Intelligent identification method, device, equipment and medium for benthic animals

Abstract

The present invention relates to a method, device, equipment and medium for intelligent identification of benthic animals. The method uses an ultrasonic distance sensor preset on a benthic exploration device to calculate the distance information from the water bottom, and starts a visual acquisition processing unit based on the distance information; instructs the visual acquisition processing unit to obtain water environment image information, analyzes whether the water environment image meets the preset identification rules, and if so, performs the identification work; otherwise, further searches for places that meet the preset identification rules nearby; fuses gyroscope floating data to correct image pixels, and performs image enhancement. The water environment image information is input into a frequency domain multi-scale feature fusion neural network model in the visual acquisition processing unit to obtain processed benthic animal information, so as to provide reliable information for further analysis or sampling of benthic animals.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method, device, equipment and medium for intelligently identifying benthic animals. Background Art

[0002] Benthic animals refer to aquatic animals that live at the bottom of water bodies for all or most of their life history. They are an indispensable part of the ecological environment of rivers and lakes, and their community species and density have become important indicators reflecting the quality of the ecological environment of rivers and lakes. Therefore, ecological science researchers at home and abroad often study the changes in the ecological environment of rivers and lakes by investigating the changes in various indicators of benthic animals, and find out the specific influencing factors, so as to provide a scientific basis for the formulation of relevant water ecological environment protection policies and programs.

[0003] Identification of benthic animals has always been a challenging task. Since benthic animals live at the bottom of the water, they are often affected by many factors such as water quality, water depth, and water flow. Traditional capture and identification methods often have difficulties in accurately capturing and identifying these animals, and this method has disadvantages such as low accuracy and efficiency, and long time consumption. Due to complex factors such as different light levels in the underwater environment and different turbidity of the water body, it is difficult to accurately identify them using deep three-dimensional vision methods. Summary of the invention

[0004] The main purpose of the present invention is to provide a method, device, equipment and medium for intelligent identification of benthic animals to solve the problem of difficulty in identifying benthic animals.

[0005] To achieve the above object, the present invention provides a method for intelligent identification of benthic animals, comprising the following steps:

[0006] An ultrasonic distance sensor preset on the benthic exploration equipment is used to calculate the distance information from the water bottom, and a visual acquisition processing unit is started based on the distance information, wherein an instruction corresponding to starting the visual acquisition processing unit is generated by judging whether the distance information reaches a preset threshold;

[0007] Instruct the visual acquisition processing unit to acquire water environment image information, analyze whether the water environment image information meets the preset recognition rules, and if so, perform recognition work; otherwise, further search for a location that meets the preset recognition rules nearby;

[0008] Using a preset gyroscope to collect gyroscope floating data, and performing pixel correction on water environment image information according to the gyroscope floating data, and performing image target enhancement processing on the corrected image;

[0009] The water environment image information is input into a frequency domain multi-scale feature fusion neural network model in a visual acquisition processing unit to obtain benthic animal information after model processing, wherein the benthic animal information includes the types, geometric features and vital sign activity of benthic animals on the bottom surface of the water area.

[0010] Furthermore, the step of performing pixel correction on the water environment image information according to the gyroscope floating data comprises:

[0011] The gyroscope floating data is optimized by a preset first-order low-pass filter to construct an image rotation transformation matrix, and the gyroscope floating data is used to perform pixel point correction on the image pixel coordinates based on the image rotation transformation matrix, wherein the image rotation transformation matrix is ​​calculated as follows:

[0012] ;

[0013] in is the identity matrix, is the image rotation transformation matrix, Represents the angles of rotation around the x-axis, y-axis, and z-axis respectively. Indicates the overall rotation angle of the benthic exploration equipment.

[0014] Furthermore, the algorithm for optimizing the gyroscope floating data is performed by a preset first-order low-pass filter, including: ;

[0015] in, is the angular velocity measured by the gyroscope at the current moment, is the angular velocity measured by the gyroscope at the last moment, is the overall rotation angle from time m to time n, is the sampling period, is the cut-off frequency.

[0016] Furthermore, the step of correcting the pixel coordinate values ​​of the image pixel coordinates by the visual acquisition processing unit through the gyroscope floating data also includes:

[0017] Obtain the pixel coordinate values ​​of the image of the previous frame by optimizing the gyroscope floating data;

[0018] Obtain pixel coordinate values ​​of the image of the intermediate frame during the gyroscope floating data optimization process;

[0019] Obtain pixel coordinate values ​​of an image frame after gyroscope floating data optimization processing;

[0020] Calculate the rotation matrix of the image of the frame before the gyroscope floating data optimization processing to the image of the frame after the gyroscope floating data optimization processing, and apply the rotation matrix to weightedly calculate the corresponding new pixel point coordinate value. The algorithm is: ;

[0021] in is the camera intrinsic parameter matrix, is the rotation matrix, is the pixel coordinate point of the new image, is the pixel coordinate value of the previous frame image, is the pixel coordinate value of the image in the middle frame, is the pixel coordinate value of the next frame image, is the weight coefficient.

[0022] Furthermore, the step of performing image target enhancement processing on the corrected image includes:

[0023] The corrected original image is obtained by performing pixel correction on the water environment image information collected by the visual collection and processing unit, and then image enhancement is performed on each pixel of the corrected original image. The algorithm of the image enhancement is:

[0024] ;

[0025] in is the pixel value after image enhancement, is the pixel value of the original image, and H is the water depth measured by the ultrasonic distance sensor.

[0026] Furthermore, the frequency domain multi-scale feature fusion neural network model specifically includes:

[0027] The frequency domain multi-scale feature fusion neural network model uses ACDNET as a feature extractor to extract the collected water environment image information; the ACDNET adopts a dense connection method to connect each layer with all previous layers respectively, and adopts asymmetric convolution, that is, for n*n convolution, convolutions of 1*n and n*1 convolution kernels are performed in parallel on its side, and then the three are fused;

[0028] The output ACDNET features of the first layer, the third layer and the fifth layer of ACDNET are extracted respectively, and then the shallow features of the first layer and the third layer are input into the DCT module for processing. The DCT processing method includes:

[0029] Perform discrete cosine transform to obtain the low-frequency component, horizontal high-frequency component, vertical high-frequency component and diagonal high-frequency component of ACDNET features;

[0030] The horizontal high-frequency component and the vertical high-frequency component are aggregated and normalized using a softmax function to obtain detail features;

[0031] Then multiply the detail features with the low-frequency components to obtain a detailed feature map;

[0032] The low-frequency components are added to the detailed feature map to generate a complete feature map, and then the complete feature map enters the pyramid feature multi-scale fusion part; among them, the shallow layer of the ACDNET network is responsible for extracting the shallow features at the edge of the complete feature map, while the deep layer of the ACDNET network extracts the deep features of the high-level semantic information in the complete feature map, and the deep features are upsampled in turn and fused with the shallow features processed by the DCT module to gradually restore the size of the water environment image information.

[0033] Furthermore, the algorithm for performing discrete cosine transform includes: ;

[0034] in, It is an element of the discrete cosine transform matrix, which represents the frequency component of the signal; u and v are frequency indices, which represent the position of the frequency component in the frequency domain; is an M×N digital image matrix, M is the row of the digital image matrix, and N is the column of the digital image matrix; is the normalization coefficient; the obtained results are separated and extracted to obtain , respectively low-frequency component, horizontal high-frequency component and vertical high-frequency component; the entire DCT module is expressed as: ; Where Z is the output of the DCT module, the function G is the addition of elements, S is the softmax function, and δ is the multiplication of elements.

[0035] The present invention also provides an intelligent identification device for benthic animals, comprising:

[0036] A distance unit, used to calculate the distance information from the bottom of the water by using an ultrasonic distance sensor preset on the benthic exploration equipment, and start the visual acquisition processing unit corresponding to the distance information, wherein the instruction corresponding to starting the visual acquisition processing unit is generated by judging whether the distance information reaches a preset threshold;

[0037] An analysis unit, used to instruct the visual acquisition processing unit to obtain water environment image information, analyze whether the water environment image information meets the preset recognition rules, and if so, perform recognition work; otherwise, further search for a location that meets the preset recognition rules nearby;

[0038] A gyroscope unit collects gyroscope floating data using a preset gyroscope, performs pixel correction on water environment image information according to the gyroscope floating data, and performs image target enhancement processing on the corrected image;

[0039] The recognition unit is used to input the water environment image information into the frequency domain multi-scale feature fusion neural network model in the visual acquisition and processing unit to obtain benthic animal information after model processing, wherein the benthic animal information includes the types, geometric characteristics and vital sign activity of benthic animals on the bottom surface of the water area.

[0040] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned benthic animal intelligent identification method when executing the computer program.

[0041] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for intelligent identification of benthic animals are implemented.

[0042] The method, device, equipment and medium for intelligent identification of benthic animals provided by the present invention have the following beneficial effects:

[0043] Benthic animals can be identified normally in complex situations such as waters with poor water environment. Through the cooperation of high-definition industrial cameras and embedded image processors in the visual acquisition and processing unit, image information is optimized and enhanced, and the frequency domain multi-scale feature fusion neural network model is used for target recognition. Benthic animals can be found, identified and analyzed in real time, realizing online recognition and analysis of benthic organisms. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 1 is a schematic diagram of a process of an intelligent identification method for benthic animals in one embodiment of the present invention;

[0045] Figure 2 It is a processing schematic diagram of a frequency domain multi-scale feature fusion neural network model in a method for intelligent identification of benthic animals in one embodiment of the present invention;

[0046] Figure 3 It is a target recognition neural network flow chart of an intelligent benthic animal recognition sampling method of the present invention;

[0047] Figure 4 is a structural block diagram of an intelligent identification device for benthic animals in one embodiment of the present invention;

[0048] Figure 5 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0049] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0051] Reference Figure 1 The present invention provides a flow chart of a method for intelligent identification of benthic animals, which comprises the following steps:

[0052] S1, using an ultrasonic distance sensor preset on a benthic exploration device to calculate distance information from the bottom of the water, and starting a visual acquisition processing unit based on the distance information, wherein an instruction corresponding to starting the visual acquisition processing unit is generated by judging whether the distance information reaches a preset threshold;

[0053] S2, instructing the visual acquisition processing unit to acquire water environment image information, analyze whether the water environment image information meets the preset recognition rules, and if so, perform recognition work; otherwise, further search for a location that meets the preset recognition rules nearby;

[0054] S3, using a preset gyroscope to collect gyroscope floating data, and performing pixel correction on the water environment image information according to the gyroscope floating data, and performing image target enhancement processing on the corrected image;

[0055] S4, inputting the water environment image information into the frequency domain multi-scale feature fusion neural network model in the visual acquisition processing unit to obtain benthic animal information after model processing, wherein the benthic animal information includes the types, geometric characteristics and vital sign activity of benthic animals on the bottom surface of the water area.

[0056] Specifically,

[0057] In S1, the visual acquisition processing unit includes an ultrasonic distance sensor and an image acquisition component, wherein the image acquisition component is used to observe the underwater environment and identify benthic animals; the ultrasonic distance sensor is used to measure the water depth; and the image acquisition component is a waterproof industrial camera with an integrated gyroscope.

[0058] In S2, it is determined whether the turbidity of the water body has a significant impact on the image clarity. If the image clarity is greatly affected by the turbidity of the water body or the gyroscope data is abnormal due to excessive undercurrent, the recognition rule is not met.

[0059] The step of instructing the visual acquisition processing unit to acquire water environment image information comprises:

[0060] Using the embedded image processor included in the visual acquisition processing unit to acquire water environment images and gyroscope floating data;

[0061] In order to suppress the image blur caused by the impact and shaking of the benthic exploration equipment by the water flow, the visual system integrates the gyroscope data to correct the image pixels. The gyroscope floating data is optimized by a preset first-order low-pass filter, and the image rotation transformation matrix is ​​constructed in combination with the water environment image. The pixel points of the water environment image are transformed using the gyroscope data to obtain the transformed image, and then the transformed image is subjected to image target enhancement processing.

[0062] The calculation method of the image rotation transformation matrix is: ;

[0063] in is the identity matrix, is the image rotation transformation matrix, Represents the angles of rotation around the x-axis, y-axis, and z-axis respectively. Indicates the overall rotation angle of the benthic exploration equipment.

[0064] When using gyroscope data to perform pixel correction on water environment images, an algorithm is used to optimize the angular velocity read by the gyroscope using a specific first-order low-pass filter, including: ;

[0065] in, is the angular velocity measured by the gyroscope at the current moment, is the angular velocity measured by the gyroscope at the last moment, is the overall rotation angle from time m to time n, is the sampling period, is the cut-off frequency.

[0066] The visual acquisition processing unit performs pixel coordinate point correction on the image pixel coordinates through gyroscope data, including:

[0067] Obtain the pixel coordinate values ​​of the image of the previous frame by optimizing the gyroscope floating data;

[0068] Obtain pixel coordinate values ​​of the image of the intermediate frame during the gyroscope floating data optimization process;

[0069] Obtain pixel coordinate values ​​of an image frame after gyroscope floating data optimization processing;

[0070] Calculate the rotation matrix of the image of the frame before the gyroscope floating data optimization processing to the image of the frame after the gyroscope floating data optimization processing, and apply the rotation matrix to weightedly calculate the corresponding new pixel point coordinate value. The algorithm is: ;

[0071] in is the camera intrinsic parameter matrix, is the rotation matrix, is the pixel coordinate point of the new image, is the pixel coordinate value of the previous frame image, is the pixel coordinate value of the image in the middle frame, is the pixel coordinate value of the next frame image, is the weight coefficient.

[0072] The visual acquisition processing unit enhances the image target, including:

[0073] After pixel point correction transformation, the corrected original image is obtained, and then each pixel of the corrected original image is enhanced. The image enhancement method is: ;

[0074] Where A is the mapping position of the atmospheric light value, is the perspective ratio, is the red and blue light perspective ratio, w is the smoothness, and its value is 0.2;

[0075] ;

[0076] in is the corrected pixel value, is the pixel value of the original image, and H is the water depth measured by the ultrasonic sensor.

[0077] like Figure 2-Figure 3 , the frequency domain multi-scale feature fusion neural network model specifically includes:

[0078] The frequency domain multi-scale feature fusion neural network model uses ACDNET as a feature extractor to extract the collected water environment image information. ACDNET is improved on the basis of DenseNet. The ACDNET network adopts a dense connection method to connect each layer to all previous layers respectively, so as to achieve feature reuse and improve efficiency. In addition, asymmetric convolution is used instead of conventional convolution, that is, for n*n convolution, 1*n and n*1 convolution kernels are convolved in parallel on its side, and then the three are fused. The parallel convolution makes the extraction of horizontal and vertical features more obvious, which helps to improve the recognition accuracy of the model;

[0079] The output ACDNET features of the 1st layer, the 3rd layer and the 5th layer of ACDNET are extracted respectively, and then the shallow features of the 1st layer and the 3rd layer enter the DCT module for processing. The DCT module processing method includes:

[0080] Discrete cosine transform is performed to obtain the low-frequency component, horizontal high-frequency component, vertical high-frequency component and diagonal high-frequency component of ACDNET features. The low-frequency component retains the main structural information of the feature map, and the three high-frequency components not only retain detailed information but also retain a lot of noise information, so the diagonal high-frequency component is discarded and its low-frequency component, horizontal high-frequency component and vertical high-frequency component are extracted.

[0081] After aggregating the horizontal and vertical high-frequency components, they are normalized using the softmax function to obtain detailed features;

[0082] Then multiply the detail features with the low-frequency components to obtain a detailed feature map;

[0083] The low-frequency components are added to the detailed feature map to generate a complete feature map, which then enters the pyramid feature multi-scale fusion part. The shallow layer (layer 1) of the network is responsible for extracting shallow features such as target edges, while the deep layer (layer 3, layer 5) of the network extracts deep features such as advanced semantic information. By using this method, the information of the higher-level semantic part of the deep network can be integrated into the shallower network to guide the shallow network to identify, and the deep features are sequentially upsampled and fused with the shallow features processed by the DCT module, so as to gradually restore the scale of the original image.

[0084] The DCT module processing method comprises:

[0085] The discrete cosine transform process is: ;

[0086] in, It is an element of the discrete cosine transform matrix, which represents the frequency component of the signal; u and v are frequency indices, which represent the position of the frequency component in the frequency domain; is an M×N digital image matrix, M is the row of the digital image matrix, and N is the column of the digital image matrix; is the normalization coefficient;

[0087] The discrete cosine transform result is separated and extracted to obtain , respectively low-frequency component, horizontal high-frequency component and vertical high-frequency component; the entire DCT module is expressed as: ; Where Z is the output of the DCT module, the function G is the addition of elements, S is the softmax function, and δ is the multiplication of elements.

[0088] Since it is a multi-classification model, the multi-classification cross entropy loss function is calculated by calculating the value of the actual label and the forward propagation output, as follows: ;

[0089] Among them, L is the loss function, and M represents the number of categories required; Represents 0 or 1, if the real number class of sample i is c, it is 1, otherwise it is 0; represents sample i belonging to the prediction function of category c; N represents the number of samples. The weight update uses the fastest gradient method and calculates partial derivatives based on the multi-classification cross entropy loss function. The weight update formula is as follows: ; where α represents the weight update learning rate. In the Adam optimization method used, the learning rate is set to 0.001. is the current weight value under sample i, is the new partial derivative weight value under sample i, is the partial derivative coefficient, is the number of predicted samples, The actual number of samples collected.

[0090] like Figure 4 , which is a structural block diagram of a benthic animal intelligent identification device proposed by the present invention, comprising:

[0091] A distance unit, used to calculate the distance information from the bottom of the water by using an ultrasonic distance sensor preset on the benthic exploration equipment, and start the visual acquisition processing unit corresponding to the distance information, wherein the instruction corresponding to starting the visual acquisition processing unit is generated by judging whether the distance information reaches a preset threshold;

[0092] An analysis unit, used to instruct the visual acquisition processing unit to obtain water environment image information, analyze whether the water environment image information meets the preset recognition rules, and if so, perform recognition work; otherwise, further search for a location that meets the preset recognition rules nearby;

[0093] The gyroscope unit collects gyroscope floating data using a preset gyroscope, performs pixel correction on water environment image information according to the gyroscope floating data, and performs image target enhancement processing on the corrected image.

[0094] The recognition unit is used to input the water environment image information into the frequency domain multi-scale feature fusion neural network model in the visual acquisition and processing unit to obtain benthic animal information after model processing, wherein the benthic animal information includes the types, geometric characteristics and vital sign activity of benthic animals on the bottom surface of the water area.

[0095] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0096] Reference Figure 5 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 5As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0097] Those skilled in the art will understand that Figure 5 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0098] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0099] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0100] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0101] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for intelligent identification of benthic animals, characterized in that: The following steps are involved: An ultrasonic distance sensor preset on the benthic exploration equipment is used to calculate the distance information from the water bottom, and a visual acquisition processing unit is started based on the distance information, wherein an instruction corresponding to starting the visual acquisition processing unit is generated by judging whether the distance information reaches a preset threshold; Instruct the visual acquisition processing unit to acquire water environment image information, analyze whether the water environment image information meets the preset recognition rules, and if so, perform recognition work; otherwise, further search for a location that meets the preset recognition rules nearby; Using a preset gyroscope to collect gyroscope floating data, and performing pixel correction on water environment image information according to the gyroscope floating data, and performing image target enhancement processing on the corrected image; Inputting the water environment image information into a frequency domain multi-scale feature fusion neural network model in a visual acquisition processing unit to obtain benthic animal information after model processing, wherein the benthic animal information includes the types, geometric features, and vital sign activity of benthic animals on the bottom surface of the water area; The frequency domain multi-scale feature fusion neural network model specifically includes: The frequency domain multi-scale feature fusion neural network model uses ACDNET as a feature extractor to extract the collected water environment image information; the ACDNET adopts a dense connection method to connect each layer with all previous layers respectively, and adopts asymmetric convolution, that is, for n*n convolution, convolutions of 1*n and n*1 convolution kernels are performed in parallel on its side, and then the three are fused; The output ACDNET features of the first layer, the third layer and the fifth layer of ACDNET are extracted respectively, and then the shallow features of the first layer and the third layer are input into the DCT module for processing. The DCT processing method includes: Perform discrete cosine transform to obtain the low-frequency component, horizontal high-frequency component, and vertical high-frequency component of the ACDNET feature; The horizontal high-frequency component and the vertical high-frequency component are aggregated and normalized using a softmax function to obtain detail features; Then multiply the detail features with the low-frequency components to obtain a detailed feature map; The low-frequency components are added to the detailed feature map to generate a complete feature map, and then the complete feature map enters the pyramid feature multi-scale fusion part; among them, the shallow layer of the ACDNET network is responsible for extracting shallow features, while the deep layer of the ACDNET network extracts deep features. The deep features are upsampled in turn and then fused with the complete feature map processed by the shallow layer through the DCT module, and gradually restored to the size of the water environment image information.

2. The method for intelligent identification of benthic animals according to claim 1, characterized in that: The step of performing pixel correction on the water environment image information according to the gyroscope floating data comprises: The gyroscope floating data is optimized by a preset first-order low-pass filter to construct an image rotation transformation matrix, and the gyroscope floating data is used to perform pixel point correction on the image pixel coordinates based on the image rotation transformation matrix, wherein the image rotation transformation matrix is ​​calculated as follows: ； in is the identity matrix, is the image rotation transformation matrix, Represents the angles of rotation around the x-axis, y-axis, and z-axis respectively. Indicates the overall rotation angle of the benthic exploration equipment.

3. The method for intelligent identification of benthic animals according to claim 2, characterized in that: The algorithm optimizes the gyroscope float data through a preset first-order low-pass filter, including: ； in, is the angular velocity measured by the gyroscope at the current moment, is the angular velocity measured by the gyroscope at the last moment, is the overall rotation angle from time m to time n, is the sampling period, is the cut-off frequency.

4. The method for intelligent identification of benthic animals according to claim 1, characterized in that: The step of correcting the pixel coordinate value of the image pixel coordinate by the visual acquisition processing unit through the gyroscope floating data also includes: Obtain the pixel coordinate values ​​of the image of the previous frame by optimizing the gyroscope floating data; Obtain pixel coordinate values ​​of the image of the intermediate frame during the gyroscope floating data optimization process; Obtain pixel coordinate values ​​of an image frame after gyroscope floating data optimization processing; Calculate the rotation matrix of the image of the frame before the gyroscope floating data optimization processing to the image of the frame after the gyroscope floating data optimization processing, and apply the rotation matrix to weightedly calculate the corresponding new pixel point coordinate value. The algorithm is: ； in is the camera intrinsic parameter matrix, is the rotation matrix, is the pixel coordinate point of the new image, is the pixel coordinate value of the previous frame image, is the pixel coordinate value of the image in the middle frame, is the pixel coordinate value of the next frame image, is the weight coefficient.

5. The method for intelligent identification of benthic animals according to claim 1, characterized in that: The steps of performing image target enhancement processing on the rectified image include: The corrected original image is obtained by performing pixel correction on the water environment image information collected by the visual collection and processing unit, and then image enhancement is performed on each pixel of the corrected original image. The algorithm of the image enhancement is: ;in, They correspond to the red, green and blue color channels in the water environment image respectively, is the minimum dark light value in the channel; ; ; ; Where A is the mapping position of the atmospheric light value, is the perspective ratio, is the red and blue light perspective ratio, w is the smoothness; ； in is the pixel value after image enhancement, is the pixel value of the original image, and H is the water depth measured by the ultrasonic distance sensor.

6. The method for intelligent identification of benthic animals according to claim 1, characterized in that: Algorithms for performing discrete cosine transforms include: ; in, It is an element of the discrete cosine transform matrix, which represents the frequency component of the signal; u and v are frequency indices, which represent the position of the frequency component in the frequency domain; is an M×N digital image matrix, M is the row of the digital image matrix, and N is the column of the digital image matrix; is the normalization coefficient; the obtained results are separated and extracted to obtain , respectively low-frequency component, horizontal high-frequency component and vertical high-frequency component; the entire DCT module is expressed as: ; Where Z is the output of the DCT module, the function G is the addition of elements, S is the softmax function, and δ is the multiplication of elements.

7. An intelligent identification device for benthic animals, characterized in that: include: A distance unit, used to calculate the distance information from the bottom of the water by using an ultrasonic distance sensor preset on the benthic exploration equipment, and start the visual acquisition processing unit corresponding to the distance information, wherein the instruction corresponding to starting the visual acquisition processing unit is generated by judging whether the distance information reaches a preset threshold; An analysis unit, used to instruct the visual acquisition processing unit to obtain water environment image information, analyze whether the water environment image information meets the preset recognition rules, and if so, perform recognition work; otherwise, further search for a location that meets the preset recognition rules nearby; A gyroscope unit collects gyroscope floating data using a preset gyroscope, performs pixel correction on water environment image information according to the gyroscope floating data, and performs image target enhancement processing on the corrected image; The recognition unit is used to input the water environment image information into the frequency domain multi-scale feature fusion neural network model in the visual acquisition and processing unit to obtain benthic animal information after model processing, wherein the benthic animal information includes the types, geometric characteristics and vital sign activity of benthic animals on the bottom surface of the water area.

8. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method for intelligent identification of benthic animals according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for intelligent identification of benthic animals according to any one of claims 1 to 6 are implemented.

Images