthic animal candidate image area according to the segmentation result; Step S4: identifying the benthic animal target image area according to the edge feature change of the benthic animal candidate image area, and binding the benthic animal target image area with the target water depth to generate a benthic animal water depth distribution report.

2. The benthic organism intelligent identification method according to claim 1, characterized in that, The determination of the light attenuation amount of different water depths in step S2 comprises: According to the underwater image sets of different water depths, the difference between the average gray levels of the pixel points in adjacent water depths within the same sampling time window is calculated; Based on the difference between the average gray levels of the pixel points in adjacent water depths, the gray attenuation value of different water depths is determined, and the gray attenuation value is corrected according to the pre-acquired environmental sensing data to obtain the light attenuation amount of different water depths.

3. The benthic organism intelligent identification method according to claim 2, characterized in that, The environmental influence correction comprises: The underwater image set in the underwater image set of different water depths within the range of zero-point to near-surface water collection depth is taken as the reference water surface image set, and the average gray level of the pixels in the reference water surface image set is taken as the gray reference value; Based on the gray reference value, the mutation attenuation point in the gray attenuation value of different water depths is screened, and the environmental sensing parameters of the adjacent water depth of the mutation attenuation point are extracted from the environmental sensing data; The gray attenuation value adjacent to the mutation attenuation point is regressed and associated with the adjacent water depth environmental sensing parameter to determine the associated environmental sensing parameter; The gray attenuation value of the mutation attenuation point is corrected according to the associated environmental sensing parameter, and the gray attenuation value correction result is converted into the light attenuation amount based on the regression relationship between the gray attenuation value and the light intensity in the regression association result, so as to obtain the light attenuation amount of different water depths.

4. The benthic organism intelligent identification method according to claim 2, characterized in that, The correction of the underwater image set of the target water depth in step S2 comprises: Comparing the environmental sensing data with the preset benthic animal characteristic library to determine the suitable water depth of the benthic animal in the current season environment, and taking the suitable water depth as the target water depth; Extracting the underwater image set of the target water depth from the underwater image set of different water depths, and extracting the gray distribution of the pixel points in the underwater image set of the target water depth; Converting the gray reference value into a light intensity reference value, and calculating the light compensation coefficient of the target water depth according to the light intensity reference value and the light attenuation amount of the target water depth; Using the light compensation coefficient to correct the gray value of each pixel point in the underwater image set of the target water depth to obtain the target underwater image set.

5. The benthic organism intelligent identification method according to claim 1, characterized in that, The method for obtaining the benthic animal characteristic library comprises: Obtaining regional sampling data in different seasons to determine the environmental characteristic parameters of the regional water body in each season; The environmental characteristic parameters of each season are compared with the preset benthic animal growth rule, if the environmental characteristic parameters of any season are in the suitable growth range of the benthic animal growth rule, the environmental characteristic parameters of the season are determined as the suitable growth environment of the benthic animal, and the suitable growth environments of the benthic animal in each season are collected to construct the benthic animal characteristic library.

6. The benthic organism intelligent identification method according to claim 1, characterized in that, The minimum observation unit segmentation in step S3 includes: Edge detection is performed on the target underwater image set, and a contour closed region is determined according to the connectivity of the edge pixels in the edge detection result; The minimum pixel area of the contour closed region is calculated, and the scale of the minimum pixel area is adjusted according to the resolution of the target underwater image set to obtain a minimum observation unit segmentation step; Image segmentation is performed on the target underwater image set using the minimum observation unit segmentation step to obtain a plurality of minimum observation units.

7. The benthic organism intelligent identification method according to claim 1, characterized in that, The division of the benthic animal candidate image region in step S3 includes: The minimum observation unit in which the edge pixel density is higher than the preset density threshold is taken as a candidate observation unit; According to the connectivity of each minimum observation unit, the candidate observation units are aggregated to generate a preliminary candidate region set; The local texture direction correlation degree and the gray mean square deviation of adjacent units in the preliminary candidate region set are counted, if the local texture direction correlation degree exceeds the preset correlation degree threshold, and the gray mean square deviation is in the biological tissue feature interval in the benthic animal growth rule, the adjacent unit is marked as a benthic animal candidate image region.

8. The benthic organism intelligent identification method of claim 1, wherein, The identification of the benthic animal target image region in step S4 includes: The gray gradient direction angle of each edge pixel in the benthic animal candidate image region is identified, and the gradient direction concentration degree of the gray gradient direction angle in the same local neighborhood is counted; When the gradient direction concentration degree of the gray gradient direction angle in any local neighborhood shows unimodal or bimodal characteristics, the local neighborhood is identified as a continuous edge region; The change rate curve of the edge intensity along the gray gradient direction angle in the continuous edge region is calculated, the intensity reversal points of the change rate curve are identified, and adjacent intensity reversal points are connected to obtain an edge closed structure; The edge closed structure is matched with the shape statistical characteristics of the benthic animal in the benthic animal growth rule, if the edge direction change of the edge closed structure is smooth and the roundness is in the similar interval of the shape statistical characteristics, the edge closed structure is taken as the benthic animal target image region.

9. A benthic animal intelligent identification system, characterized in that, The benthic animal intelligent identification system for performing the benthic animal intelligent identification method as claimed in claim 1 includes: An image acquisition module for acquiring an underwater image set of different water depths in a target region using a preset underwater camera; An image preprocessing module for determining the light attenuation amount of different water depths according to the pixel gray variation of the underwater image set of different water depths, and correcting the underwater image set of the target water depth according to the light attenuation amount of different water depths to obtain the target underwater image set; An observation unit segmentation module for performing minimum observation unit segmentation on the target underwater image set, and dividing the benthic animal candidate image region according to the segmentation result; The benthic animal identification module is used for identifying a benthic animal target image area according to edge feature variation of the benthic animal candidate image area, and binding the benthic animal target image area with a target water depth to generate a benthic animal water depth distribution report.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer readable medium, and the computer program is executed to implement the benthic animal intelligent identification method according to any one of claims 1-8.

Images