A non-periodic sampling remote control method for unmanned ships

Abstract

The present invention relates to the technical field of data sampling, and discloses a non-periodic sampling remote control method for unmanned ships, the method comprising: enhancing river channel remote sensing image data; training a river channel image representation extraction model and extracting image representation vectors of enhanced river channel remote sensing image data; constructing a river channel bend recognition model and identifying the river channel bend; marking the river channel bend position in the river channel remote sensing image data at the actual geographical location; controlling the unmanned ship to perform key sampling when passing through the turning point of the river channel bend, and perform under-sampling when passing through other positions. The present invention performs linear stretching and pixel transformation on the river channel remote sensing image to improve pixel contrast, realize image enhancement, realize river channel bend position recognition based on the curve curvature of the core pixel coordinates in the river channel bend area image and the image representation vector, controls the unmanned ship to perform key sampling when passing through the river channel bend, and realizes non-periodic sampling remote control of the unmanned ship.

Description

Technical Field

[0001] The invention relates to the technical field of data sampling, and in particular to a non-periodic sampling remote control method for an unmanned ship. Background Art

[0002] The monitoring and maintenance of waterway regulation structures are the key to ensuring the stable and normal operation of waterways. Conventional monitoring methods are manual observation, photography, and measurement. For large-scale monitoring, manual observation and measurement are inefficient and costly. At the same time, some sections of the Yangtze River waterway have poor accessibility, making manual measurement difficult. By utilizing the photogrammetry method of unmanned ships, high-precision digital surface models and digital orthophotos of the survey area are obtained, which greatly reduces the manual workload and can complete waterway monitoring tasks in a short period of time, at low cost, and with high efficiency. However, unmanned ships can only collect waterway environment data periodically during the process of waterway data collection, and cannot adaptively sample according to the importance of the waterway environment. In response to this problem, the present invention proposes a non-periodic sampling remote control method for unmanned ships. Summary of the invention

[0003] In view of this, the present invention provides a non-periodic sampling remote control method for unmanned ships, the purpose of which is to: 1) use a linear stretching method to expand the overall pixel contrast of river remote sensing image data, and combine the information entropy of pixel values ​​to transform pixel values ​​of pixels in different intervals, so that the pixel values ​​of bright intervals are brighter and the color values ​​of dark intervals are darker, thereby improving the pixel contrast between river areas and riverbank areas in river remote sensing image data, realizing river remote sensing image data enhancement, and constructing positive and negative samples of the river respectively by random cropping and multi-angle transformation of images, so that images of the same river have a higher similarity, while images of different rivers have The model is trained with a lower similarity as the goal, and the image representation vector that can effectively distinguish different rivers is obtained through training; 2) the river bend position is recognized based on the curve curvature and image representation vector of the core pixel coordinates in the river bend area image, where the higher the curve curvature and the lower the similarity with the image representation vector of the adjacent image, the higher the probability of the river bend position, and the river bend position in the river remote sensing image data is matched in the river geographic reference image, and the unmanned ship is controlled to perform key sampling when passing the turning point of the river bend and under-sample when passing other positions, so as to realize non-periodic sampling and remote control of the unmanned ship.

[0004] To achieve the above object, the present invention provides a non-periodic sampling remote control method for an unmanned ship, comprising the following steps:

[0005] S1: Collect river remote sensing image data and enhance the river remote sensing image data;

[0006] S2: construct a river channel positive and negative sample pair data set to train a river channel image representation extraction model, and use the trained river channel image representation extraction model to extract image representation vectors of enhanced river channel remote sensing image data;

[0007] S3: constructing a river bend recognition model and identifying the river bend, wherein the river bend recognition model takes the image representation vector as input and takes the river bend judgment result as output;

[0008] S4: matching the river bend position in the river remote sensing image data with the river geographic reference image to obtain the actual coordinate position of the river bend;

[0009] S5: Control the unmanned vessel to perform key sampling when passing through the turning point of the river bend, and perform under-sampling when passing through other locations.

[0010] As a further improvement method of the present invention:

[0011] Optionally, the collecting of river channel remote sensing image data and enhancing of river channel remote sensing image data in step S1 include:

[0012] The remote sensing image data I of the river channel is collected by using the remote sensing device carried by the remote sensing platform, wherein the remote sensing platform is an artificial satellite and the remote sensing device is a multispectral scanner. The collection process of the river channel remote sensing image data is as follows:

[0013] Select filters corresponding to the red band, blue band and green band respectively for projection synthesis in the remote sensor, where the wavelength of the selected red band is 0.66um, the wavelength of the blue band is 0.45um, and the wavelength of the green band is 0.51um;

[0014] The infrared radiation and visible light radiation emitted from the river area will enter the remote sensor and obtain the river remote sensing image data I through the filter after projection synthesis;

[0015] The river remote sensing image data is enhanced, and the enhancement process is as follows:

[0016] S11: grayscale processing is performed on the river remote sensing image data I, wherein the grayscale processing formula is:

[0017] ;

[0018] in:

[0019] Represents pixels Color value in RGB color channels; pixel represents the pixel in the i-th row and j-th column in the river remote sensing image data I;

[0020] Represents pixels Gray value of

[0021] The gray value of the pixel is used as the pixel value of the corresponding pixel to obtain the grayscale processing result of the river remote sensing image data. ;

[0022] S12: Remote sensing image data of river channels The grayscale value of the pixel in is linearly stretched, where the grayscale value The linear stretching processing formula is:

[0023] ;

[0024] in:

[0025] is a preset threshold parameter; in the embodiment of the present invention, , ;

[0026] Represents river remote sensing image data The first quartile of the pixel gray value, Represents river remote sensing image data The third quartile of the gray value of the middle pixel;

[0027] Represents grayscale value The linear stretching result of

[0028] S13: Calculate the river channel remote sensing image data after linear stretching Information entropy :

[0029] ;

[0030] in:

[0031] Represents river channel remote sensing image data after linear stretching The proportion of pixels with gray value L;

[0032] S14: Remote sensing image data of river channels The gray value of the pixel is enhanced to obtain the enhanced river remote sensing image data , where the gray value The enhancement formula is:

[0033] ;

[0034] in:

[0035] Gray value The enhanced results;

[0036] The enhanced grayscale value is used as the pixel value of the corresponding pixel to form the enhanced river remote sensing image data .

[0037] Optionally, in step S2, constructing river channel positive and negative sample pair data and training a river channel image representation extraction model includes:

[0038] The river channel positive and negative sample pair data are constructed, and the steps for constructing the river channel positive and negative sample pair data are as follows:

[0039] M pieces of local remote sensing image data of the river channel are collected to form positive samples, wherein the collected local remote sensing image data of the river channel are local remote sensing image data of the river channel after enhancement processing and only include remote sensing image data of the river channel portion;

[0040] For any m-th river local remote sensing image data Perform two random cuts to obtain Two positive samples of , and perform random chromaticity transformation, rotation, color distortion and regional blurring on the two positive samples to obtain the local remote sensing image data of the river channel The corresponding M-2 different negative samples ;

[0041] Construct the river positive and negative sample data ;

[0042] Constructing a river channel image representation extraction model, the constructed river channel image representation extraction model includes an input layer, a feature extraction module, a feature mapping module and an output layer, wherein the input layer is used to receive the enhanced river channel remote sensing image data and perform image segmentation processing, and the plurality of sub-image data obtained by segmentation are sequentially input into the feature extraction module, the feature extraction module is used to perform encoding processing on the received image data, and the feature mapping module is used to perform feature mapping on the encoding processing result, and the mapped image representation vector is output;

[0043] The river channel image representation extraction model is trained based on the river channel positive and negative sample data. The parameters to be trained of the river channel image representation extraction model are the weight parameters and bias parameters in the feature extraction module and the mapping matrix in the feature mapping module. The training process of the river channel image representation extraction model is as follows:

[0044] S21: Construct the training objective function of the river image representation extraction model:

[0045] ;

[0046] ;

[0047] ;

[0048] ;

[0049] ;

[0050] in:

[0051] T stands for transpose;

[0052] represents the L1 norm, represents an exponential function with a natural constant as base;

[0053] represents the training objective function of the river channel image representation extraction model, are the weight parameters and bias parameters in the feature extraction module, respectively. is the mapping matrix in the feature mapping module;

[0054] To make positive samples The encoding result obtained by inputting into the feature extraction module, Represents a positive sample Image representation vector, j=1,2;

[0055] Represents the mapping matrix The second derivative of Represents the mapping matrix The first derivative of ReLU activation function.

[0056] express The similarity between

[0057] It represents the sum of similarities between the negative samples of any m-th river local remote sensing image data and the negative samples of different river local remote sensing image data;

[0058] S22: Randomly generate initialized model parameters , set the current iteration number of parameter training to t, and the maximum iteration number to Max, then the model parameters obtained in the tth iteration are ,make ;

[0059] S23: If , then As the model parameters obtained by training, and based on The river channel image representation extraction model is constructed, otherwise, the process goes to step S24, where represents the preset gradient threshold, represents the L1 norm;

[0060] S24: Generate gradient update step size :

[0061] ;

[0062] in:

[0063] , represents the L2 norm;

[0064] S25: Generate an iteration step length that satisfies the following formula :

[0065] ;

[0066] S26: Update model parameters based on gradient step size and iteration step size:

[0067] ;

[0068] Let t=t+1 and return to step S23.

[0069] Optionally, in step S2, extracting the image representation vector of the enhanced river channel remote sensing image data using the trained river channel image representation extraction model includes:

[0070] The trained river channel image representation extraction model is used to extract the image representation vector of the enhanced river channel remote sensing image data, where the image representation vector extraction process is as follows:

[0071] The input layer receives enhanced river remote sensing image data , and divide the river remote sensing image data into K sub-image data of equal size, where the kth sub-image data obtained by segmentation is , ;

[0072] The feature extraction module extracts the received sub-image data Perform encoding processing, where the encoding processing formula is:

[0073] ;

[0074] in:

[0075] Represents sub-image data The encoding processing result;

[0076] They are the weight parameters and bias parameters in the feature extraction module after model training;

[0077] The feature mapping module processes the encoding results Perform feature mapping and map the image representation vector for output, where the mapping formula is:

[0078] ;

[0079] in:

[0080] It is the mapping matrix in the feature mapping module after model training;

[0081] Represents sub-image data The image representation vector.

[0082] Optionally, constructing a river bend recognition model and identifying a river bend in step S3 includes:

[0083] A river bend recognition model is constructed, and the river bend area in the river remote sensing image data is identified based on the river bend recognition model. The river bend recognition process is as follows:

[0084] S31: Calculate the image representation vector similarity of adjacent sub-image data in the river remote sensing image data to form an image representation vector similarity set, the sub-image data The similarity set of image representation vectors with adjacent sub-image data is:

[0085] ;

[0086] in:

[0087] Sub-image data The similarity with the image representation vector of the u-th adjacent sub-image data, wherein the image representation vector similarity is calculated by using a cosine similarity algorithm to perform similarity calculation on the image representation vector;

[0088] S32: Select the result with the lowest similarity in the image representation vector similarity set as the neighboring image representation vector similarity of the sub-image data, where the sub-image data The similarity of the neighboring image representation vector is ;

[0089] S33: if the similarity of the representation vectors of the adjacent images is less than a preset similarity threshold, marking the corresponding sub-image data as a candidate river bend area image;

[0090] S34: For any pixel in the candidate river bend region image, take the pixel as the center In the pixel area, the pixel value of the central pixel is compared with the pixel values ​​of the adjacent 8 pixels in the order from left to right and from top to bottom. If the pixel value of the central pixel is greater than the pixel value of the adjacent pixel, 0 is output, otherwise 1 is output to obtain an 8-bit binary number, and the 8-bit binary number is converted to a decimal number. If the decimal number conversion result is greater than the preset pixel threshold, the corresponding pixel is marked as a core pixel;

[0091] S35: Using the cubic spline interpolation method, the core pixel coordinates in the candidate river bend area image are interpolated to obtain an interpolation curve , where x represents the horizontal coordinate of the interpolation curve, is the vertical coordinate corresponding to the horizontal coordinate;

[0092] S36: Convert the interpolation curve to a quantization function:

[0093] ;

[0094] Measure the curvature of the core pixel coordinates in the candidate river bend area image:

[0095] ;

[0096] in:

[0097] num represents the number of core pixels in the candidate river bend region image;

[0098] Represents the horizontal coordinate of the rth core pixel in the candidate river bend area image;

[0099] The curvature of the curve representing the core pixel coordinates in the candidate river bend region image;

[0100] If C is greater than a preset curve curvature threshold, the candidate river channel bend area image is marked as a river channel bend location.

[0101] Optionally, in step S4, matching the river channel bend position in the river channel remote sensing image data with the river channel geographic reference image includes:

[0102] The river bend position in the river remote sensing image data is matched in the river georeferenced image, where the river bend position matching process is as follows:

[0103] Grayscale the georeferenced image of the river channel;

[0104] Extract the SIFT feature vector of each pixel in the river georeferenced image after grayscale processing;

[0105] Extract SIFT feature vectors of core pixels at the river bend position in river remote sensing image data;

[0106] The feature similarity matching is performed on the SIFT feature vectors of the pixels in the river channel georeferenced image and the river channel bend position, so that the feature similarity after the overall matching is maximized, wherein the feature similarity calculation method is the cosine similarity algorithm;

[0107] The successfully matched river bend position is marked at the corresponding position of the river geographic reference image to obtain the real geographic coordinates of the river bend position.

[0108] Optionally, in step S5, controlling the unmanned vessel to perform key sampling when passing through a river bend turning point includes:

[0109] The unmanned ship is controlled to conduct key sampling when passing through the turning point of the river bend, and the balanced key sampling process of the unmanned ship is as follows:

[0110] S51: Calculate the ratio P of the area of ​​the river bend location to the river area;

[0111] S52: Determine the sampling rate of the river bend position according to the ratio P :

[0112] ;

[0113] S53: Control the unmanned ship to travel in the river. When it does not reach the river bend area, it will follow the sampling rate. Take river samples and when you reach the river bend area, follow the sampling rate Carry out river sampling; according to the sampling rate Sampling the river channel means controlling the unmanned ship every time it travels distance, Sampling, where Indicates the unit travel distance of the unmanned ship. In the embodiment of the present invention, Set to 5 km. Set to 1 km.

[0114] In order to solve the above problem, the present invention provides an electronic device, the electronic device comprising:

[0115] A memory storing at least one instruction;

[0116] Communication interface, enabling electronic equipment to communicate; and

[0117] The processor executes the instructions stored in the memory to implement the above-mentioned unmanned ship non-periodic sampling remote control method.

[0118] In order to solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored. The at least one instruction is executed by a processor in an electronic device to implement the above-mentioned unmanned ship non-periodic sampling remote control method.

[0119] Compared with the prior art, the present invention proposes a non-periodic sampling remote control method for unmanned ships, which has the following advantages:

[0120] First, this scheme proposes an image enhancement processing method to enhance the river remote sensing image data, wherein the enhancement process is: grayscale processing is performed on the river remote sensing image data I, wherein the grayscale processing formula is:

[0121] ;

[0122] in: Represents pixels Color value in RGB color channels; pixel represents the pixel in the i-th row and j-th column in the river remote sensing image data I; Represents pixels The grayscale value of the pixel is used as the pixel value of the corresponding pixel to obtain the grayscale processing result of the river remote sensing image data. ; Remote sensing image data of river channels The grayscale value of the pixel in is linearly stretched, where the grayscale value The linear stretching processing formula is:

[0123] ;

[0124] in: is the preset threshold parameter; Represents grayscale value Calculate the linear stretching processing results of the river remote sensing image data after linear stretching processing Information entropy :

[0125] ;

[0126] in: Represents river channel remote sensing image data after linear stretching The proportion of pixels with gray value L in the middle; for river remote sensing image data The gray value of the pixel is enhanced to obtain the enhanced river remote sensing image data , where the gray value The enhancement formula is:

[0127] ;

[0128] in: Gray value The enhanced result; the enhanced gray value is used as the pixel value of the corresponding pixel to form the enhanced river remote sensing image data This scheme uses a linear stretching method to expand the overall pixel contrast of river remote sensing image data, and combines the information entropy of pixel values ​​to transform pixel values ​​in different intervals, making the pixel values ​​in bright intervals brighter and the color values ​​in dark intervals darker, thereby improving the pixel contrast between river areas and riverbank areas in river remote sensing image data, and realizing river remote sensing image data enhancement. Positive and negative samples of the river are constructed respectively by random cropping and multi-angle transformation of images, and the model is trained with the goal that images of the same river have a higher similarity, while images of different rivers have a lower similarity, and the image representation vector that can effectively distinguish different rivers is obtained through training.

[0129] At the same time, this scheme proposes a river bend recognition and key sampling method, constructs a river bend recognition model, and identifies the river bend area in the river remote sensing image data based on the river bend recognition model. The river bend recognition process is as follows: the image representation vector similarity of adjacent sub-image data in the river remote sensing image data is calculated to form an image representation vector similarity set, and the sub-image data is The similarity set of image representation vectors with adjacent sub-image data is:

[0130] ;

[0131] in: Sub-image data The image representation vector similarity with the u-th adjacent sub-image data, wherein the image representation vector similarity is calculated by using the cosine similarity algorithm to calculate the similarity of the image representation vector; the result with the lowest similarity in the image representation vector similarity set is selected as the adjacent image representation vector similarity of the sub-image data, wherein the sub-image data The similarity of the neighboring image representation vector is ; If the similarity of the adjacent image representation vector is less than the preset similarity threshold, the corresponding sub-image data is marked as a candidate river bend area image; for any pixel in the candidate river bend area image, the pixel is taken as the center In the pixel area, the pixel value of the central pixel is compared with the pixel values ​​of the adjacent 8 pixels in the order from left to right and from top to bottom. If the pixel value of the central pixel is greater than the pixel value of the adjacent pixel, 0 is output, otherwise 1 is output to obtain an 8-bit binary number, and the 8-bit binary number is converted to a decimal number. If the decimal number conversion result is greater than the preset pixel threshold, the corresponding pixel is marked as a core pixel; the core pixel coordinates in the candidate river bend area image are interpolated using the cubic spline interpolation method to obtain the interpolation curve , where x represents the horizontal coordinate of the interpolation curve, is the vertical coordinate corresponding to the horizontal coordinate; convert the interpolation curve into a quantization function:

[0132] ;

[0133] Measure the curvature of the core pixel coordinates in the candidate river bend area image:

[0134] ;

[0135] Where: num represents the number of core pixels in the candidate river bend area image; Represents the horizontal coordinate of the rth core pixel in the candidate river bend area image; Indicates the curve curvature of the core pixel coordinates in the candidate river bend area image; if C is greater than the preset curve curvature threshold, the candidate river bend area image is marked as the river bend position. Control the unmanned ship to perform key sampling when passing through the turning point of the river bend, where the balanced key sampling process of the unmanned ship is: calculate the ratio P of the river bend position area to the river area; determine the sampling rate of the river bend position according to the ratio P :

[0136] ;

[0137] Control the unmanned ship to travel in the river. When it does not reach the bend area of ​​the river, it will follow the sampling rate. Take river samples and when you reach the river bend area, follow the sampling rate Carry out river sampling; according to the sampling rate Sampling the river channel means controlling the unmanned ship every time it travels distance, Sampling, where Indicates the unit travel distance of the unmanned ship. This scheme realizes the recognition of the river bend position based on the curve curvature of the core pixel coordinates in the river bend area image and the image representation vector. The higher the curve curvature, the lower the similarity with the image representation vector of the adjacent image, and the higher the probability of the river bend position. The river bend position in the river remote sensing image data is matched in the river geographic reference image, and the unmanned ship is controlled to perform key sampling when passing the turning point of the river bend, and under-sampling when passing other positions, so as to realize the non-periodic sampling remote control of the unmanned ship. BRIEF DESCRIPTION OF THE DRAWINGS

[0138] Figure 1 A flowchart of a non-periodic sampling remote control method for an unmanned ship provided by an embodiment of the present invention;

[0139] Figure 2 A schematic diagram of the structure of an electronic device for implementing a method for remotely controlling an unmanned ship by non-periodic sampling provided by an embodiment of the present invention.

[0140] 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

[0141] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0142] The embodiment of the present application provides a remote control method for non-periodic sampling of unmanned ships. The execution subject of the remote control method for non-periodic sampling of unmanned ships includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided in the embodiment of the present application. In other words, the remote control method for non-periodic sampling of unmanned ships can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0143] Embodiment 1:

[0144] S1: Collect river remote sensing image data and enhance the river remote sensing image data.

[0145] The step S1 includes collecting river channel remote sensing image data and enhancing the river channel remote sensing image data, including:

[0146] The remote sensing image data I of the river channel is collected by using the remote sensing device carried by the remote sensing platform, wherein the remote sensing platform is an artificial satellite and the remote sensing device is a multispectral scanner. The collection process of the river channel remote sensing image data is as follows:

[0147] Select filters corresponding to the red band, blue band and green band respectively for projection synthesis in the remote sensor, where the wavelength of the selected red band is 0.66um, the wavelength of the blue band is 0.45um, and the wavelength of the green band is 0.51um;

[0148] The infrared radiation and visible light radiation emitted from the river area will enter the remote sensor and obtain the river remote sensing image data I through the filter after projection synthesis;

[0149] The river remote sensing image data is enhanced, and the enhancement process is as follows:

[0150] S11: grayscale processing is performed on the river remote sensing image data I, wherein the grayscale processing formula is:

[0151] ;

[0152] in:

[0153] Represents pixels Color value in RGB color channels; pixel represents the pixel in the i-th row and j-th column in the river remote sensing image data I;

[0154] Represents pixels Gray value of

[0155] The gray value of the pixel is used as the pixel value of the corresponding pixel to obtain the grayscale processing result of the river remote sensing image data. ;

[0156] S12: Remote sensing image data of river channels The grayscale value of the pixel in is linearly stretched, where the grayscale value The linear stretching processing formula is:

[0157] ;

[0158] in:

[0159] is a preset threshold parameter; in the embodiment of the present invention, , ;

[0160] Represents river remote sensing image data The first quartile of the pixel gray value, Represents river remote sensing image data The third quartile of the gray value of the middle pixel;

[0161] Represents grayscale value The linear stretching result of

[0162] S13: Calculate the river channel remote sensing image data after linear stretching Information entropy :

[0163] ;

[0164] in:

[0165] Represents river channel remote sensing image data after linear stretching The proportion of pixels with gray value L;

[0166] S14: Remote sensing image data of river channels The gray value of the pixel is enhanced to obtain the enhanced river remote sensing image data , where the gray value The enhancement formula is:

[0167] ;

[0168] in:

[0169] Gray value The enhanced results;

[0170] The enhanced grayscale value is used as the pixel value of the corresponding pixel to form the enhanced river remote sensing image data .

[0171] S2: Construct a river channel positive and negative sample pair data set to train a river channel image representation extraction model, and use the trained river channel image representation extraction model to extract image representation vectors of enhanced river channel remote sensing image data.

[0172] In the step S2, the positive and negative sample pairs of river channels are constructed and the river channel image representation extraction model is trained, including:

[0173] The river channel positive and negative sample pair data are constructed, and the steps for constructing the river channel positive and negative sample pair data are as follows:

[0174] M pieces of local remote sensing image data of the river channel are collected to form positive samples, wherein the collected local remote sensing image data of the river channel are local remote sensing image data of the river channel after enhancement processing and only include remote sensing image data of the river channel portion;

[0175] For any m-th river local remote sensing image data Perform two random cuts to obtain Two positive samples of , and perform random chromaticity transformation, rotation, color distortion and regional blurring on the two positive samples to obtain the local remote sensing image data of the river channel The corresponding M-2 different negative samples ;

[0176] Construct the river positive and negative sample data ;

[0177] Constructing a river channel image representation extraction model, the constructed river channel image representation extraction model includes an input layer, a feature extraction module, a feature mapping module and an output layer, wherein the input layer is used to receive the enhanced river channel remote sensing image data and perform image segmentation processing, and the plurality of sub-image data obtained by segmentation are sequentially input into the feature extraction module, the feature extraction module is used to perform encoding processing on the received image data, and the feature mapping module is used to perform feature mapping on the encoding processing result, and the mapped image representation vector is output;

[0178] The river channel image representation extraction model is trained based on the river channel positive and negative sample data. The parameters to be trained of the river channel image representation extraction model are the weight parameters and bias parameters in the feature extraction module and the mapping matrix in the feature mapping module. The training process of the river channel image representation extraction model is as follows:

[0179] S21: Construct the training objective function of the river image representation extraction model:

[0180] ;

[0181] ;

[0182] ;

[0183] ;

[0184] ;

[0185] in:

[0186] T stands for transpose;

[0187] represents the L1 norm, represents an exponential function with a natural constant as base;

[0188] represents the training objective function of the river channel image representation extraction model, are the weight parameters and bias parameters in the feature extraction module, respectively. is the mapping matrix in the feature mapping module;

[0189] To make positive samples The encoding result obtained by inputting into the feature extraction module, Represents a positive sample Image representation vector, j=1,2;

[0190] Represents the mapping matrix The second derivative of Represents the mapping matrix The first derivative of ReLU activation function.

[0191] express The similarity between

[0192] It represents the sum of similarities between the negative samples of any m-th river local remote sensing image data and the negative samples of different river local remote sensing image data;

[0193] S22: Randomly generate initialized model parameters , set the current iteration number of parameter training to t, and the maximum iteration number to Max, then the model parameters obtained in the tth iteration are ,make ;

[0194] S23: If , then As the model parameters obtained by training, and based on The river channel image representation extraction model is constructed, otherwise, the process goes to step S24, where represents the preset gradient threshold, represents the L1 norm;

[0195] S24: Generate gradient update step size :

[0196] ;

[0197] in:

[0198] , represents the L2 norm;

[0199] S25: Generate an iteration step length that satisfies the following formula :

[0200]

[0201] S26: Update model parameters based on gradient step size and iteration step size:

[0202] ;

[0203] Let t=t+1 and return to step S23.

[0204] In the step S2, the image representation vector of the enhanced river channel remote sensing image data is extracted using the trained river channel image representation extraction model, including:

[0205] The trained river channel image representation extraction model is used to extract the image representation vector of the enhanced river channel remote sensing image data, where the image representation vector extraction process is as follows:

[0206] The input layer receives enhanced river remote sensing image data , and divide the river remote sensing image data into K sub-image data of equal size, where the kth sub-image data obtained by segmentation is , ;

[0207] The feature extraction module extracts the received sub-image data Perform encoding processing, where the encoding processing formula is:

[0208] ;

[0209] in:

[0210] Represents sub-image data The encoding processing result;

[0211] They are the weight parameters and bias parameters in the feature extraction module after model training;

[0212] The feature mapping module processes the encoding results Perform feature mapping and map the image representation vector for output, where the mapping formula is:

[0213] ;

[0214] in:

[0215] It is the mapping matrix in the feature mapping module after model training;

[0216] Represents sub-image data The image representation vector.

[0217] S3: constructing a river bend recognition model and identifying the river bend, wherein the river bend recognition model takes the image representation vector as input and takes the river bend judgment result as output.

[0218] In the step S3, a river bend recognition model is constructed and a river bend is identified, including:

[0219] A river bend recognition model is constructed, and the river bend area in the river remote sensing image data is identified based on the river bend recognition model. The river bend recognition process is as follows:

[0220] S31: Calculate the image representation vector similarity of adjacent sub-image data in the river remote sensing image data to form an image representation vector similarity set, the sub-image data The similarity set of image representation vectors with adjacent sub-image data is:

[0221] ;

[0222] in:

[0223] Sub-image data The similarity with the image representation vector of the u-th adjacent sub-image data, wherein the image representation vector similarity is calculated by using a cosine similarity algorithm to perform similarity calculation on the image representation vector;

[0224] S32: Select the result with the lowest similarity in the image representation vector similarity set as the neighboring image representation vector similarity of the sub-image data, where the sub-image data The similarity of the neighboring image representation vector is ;

[0225] S33: if the similarity of the representation vectors of the adjacent images is less than a preset similarity threshold, marking the corresponding sub-image data as a candidate river bend area image;

[0226] S34: For any pixel in the candidate river bend region image, take the pixel as the center In the pixel area, the pixel value of the central pixel is compared with the pixel values ​​of the adjacent 8 pixels in the order from left to right and from top to bottom. If the pixel value of the central pixel is greater than the pixel value of the adjacent pixel, 0 is output, otherwise 1 is output to obtain an 8-bit binary number, and the 8-bit binary number is converted to a decimal number. If the decimal number conversion result is greater than the preset pixel threshold, the corresponding pixel is marked as a core pixel;

[0227] S35: Using the cubic spline interpolation method, the core pixel coordinates in the candidate river bend area image are interpolated to obtain an interpolation curve , where x represents the horizontal coordinate of the interpolation curve, is the vertical coordinate corresponding to the horizontal coordinate;

[0228] S36: Convert the interpolation curve to a quantization function:

[0229] ;

[0230] Measure the curvature of the core pixel coordinates in the candidate river bend area image:

[0231] ;

[0232] in:

[0233] num represents the number of core pixels in the candidate river bend region image;

[0234] Represents the horizontal coordinate of the rth core pixel in the candidate river bend area image;

[0235] The curvature of the curve representing the core pixel coordinates in the candidate river bend region image;

[0236] If C is greater than a preset curve curvature threshold, the candidate river channel bend area image is marked as a river channel bend location.

[0237] S4: Matching the river bend position in the river remote sensing image data with the river geographic reference image to obtain the actual coordinate position of the river bend.

[0238] In the step S4, the river channel bend position in the river channel remote sensing image data is matched in the river channel geographic reference image, including:

[0239] The river bend position in the river remote sensing image data is matched in the river georeferenced image, where the river bend position matching process is as follows:

[0240] Grayscale the georeferenced image of the river channel;

[0241] Extract the SIFT feature vector of each pixel in the river georeferenced image after grayscale processing;

[0242] Extract SIFT feature vectors of core pixels at the river bend position in river remote sensing image data;

[0243] The feature similarity matching is performed on the SIFT feature vectors of the pixels in the river channel georeferenced image and the river channel bend position, so that the feature similarity after the overall matching is maximized, wherein the feature similarity calculation method is the cosine similarity algorithm;

[0244] The successfully matched river bend position is marked at the corresponding position of the river geographic reference image to obtain the real geographic coordinates of the river bend position.

[0245] S5: Control the unmanned vessel to perform key sampling when passing through the turning point of the river bend, and perform under-sampling when passing through other locations.

[0246] In step S5, the unmanned vessel is controlled to perform key sampling when passing through a bend in the river, including:

[0247] The unmanned ship is controlled to conduct key sampling when passing through the turning point of the river bend, and the balanced key sampling process of the unmanned ship is as follows:

[0248] S51: Calculate the ratio P of the area of ​​the river bend location to the river area;

[0249] S52: Determine the sampling rate of the river bend position according to the ratio P :

[0250] ;

[0251] S53: Control the unmanned ship to travel in the river. When it does not reach the river bend area, it will follow the sampling rate. Take river samples and when you reach the river bend area, follow the sampling rate Carry out river sampling; according to the sampling rate Sampling the river channel means controlling the unmanned ship every time it travels distance, Sampling, where Indicates the unit travel distance of the unmanned ship.

[0252] Embodiment 2:

[0253] like Figure 2 FIG. 1 is a schematic diagram of the structure of an electronic device for implementing a method for remotely controlling an unmanned ship by non-periodic sampling provided by an embodiment of the present invention.

[0254] The electronic device 1 may include a processor 10 , a memory 11 , a communication interface 13 and a bus, and may also include a computer program stored in the memory 11 and executable on the processor 10 , such as a program 12 .

[0255] The memory 11 includes at least one type of readable storage medium, and the readable storage medium includes a flash memory, a mobile hard disk, a multimedia card, a card-type memory (e.g., SD or DX memory, etc.), a magnetic memory, a disk, an optical disk, etc. In some embodiments, the memory 11 may be an internal storage unit of the electronic device 1, such as a mobile hard disk of the electronic device 1. In other embodiments, the memory 11 may also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 1. Further, the memory 11 may also include both an internal storage unit of the electronic device 1 and an external storage device. The memory 11 may not only be used to store application software and various types of data installed in the electronic device 1, such as the code of the program 12, etc., but also be used to temporarily store data that has been output or is to be output.

[0256] The processor 10 may be composed of an integrated circuit in some embodiments, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips, etc. The processor 10 is the control core (Control Unit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes or executes programs or modules stored in the memory 11 (such as the program 12 for realizing non-periodic sampling remote control of unmanned ships), and calls data stored in the memory 11, so as to execute various functions of the electronic device 1 and process data.

[0257] The communication interface 13 may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices, and to achieve connection and communication between internal components of the electronic device.

[0258] The bus may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The bus is configured to realize connection and communication between the memory 11 and at least one processor 10, etc.

[0259] Figure 2 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 2 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0260] For example, although not shown, the electronic device 1 may also include a power source (such as a battery) for supplying power to various components. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power source may also include one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device 1 may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0261] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), or a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode) touch device. The display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device 1 and to display a visual user interface.

[0262] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0263] The program 12 stored in the memory 11 in the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can achieve:

[0264] Collect river remote sensing image data and enhance the river remote sensing image data;

[0265] A river channel positive and negative sample pair data set is constructed to train a river channel image representation extraction model, and the trained river channel image representation extraction model is used to extract image representation vectors of enhanced river channel remote sensing image data;

[0266] Constructing a river bend recognition model and identifying the river bend;

[0267] Matching the river bend position in the river remote sensing image data with the river geographic reference image to obtain the actual coordinate position of the river bend;

[0268] Control the unmanned ship to conduct key sampling when passing the turning points of the river bend, and conduct under-sampling when passing other locations.

[0269] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to Figure 1 to Figure 2 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0270] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article 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 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 "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0271] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0272] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made 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 non-periodic sampling remote control method for unmanned ships, characterized in that: The method comprises: S1: Collect river remote sensing image data and enhance the river remote sensing image data; S2: construct a river channel positive and negative sample pair data set to train a river channel image representation extraction model, and use the trained river channel image representation extraction model to extract image representation vectors of enhanced river channel remote sensing image data; S3: constructing a river bend recognition model and identifying the river bend, wherein the river bend recognition model takes the image representation vector as input and takes the river bend judgment result as output; The process of river bend identification is as follows: S31: Calculate the image representation vector similarity of adjacent sub-image data in the river remote sensing image data to form an image representation vector similarity set, the sub-image data The similarity set of image representation vectors with adjacent sub-image data is: in: Sub-image data The similarity with the image representation vector of the u-th adjacent sub-image data, wherein the image representation vector similarity is calculated by using a cosine similarity algorithm to perform similarity calculation on the image representation vector; S32: Select the result with the lowest similarity in the image representation vector similarity set as the neighboring image representation vector similarity of the sub-image data, where the sub-image data The similarity of the neighboring image representation vector is ; S33: if the similarity of the representation vectors of the adjacent images is less than a preset similarity threshold, marking the corresponding sub-image data as a candidate river bend area image; S34: For any pixel in the candidate river bend region image, take the pixel as the center In the pixel area, the pixel value of the central pixel is compared with the pixel values ​​of the adjacent 8 pixels in the order from left to right and from top to bottom. If the pixel value of the central pixel is greater than the pixel value of the adjacent pixel, 0 is output, otherwise 1 is output to obtain an 8-bit binary number, and the 8-bit binary number is converted to a decimal number. If the decimal number conversion result is greater than the preset pixel threshold, the corresponding pixel is marked as a core pixel; S35: Using the cubic spline interpolation method, the core pixel coordinates in the candidate river bend area image are interpolated to obtain an interpolation curve , where x represents the horizontal coordinate of the interpolation curve, is the vertical coordinate corresponding to the horizontal coordinate; S36: Convert the interpolation curve to a quantization function: Measure the curvature of the core pixel coordinates in the candidate river bend area image: in: num represents the number of core pixels in the candidate river bend region image; Represents the horizontal coordinate of the rth core pixel in the candidate river bend area image; The curvature of the curve representing the core pixel coordinates in the candidate river bend region image; If C is greater than a preset curve curvature threshold, the candidate river channel bend area image is marked as a river channel bend location; S4: matching the river bend position in the river remote sensing image data with the river geographic reference image to obtain the actual coordinate position of the river bend; S5: Control the unmanned vessel to perform key sampling when passing through the turning point of the river bend, and perform under-sampling when passing through other locations.

2. The non-periodic sampling remote control method for an unmanned ship according to claim 1, characterized in that: The step S1 includes collecting river channel remote sensing image data and enhancing the river channel remote sensing image data, including: The remote sensing image data I of the river channel is collected by using the remote sensing device carried by the remote sensing platform, wherein the remote sensing platform is an artificial satellite and the remote sensing device is a multispectral scanner. The collection process of the river channel remote sensing image data is as follows: Select filters corresponding to the red band, blue band and green band respectively for projection synthesis in the remote sensor, where the wavelength of the selected red band is 0.66um, the wavelength of the blue band is 0.45um, and the wavelength of the green band is 0.51um; The infrared radiation and visible light radiation emitted from the river area will enter the remote sensor and obtain the river remote sensing image data I through the filter after projection synthesis; The river remote sensing image data is enhanced, and the enhancement process is as follows: S11: grayscale processing is performed on the river remote sensing image data I, wherein the grayscale processing formula is: in: Represents pixels Color value in RGB color channels; pixel represents the pixel in the i-th row and j-th column in the river remote sensing image data I; Represents pixels Gray value of The gray value of the pixel is used as the pixel value of the corresponding pixel to obtain the gray processing result of the river remote sensing image data. ; S12: Remote sensing image data of river channels The grayscale value of the pixel in is linearly stretched, where the grayscale value The linear stretching processing formula is: in: is the preset threshold parameter; Represents river remote sensing image data The first quartile of the pixel gray value, Represents river remote sensing image data The third quartile of the gray value of the middle pixel; Represents grayscale value The linear stretching result of S13: Calculate the river channel remote sensing image data after linear stretching Information entropy : in: Represents river channel remote sensing image data after linear stretching The proportion of pixels with gray value L; S14: Remote sensing image data of river channels The gray value of the pixel is enhanced to obtain the enhanced river remote sensing image data , where the gray value The enhancement formula is: in: Gray value The enhanced results; The enhanced grayscale value is used as the pixel value of the corresponding pixel to form the enhanced river remote sensing image data .

3. The non-periodic sampling remote control method for an unmanned ship according to claim 1, characterized in that: In the step S2, the positive and negative sample pairs of river channels are constructed and the river channel image representation extraction model is trained, including: The river channel positive and negative sample pair data are constructed, and the steps for constructing the river channel positive and negative sample pair data are as follows: M pieces of local remote sensing image data of the river channel are collected to form positive samples, wherein the collected local remote sensing image data of the river channel are local remote sensing image data of the river channel after enhancement processing and only include remote sensing image data of the river channel portion; For any m-th river local remote sensing image data Perform two random cuts to obtain Two positive samples of , and perform random chromaticity transformation, rotation, color distortion and regional blurring on the two positive samples to obtain the local remote sensing image data of the river channel The corresponding M-2 different negative samples ; Construct the river positive and negative sample data ; Constructing a river channel image representation extraction model, the constructed river channel image representation extraction model includes an input layer, a feature extraction module, a feature mapping module and an output layer, wherein the input layer is used to receive the enhanced river channel remote sensing image data and perform image segmentation processing, and the plurality of sub-image data obtained by segmentation are sequentially input into the feature extraction module, the feature extraction module is used to perform encoding processing on the received image data, and the feature mapping module is used to perform feature mapping on the encoding processing result, and the mapped image representation vector is output; The river channel image representation extraction model is trained based on the river channel positive and negative sample data. The parameters to be trained of the river channel image representation extraction model are the weight parameters and bias parameters in the feature extraction module and the mapping matrix in the feature mapping module. The training process of the river channel image representation extraction model is as follows: S21: Construct the training objective function of the river image representation extraction model: in: T stands for transpose; represents the L1 norm, represents an exponential function with a natural constant as base; represents the training objective function of the river channel image representation extraction model, are the weight parameters and bias parameters in the feature extraction module, respectively. is the mapping matrix in the feature mapping module; To make positive samples The encoding result obtained by inputting into the feature extraction module, Represents a positive sample Image representation vector, j=1,2; Represents the mapping matrix The second derivative of Represents the mapping matrix The first derivative of ReLU activation function. express The similarity between It represents the sum of similarities between the negative samples of any m-th river local remote sensing image data and the negative samples of different river local remote sensing image data; S22: Randomly generate initialized model parameters , set the current iteration number of parameter training to t, and the maximum iteration number to Max, then the model parameters obtained in the tth iteration are ,make ; S23: If , then As the model parameters obtained by training, and based on The river channel image representation extraction model is constructed, otherwise, the process goes to step S24, where represents the preset gradient threshold, represents the L1 norm; S24: Generate gradient update step : in: , represents the L2 norm; S25: Generate an iteration step length that satisfies the following formula : S26: Update model parameters based on gradient step size and iteration step size: Let t=t+1 and return to step S23.

4. The non-periodic sampling remote control method for an unmanned ship as claimed in claim 3, characterized in that: In the step S2, the image representation vector of the enhanced river channel remote sensing image data is extracted using the trained river channel image representation extraction model, including: The trained river channel image representation extraction model is used to extract the image representation vector of the enhanced river channel remote sensing image data, where the image representation vector extraction process is as follows: The input layer receives enhanced river remote sensing image data , and divide the river remote sensing image data into K sub-image data of equal size, where the kth sub-image data obtained by segmentation is , ; The feature extraction module extracts the received sub-image data Perform encoding processing, where the encoding processing formula is: in: Represents sub-image data The encoding processing result; They are the weight parameters and bias parameters in the feature extraction module after model training; The feature mapping module processes the encoding results Perform feature mapping and map the image representation vector for output, where the mapping formula is: in: It is the mapping matrix in the feature mapping module after model training; Represents sub-image data The image representation vector.

5. The non-periodic sampling remote control method for an unmanned ship according to claim 1, characterized in that: In the step S4, the river channel bend position in the river channel remote sensing image data is matched in the river channel geographic reference image, including: The river bend position in the river remote sensing image data is matched in the river georeferenced image, where the river bend position matching process is as follows: Grayscale the georeferenced image of the river channel; Extract the SIFT feature vector of each pixel in the river georeferenced image after grayscale processing; Extract SIFT feature vectors of core pixels at the river bend position in river remote sensing image data; The feature similarity matching is performed on the SIFT feature vectors of the pixels in the river channel georeferenced image and the river channel bend position, so that the feature similarity after the overall matching is maximized, wherein the feature similarity calculation method is the cosine similarity algorithm; The successfully matched river bend position is marked at the corresponding position of the river geographic reference image to obtain the real geographic coordinates of the river bend position.

6. The non-periodic sampling remote control method for an unmanned ship according to claim 1, characterized in that: In step S5, the unmanned vessel is controlled to perform key sampling when passing through a bend in the river, including: The unmanned ship is controlled to conduct key sampling when passing through the turning point of the river bend, and the balanced key sampling process of the unmanned ship is as follows: S51: Calculate the ratio P of the area of ​​the river bend location to the river area; S52: Determine the sampling rate of the river bend position according to the ratio P : S53: Control the unmanned ship to travel in the river. When it does not reach the river bend area, it will follow the sampling rate. Take river samples and when you reach the river bend area, follow the sampling rate Carry out river sampling; according to the sampling rate Sampling the river channel means controlling the unmanned ship every time it travels distance, Sampling, where Indicates the unit travel distance of the unmanned ship.

Images