Autonomous acousto-optic fusion target detection method and system for amphibious unmanned platform

Abstract

The invention discloses an autonomous acousto-optic fusion target detection method and system for an amphibious unmanned platform, relates to the technical field of underwater target detection, and aims to solve the problem that a method for detecting a complex underwater environment target by combining an acoustic sensor and an optical sensor is lacked in the prior art. Comprising the following steps: S1, synchronously acquiring underwater environment information through an optical camera and a foresight sonar of the amphibious unmanned platform; s2, preprocessing the image data and the acoustic data respectively, and then performing segmentation and target identification on the preprocessed image data and acoustic data; s3, respectively converting target identification results under the optical camera coordinate system and the foresight sonar coordinate system to a joint coordinate system; and S4, establishing a support degree matrix between the target identification results of the optical camera and the foresight sonar, rejecting outliers with low support degree according to the obtained comprehensive support degree matrix of each data, and performing optimal weighted fusion on measured values with high support degree to obtain an acousto-optic combined underwater target positioning result.

Description

Technical Field

[0001] The present invention relates to the field of underwater target detection technology, and in particular to an autonomous acoustic-optical fusion target detection method and system for an amphibious unmanned platform. Background Art

[0002] In the field of underwater target detection, optical cameras and forward-looking sonar are the most commonly used detection sensors. However, both sensors have unique characteristics and are suitable for different purposes. The advantages of optical-based sensors come from high details and color information, but their detection distance is greatly affected by water quality and water body properties, and when moving close to the seabed or using a robotic arm, silt and sediment will rise, interfering with the performance of optical sensors. The advantages of acoustic-based sensors come from long detection range and little interference from impurities in the water, but they do not provide color information, and have blurred edges and low resolution. Both sensor systems have inherent problems that can reduce or increase their applicability under certain conditions, affecting detection accuracy and positioning precision. Therefore, there is an urgent need for an underwater target detection method that combines and fuses optical and acoustic sensors.

[0003] The patent document with publication number CN 116452965 and titled "A method for underwater target detection and recognition based on acoustic-optical fusion" discloses a method for underwater target detection and recognition based on acoustic-optical fusion applied to autonomous submersibles. The method implements underwater target detection and recognition through three steps: constructing a target acoustic-optical fusion data set, constructing a compositely connected reparameterized backbone network, and obtaining an available underwater target detection and recognition model. The method addresses the problem of low stability of small target detection by multimodal target detection algorithms and the problem of insufficient detection results of multimodal target detection algorithms, and uses an integrated model of optical and acoustic sensors to improve the accuracy of underwater environmental data collection. The method only targets underwater target detection and recognition where the underwater light intensity of the target set remains unchanged, and ignores the differences in feature information of each mode under different underwater lighting conditions, resulting in insufficient detection results.

[0004] Ge Huilin from Jiangsu University of Science and Technology proposed a new fusion model YOLOv7-Fusion in the journal "Underwater Target Detection Method Based on Improved YOLOv7 Acoustic-Optical Fusion", and introduced an acoustic-optical fusion underwater target detection method by introducing the CE-Fusion module. This method aims to solve the problems of low stability and insufficient detection accuracy of underwater target detection, and improves the real-time neural network detector architecture YOLOv7 to adapt it to multi-modal input. This algorithm and model can support real applications, but the research on the perception capabilities of optical cameras, sonar and other equipment in different types of underwater scenes is not sufficient. Summary of the invention

[0005] The technical problems to be solved by the present invention are:

[0006] In the existing technology, acoustic sensors have difficulty in constructing target contours and details, while optical sensors are not suitable for long-distance target detection and are greatly affected by water quality. There is a lack of methods that combine the two for target detection in complex underwater environments.

[0007] The present invention adopts the following technical solutions to solve the above technical problems:

[0008] The present invention provides an autonomous acoustic-optical fusion target detection method for an amphibious unmanned platform, comprising the following steps:

[0009] S1, synchronously collect underwater environmental information through the optical camera and forward-looking sonar of the amphibious unmanned platform to obtain underwater image data and acoustic data;

[0010] S2, preprocessing the image data and acoustic data respectively, and then performing segmentation and target recognition on the preprocessed image data and acoustic data;

[0011] S3, converting the target recognition results in the optical camera coordinate system and the forward-looking sonar coordinate system into the joint coordinate system respectively;

[0012] S4. Establish a support matrix between the target recognition results of the optical camera and the forward-looking sonar. Eliminate outliers with low support based on the comprehensive support matrix of each data, and perform optimal weighted fusion on the measurement values ​​with high support to obtain the positioning result of the underwater target by the combination of sound and light.

[0013] Furthermore, the preprocessing of image data in S2 includes: graying, filtering and dehazing preprocessing of image data, wherein the dehazing preprocessing estimates the attenuation of each color channel in the image, adaptively constructs a dark channel prior map to estimate the transmittance, and then dehazes the data; the preprocessing of acoustic data includes: graying and filtering preprocessing.

[0014] Furthermore, bilateral filtering is used in S2 to filter the data. The specific calculation process is:

[0015]

[0016] Where (x, y) is the coordinate of the center point of the window; filtered (x, y) is the pixel value of the center point of the window after filtering; W p (x, y) is the normalization weight used to normalize the filtering result; Ω is the neighborhood window of the filter; I(x, y) is the pixel value at the center of the original image window; F spatial (i, j) is the spatial domain kernel function, which is used to calculate the spatial distance weight between pixels; F range(I(x,y),I(x+i,y+j)) is the pixel value domain kernel function, which is used to calculate the similarity weight between pixel values; variance σ s , σ r The spatial domain and pixel value domain weights are defined.

[0017] Furthermore, in S2, the image data is pre-processed for defogging, which specifically includes the following processes:

[0018] First, the underwater optical imaging model is constructed as:

[0019] I λ (x) = J λ (x)t λ (x)+B λ (1-t λ (x)); λ∈{r,g,b}

[0020] Where x is the two-dimensional coordinate of a point in the image data; λ is a color channel of the RGB channel; I λ (x) is the image formed by the optical camera; J λ (x) is the image before attenuation and scattering; B λ is the global background light in water medium; t λ (x) is the transmittance of light propagating in water medium;

[0021] The estimated attenuation of the image color channel is:

[0022] η=argmin(ps λ )stη∈{r,g,b}

[0023] Where η is the color channel with the largest attenuation;

[0024] ps λ =sum(I λ )stλ∈{r,g,b}

[0025] Where ps is the total number of pixel values ​​in a certain color channel; I is the original image;

[0026] The dark channel prior map is:

[0027]

[0028] The estimated transmittance is:

[0029]

[0030] In the formula is the transmittance obtained by estimating the attenuation degree of each channel;

[0031] The final restored image is:

[0032]

[0033] Where t 0 is the minimum transmittance.

[0034] Furthermore, in S2, the preprocessed image data and acoustic data are segmented, specifically, the preprocessed data is segmented using the maximum inter-class variance threshold segmentation method to obtain the target region of interest.

[0035] Furthermore, S2 also includes performing morphological processing on the segmented image data and acoustic data.

[0036] Furthermore, in S2, target recognition is performed on the data, specifically, a Gaussian mixture model classifier is used for target recognition.

[0037] Furthermore, in S3, the target recognition results in the optical camera coordinate system and the forward-looking sonar coordinate system are respectively converted to the joint coordinate system as follows:

[0038] The transformation matrix from the optical camera coordinate system to the joint coordinate system is:

[0039]

[0040] Where O c X c Y c Z c is the camera coordinate system; O u X u Y u Z u is the joint coordinate system; R c Indicates O c X c Y c Z c With O u X u Y u Z u The rotation relationship between c Indicates O c X c Y c Z c With O u X u Y u Z u The translation relationship between

[0041] The transformation matrix from the forward-looking sonar coordinate system to the joint coordinate system is:

[0042]

[0043] Where O sX s Y s Z s is the camera coordinate system; O u X u Y u Z u is the joint coordinate system; R s Indicates O s X s Y s Z s With O u X u Y u Z u The rotation relationship between s Indicates O s X s Y s Z s With O u X u Y u Z u The translation relationship between them.

[0044] Further, S4 includes the following steps:

[0045] First, the deviation between the target recognition result data of the confidence distance optical camera and the target recognition result data of the forward-looking sonar is used, that is, the relative position between the targets is calculated as:

[0046]

[0047] Where (x m ,y m ,z m ) and (x n ,y n ,z n ) is the mth data and nth data measured by the sensor; d mn is the relative Euclidean distance between the mth data and the nth data;

[0048] The confidence distance matrix is ​​calculated as:

[0049]

[0050] Then, introduce a mn The negatively correlated support function represents the mutual support relationship between the measurement data, that is:

[0051]

[0052] Where r mn Indicates the mutual support degree between the mth data and the nth data;

[0053] The support matrix is ​​calculated as:

[0054]

[0055] Finally, a comprehensive support function is defined to reflect the comprehensive support of the data by all other measurement data, namely:

[0056]

[0057] Where s m Indicates the comprehensive support degree of the mth data by all other measurement data;

[0058] Finally, the comprehensive support matrix of each measurement data is obtained as follows:

[0059] S=[s 1 ,s 2 ,…,s m ,s i ] T

[0060] According to the comprehensive support matrix of each measurement data, outliers with low support are eliminated, and the measurement values ​​with high support are optimally weighted fused to obtain the positioning result of underwater targets by combining acoustic and optical methods.

[0061] An autonomous sound and light fusion target detection system for an amphibious unmanned platform, the system having a program module corresponding to the steps of the method described in any one of the above technical solutions, and executing the steps of the autonomous sound and light fusion target detection method for an amphibious unmanned platform during operation.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] The present invention is based on an amphibious unmanned platform and uses multiple sensors to realize a complete underwater target detection method combined with acoustic and optical sensors through time-space registration, data processing, data fusion, etc. In the data fusion process, the wild values ​​with low support are eliminated according to the comprehensive support matrix of the data, and the measurement values ​​with high support are optimally weighted fused to obtain the positioning result of the underwater target combined with acoustic and optical sensors with high accuracy.

[0064] The present invention makes full use of multi-source data related to the target position, and through information fusion and the method of learning from each other's strengths and making up for each other's weaknesses, enables the detection system to obtain target-related characteristics more comprehensively and accurately, thereby improving the target positioning accuracy of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a flow chart of the autonomous sound and light fusion target detection method of the amphibious unmanned platform in an embodiment of the present invention;

[0066] Figure 2is a time registration diagram in an embodiment of the present invention;

[0067] Figure 3 is a schematic diagram of an optical camera coordinate system in an embodiment of the present invention;

[0068] Figure 4 Schematic diagram of a forward-looking sonar coordinate system in an embodiment of the present invention;

[0069] Figure 5 is a schematic diagram of a joint coordinate system in an embodiment of the present invention;

[0070] Figure 6 is an image processing flow chart in an embodiment of the present invention;

[0071] Figure 7 The image data and acoustic data of the underwater environment information collected by the optical camera and the forward-looking sonar in the embodiment of the present invention;

[0072] Figure 8 This is a diagram of the final target recognition result of the image data and acoustic data in an embodiment of the present invention. DETAILED DESCRIPTION

[0073] In order to enable those skilled in the art to better understand the scheme of the present invention, exemplary implementations or embodiments of the present invention will be described below in conjunction with the accompanying drawings. Obviously, the described implementations or embodiments are only implementations or embodiments of a part of the present invention, not all of them. Based on the implementations or embodiments of the present invention, all other implementations or embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present invention.

[0074] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0075] Combination Figures 1 to 6 As shown, the present invention provides an autonomous acoustic-optical fusion target detection method for an amphibious unmanned platform, comprising the following steps:

[0076] S1. The underwater environment information is collected synchronously by the optical camera and forward-looking sonar of the amphibious unmanned platform to obtain underwater image data and acoustic data. Figure 2 As shown, time synchronized sampling can unify sensors with faster sampling rates onto sensors with slower sampling rates.

[0077] S2. Preprocess the image data and acoustic data respectively, and then segment and identify the target on the preprocessed image data and acoustic data.

[0078] like Figure 6As shown, the preprocessing of image data includes: graying, filtering and dehazing preprocessing of image data, wherein the dehazing preprocessing estimates the attenuation of each color channel in the image, adaptively constructs a dark channel prior map to estimate the transmittance, and then dehazes the data; the preprocessing of acoustic data includes: graying and filtering preprocessing.

[0079] The component method is used to grayscale the data, and the brightness of the three components in the color image is used as the grayscale values ​​of the three grayscale images:

[0080] f 1 (i,j)=R(i,j)

[0081] f 2 (i,j)=G(i,j)

[0082] f 3 (i,j)=B(i,j)

[0083] where f k (i, j)(k=1,2,3) is the grayscale value of the converted grayscale image at (i, j).

[0084] Bilateral filtering is used to filter the data. The specific calculation method is:

[0085]

[0086]

[0087] Where (x, y) is the coordinate of the center point of the window; filtered (x, y) is the pixel value of the center point of the window after filtering; W p (x, y) is the normalization weight used to normalize the filtering result; Ω is the neighborhood window of the filter; I(x, y) is the pixel value at the center of the original image window; F spatial (i, j) is the spatial domain kernel function, which is used to calculate the spatial distance weight between pixels; F range (I(x,y),I(x+i,y+j)) is the pixel value domain kernel function, which is used to calculate the similarity weight between pixel values; variance σ s , σ r The spatial domain and pixel value domain weights are defined.

[0088] By combining the spatial proximity and pixel value similarity of the image through filtering processing, and considering the spatial domain and pixel value domain information at the same time, it can effectively protect the edge characteristics of the image while smoothing the image, which helps to retain the details of the image.

[0089] Dehazing the image data includes the following steps:

[0090] First, the underwater optical imaging model is constructed as:

[0091] I λ (x) = J λ (x)t λ (x)+B λ (1-t λ (x)); λ∈{r,g,b}

[0092] Where x is the two-dimensional coordinate of a point in the optical image; λ is a color channel of the RGB channel; I λ (x) is the image formed by the optical device; J λ (x) is the image before attenuation and scattering; B λ is the global background light in water medium; t λ (x) is the transmittance of light propagating in water medium;

[0093] The estimated attenuation of the image color channel is:

[0094] η=argmin(ps λ )stη∈{r,g,b}

[0095] Where η is the color channel with the largest attenuation;

[0096] ps λ =sum(I λ )stλ∈{r,g,b}

[0097] Where ps is the total number of pixel values ​​of a certain color channel; I is the original image;

[0098] The dark channel prior map is:

[0099]

[0100] The estimated transmittance is:

[0101]

[0102] Where t(x) is the transmittance obtained by estimating the attenuation degree of each channel.

[0103] The final restored image is:

[0104]

[0105] Where t 0 is the minimum transmittance.

[0106] Finally, considering that the red channel is severely attenuated underwater and is almost zero in some scenes, the range of the prior dark channel is selected as:

[0107]

[0108] Where J UDCP (x) is the dark channel prior result calculated by discarding the red channel;

[0109] The range of transmittance is selected as:

[0110]

[0111] In the formula is the transmittance in the underwater dark channel prior when only the GB channel is considered.

[0112] The present invention adaptively constructs a dark channel priori map to estimate the transmittance, and then performs defogging on the image to achieve a more accurate image processing result.

[0113] The maximum inter-class variance threshold segmentation method is used to segment the preprocessed image data and acoustic data. The grayscale value of the image is divided into two parts with the optimal threshold, so that the variance between the two parts is maximized, that is, it has the greatest separation. This method can well separate the target area. The data is morphologically processed, and the highlight area or white part in the data is amplified and coarsened by dilation processing, and then the highlight area or white part in the data is reduced and refined by corrosion processing.

[0114] Gaussian mixture model (GMM) classifier is used for target recognition, specifically:

[0115] Suppose there is a random variable x, then the Gaussian mixture model is:

[0116]

[0117] Where K is the total number of components; N(x|u k ,Σ k ) is the Gaussian distribution probability density function, that is, the kth component in the Gaussian mixture model; π k is the mixing coefficient.

[0118] The EM algorithm is used to determine the GMM parameters:

[0119] The first step is to define the number of components K and set the initial value π for each component k 、u k ,Σ k , and then calculate the log-likelihood function:

[0120]

[0121] The second step is to estimate the generation probability of each component of the data, that is, according to the current π k 、u k ,Σk Calculate the posterior probability γ(z ik ):

[0122]

[0123] Step 3: According to the calculated γ(z ik ) and then calculate the new

[0124]

[0125] The fourth step is to calculate the log-likelihood function and check whether the parameters converge or whether the log-likelihood function converges; if converged, the parameter estimation result is obtained; if not, return to the second step to continue iterating.

[0126] The complex underwater environment data distribution is modeled through the classifier, which effectively processes the nonlinearly separable data and realizes the classification and identification of targets in the data.

[0127] S3. Convert the target recognition results in the optical camera coordinate system and the forward-looking sonar coordinate system to the joint coordinate system respectively.

[0128] By obtaining the conversion relationship between the optical camera and forward-looking sonar coordinate systems, a joint coordinate system is established, that is, Figure 3-5 The optical camera, forward-looking sonar and joint coordinate system shown in the figure are used to establish the joint coordinate system by obtaining the conversion relationship between the optical camera and forward-looking sonar coordinate systems, and complete the time-space registration. The conversion matrix from the optical camera coordinate system to the joint coordinate system is:

[0129]

[0130] Where O c X c Y c Z c is the camera coordinate system; O u X u Y u Z u is the joint coordinate system; R c Indicates O c X c Y c Z c With O u X u Y u Z u The rotation relationship between c Indicates O c X c Y c Z c With O u X u Yu Z u The translation relationship between them.

[0131] The transformation matrix from the forward-looking sonar coordinate system to the joint coordinate system is:

[0132]

[0133] Where O s X s Y s Z s is the camera coordinate system; O u X u Y u Z u is the joint coordinate system; R s Indicates O s X s Y s Z s With O u X u Y u Z u The rotation relationship between s Indicates O s X s Y s Z s With O u X u Y u Z u The translation relationship between them.

[0134] S4, establishing a support matrix between the target recognition results of the optical camera and the forward-looking sonar, eliminating wild values ​​with low support according to the comprehensive support matrix of each data, and performing optimal weighted fusion on the measurement values ​​with high support to obtain the positioning result of the underwater target by the combination of sound and light, including the following steps:

[0135] First, the deviation between the target recognition result data of the confidence distance optical camera and the target recognition result data of the forward-looking sonar is used, that is, the relative position between the targets is calculated as:

[0136]

[0137] Where (x m ,y m ,z m ) and (x n ,y n ,z n ) is the mth data and nth data measured by the sensor; d mn is the relative Euclidean distance between the mth data and the nth data;

[0138] The confidence distance matrix is ​​calculated as:

[0139]

[0140] Then, introduce a mn The negatively correlated support function represents the mutual support relationship between the measurement data, that is:

[0141]

[0142] Where r mn Indicates the mutual support between the mth data and the nth data; r mn ∈[0,1].

[0143] The support matrix is ​​calculated as:

[0144]

[0145] Finally, a comprehensive support function is defined to reflect the comprehensive support of the data by all other measurement data, namely:

[0146]

[0147] Where s m Indicates the comprehensive support degree of the mth data by all other measurement data;

[0148] Finally, the comprehensive support matrix of each measurement data is obtained as follows:

[0149] S=[s 1 ,s 2 ,…,s m ,s i ] T

[0150] According to the comprehensive support matrix of each measurement data, outliers with low support are eliminated, and the measurement values ​​with high support are optimally weighted fused to obtain the positioning result of underwater targets by combining acoustic and optical methods.

[0151] The autonomous acoustic-optical fusion target detection method (algorithm) for an amphibious unmanned platform proposed in the present invention is the underlying technical core of the present invention, and various products can be derived based on the algorithm.

[0152] Based on the method proposed in the present invention, an autonomous sound and light fusion target detection system for an amphibious unmanned platform is developed using a programming language. The system has program modules corresponding to the steps of the above-mentioned technical solution, and executes the steps in the above-mentioned autonomous sound and light fusion target detection method for an amphibious unmanned platform during operation.

[0153] Example 1

[0154] At the same depth underwater, the optical camera and forward-looking sonar of the amphibious unmanned platform along a zigzag path synchronously collect image data and acoustic data multiple times. Some of the data are as follows: Figure 7 As shown in the figure, the platform is always facing the target during data collection. The target position is (6.80, 2.60, 1.30).

[0155] Target recognition based on underwater image data and acoustic data includes: graying, filtering and defogging preprocessing of image data and acoustic data respectively. The defogging preprocessing estimates the attenuation of each color channel in the image, adaptively constructs a dark channel prior map to estimate the transmittance, and then defogs the data. Finally, the preprocessed data is segmented and the target is recognized. The above processing and recognition are performed on the image data collected at different locations. The results are as follows: Figure 8 shown.

[0156] The target recognition results in the optical camera coordinate system and the forward-looking sonar coordinate system are converted to the joint coordinate system through the coordinate system conversion matrix; the results are shown in Tables 1 and 2.

[0157] Table 1. Coordinates of the target measured by the optical camera in the joint coordinate system

[0158]

[0159] Table 2. Coordinates of the target measured by forward-looking sonar in the joint coordinate system

[0160]

[0161]

[0162] Finally, the support matrix between the target recognition results of the optical camera and the forward-looking sonar is established. The outliers with low support are eliminated according to the comprehensive support matrix of each data, and the measurement values ​​with high support are optimally weighted fused to obtain the positioning results of the underwater target by the combination of sound and light. The results are shown in Table 3.

[0163] Table 3. Coordinates of targets obtained by acoustic-optical joint target detection in the joint coordinate system

[0164]

[0165] It can be seen that this embodiment achieves underwater target detection with higher accuracy compared with a single acoustic sensor or optical sensor.

[0166] Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the protection scope of the present invention.

Claims

1. An autonomous acoustic-optical fusion target detection method for an amphibious unmanned platform, characterized in that: The steps include: S1, synchronously collect underwater environmental information through the optical camera and forward-looking sonar of the amphibious unmanned platform to obtain underwater image data and acoustic data; S2, preprocessing the image data and acoustic data respectively, and then performing segmentation and target recognition on the preprocessed image data and acoustic data; S3, converting the target recognition results in the optical camera coordinate system and the forward-looking sonar coordinate system into the joint coordinate system respectively; S4. Establish a support matrix between the target recognition results of the optical camera and the forward-looking sonar. Eliminate outliers with low support based on the comprehensive support matrix of each data, and perform optimal weighted fusion on the measurement values ​​with high support to obtain the positioning result of the underwater target by the combination of sound and light.

2. The autonomous sound and light fusion target detection method for an amphibious unmanned platform according to claim 1 is characterized in that: The preprocessing of image data in S2 includes: graying, filtering and dehazing preprocessing of image data. The dehazing preprocessing estimates the attenuation of each color channel in the image, adaptively constructs a dark channel prior map to estimate the transmittance, and then dehazes the data; the preprocessing of acoustic data includes: graying and filtering preprocessing.

3. The autonomous sound and light fusion target detection method for an amphibious unmanned platform according to claim 2 is characterized in that: In S2, bilateral filtering is used to filter the data. The specific calculation process is: Where (x, y) is the coordinate of the center point of the window; filtered (x, y) is the pixel value of the center point of the window after filtering; W p (x, y) is the normalization weight, which is used to normalize the filtering result; Ω is the neighborhood window of the filter; I(x,y) is the pixel value at the center of the original image window; F spatial (i, j) is the spatial domain kernel function, which is used to calculate the spatial distance weight between pixels; F range (I(x,y),I(x+i,y+j)) is the pixel value domain kernel function, which is used to calculate the similarity weight between pixel values; variance σ s , σ r The spatial domain and pixel value domain weights are defined.

4. The autonomous sound and light fusion target detection method for an amphibious unmanned platform according to claim 3 is characterized in that: In S2, the image data is pre-processed for defogging, which specifically includes the following processes: First, the underwater optical imaging model is constructed as: I λ (x)=J λ (x)t λ (x)+B λ (1-t λ (x));λ∈{r,g,b} Where x is the two-dimensional coordinate of a point in the image data; λ is a color channel of the RGB channel; I λ (x) is the image formed by the optical camera; J λ (x) is the image before attenuation and scattering; B λ is the global background light in water medium; t λ (x) is the transmittance of light propagating in water medium; The estimated attenuation of the image color channel is: η=argmin(ps λ )s.t.η∈{r,g,b} Where η is the color channel with the largest attenuation; ps λ =sum(I λ )s.t.λ∈{r,g,b} Where ps is the total number of pixel values ​​of a certain color channel; I is the original image; The dark channel prior map is: The estimated transmittance is: In the formula is the transmittance obtained by estimating the attenuation degree of each channel; The final restored image is: Where t0 is the minimum transmittance.

5. The autonomous sound and light fusion target detection method for an amphibious unmanned platform according to claim 4 is characterized in that: In S2, the preprocessed image data and acoustic data are segmented. Specifically, the preprocessed data is segmented using the maximum inter-class variance threshold segmentation method to obtain the target region of interest.

6. The autonomous sound and light fusion target detection method for an amphibious unmanned platform according to claim 5 is characterized in that: S2 also includes morphological processing of the segmented image data and acoustic data.

7. The autonomous sound and light fusion target detection method for an amphibious unmanned platform according to claim 6 is characterized in that: In S2, target recognition is performed on the data, specifically, a Gaussian mixture model classifier is used for target recognition.

8. The autonomous sound and light fusion target detection method for an amphibious unmanned platform according to claim 1 is characterized in that: In S3, the target recognition results in the optical camera coordinate system and the forward-looking sonar coordinate system are converted to the joint coordinate system respectively as follows: The transformation matrix from the optical camera coordinate system to the joint coordinate system is: Where O c X c Y c Z c is the camera coordinate system; O u X u Y u Z u is the joint coordinate system; R c Indicates O c X c Y c Z c With O u X u Y u Z u The rotation relationship between c Indicates O c X c Y c Z c With O u X u Y u Z u The translation relationship between The transformation matrix from the forward-looking sonar coordinate system to the joint coordinate system is: Where O s X s Y s Z s is the camera coordinate system; O u X u Y u Z u is the joint coordinate system; R s Indicates O s X s Y s Z s With O u X u Y u Z u The rotation relationship between s Indicates O s X s Y s Z s With O u X u Y u Z u The translation relationship between them.

9. The autonomous sound and light fusion target detection method for an amphibious unmanned platform according to claim 1 is characterized in that: S4 includes the following steps: First, the deviation between the target recognition result data of the confidence distance optical camera and the target recognition result data of the forward-looking sonar is used, that is, the relative position between the targets is calculated as: Where (x m ,y m ,z m ) and (x n ,y n ,z n ) is the mth data and nth data measured by the sensor; d mn is the relative Euclidean distance between the mth data and the nth data; The confidence distance matrix is ​​calculated as: Then, introduce a mn The negatively correlated support function represents the mutual support relationship between the measurement data, that is: Where r mn Indicates the mutual support degree between the mth data and the nth data; The support matrix is ​​calculated as: Finally, a comprehensive support function is defined to reflect the comprehensive support of the data by all other measurement data, namely: Where s m Indicates the comprehensive support degree of the mth data by all other measurement data; Finally, the comprehensive support matrix of each measurement data is obtained as follows: S=[s1,s2,…,s m ,…s i ] T According to the comprehensive support matrix of each measurement data, outliers with low support are eliminated, and the measurement values ​​with high support are optimally weighted fused to obtain the positioning result of underwater targets by combining acoustic and optical methods.

10. An autonomous acoustic-optical fusion target detection system for an amphibious unmanned platform, characterized in that: The system has a program module corresponding to the steps of the method described in any one of claims 1 to 9 above, and executes the steps in the autonomous sound and light fusion target detection method of the above-mentioned amphibious unmanned platform when running.

Images