A battery monitoring system for pure electric unmanned ships

Abstract

The present invention relates to the field of battery monitoring technology, and discloses a battery monitoring system for a pure electric unmanned ship, including: an image acquisition module, which is used to acquire image data of six surfaces of a battery pack and the material type of the battery pack; a correction processing module, which is used to perform correction processing on the image data of the battery pack to obtain standard data; a corrosion feature map acquisition module, which is used to acquire battery pack image samples, establish a "material-corrosion feature map" database, and match them to obtain a set of corrosion feature maps of different corrosion degrees corresponding to the battery pack; a corrosion risk identification module, which is used to fit the standard data with the set of corrosion feature maps of the battery pack to obtain a fitting result, and output the corrosion risk score of the battery pack according to the fitting result; a response measure generation module, which is used to generate response measures according to the corrosion risk score. The present invention realizes efficient and accurate monitoring of the corrosion state of unmanned ship batteries.

Description

Technical Field

[0001] The present invention belongs to the technical field of battery monitoring, and in particular relates to a battery monitoring system for a pure electric unmanned ship. Background Art

[0002] With the rapid development of the new energy industry, unmanned boats, as important intelligent equipment, have a power system that is highly dependent on battery packs. The safety and reliability of battery packs are directly related to the operational stability of unmanned boats. Under long-term use or in complex environments, the battery pack shell of unmanned boats is easily corroded by factors such as seawater, high humidity, and salt spray, causing rust. Rust not only reduces the mechanical strength of the battery pack shell and affects its protective performance, but may also cause damage to the internal battery cells, leading to electrical failures or even safety accidents. Therefore, corrosion monitoring and risk assessment of unmanned boat battery packs have become important technical means to ensure the safe operation of unmanned boats.

[0003] At present, battery monitoring technology mainly focuses on the detection of internal electrical parameters, such as voltage, temperature and current, but the monitoring of the outer shell status is relatively weak. For the corrosion monitoring of unmanned ship battery packs, traditional methods include manual inspection, infrared imaging and acoustic wave detection. However, these methods have obvious defects in practical applications: manual inspection is inefficient and highly subjective, and it is difficult to adapt to the complex operating scenarios of unmanned ships; infrared imaging has high requirements on the thermal conductivity of materials and cannot effectively detect fine corrosion; acoustic wave detection is easily affected by environmental noise, and it is difficult to accurately locate the rusted area in the marine environment. In addition, the main shortcoming of the existing technology in the corrosion problem is the lack of accurate quantification and classification capabilities of the corrosion characteristics of battery packs of different materials. Summary of the invention

[0004] The present invention provides a battery monitoring system for a pure electric unmanned ship, which solves the technical problems in the related art of low monitoring efficiency, insufficient accuracy, and difficulty in meeting the requirements for quantifying the corrosion characteristics of battery packs of unmanned ships in complex environments.

[0005] The present invention provides a battery monitoring system for a pure electric unmanned ship, comprising:

[0006] An image acquisition module, which is used to obtain image data of six surfaces of the battery pack and the material type of the battery pack when replacing the battery of the unmanned boat;

[0007] A correction processing module, which is used to perform correction processing on the image data of the battery pack to obtain standard data;

[0008] A corrosion feature map acquisition module is used to obtain battery pack image samples of different material types and different corrosion degrees, establish a "material-corrosion feature map" database, match the material type of the battery pack with the "material-corrosion feature map" database, and obtain a set of corrosion feature maps of different corrosion degrees corresponding to the battery pack;

[0009] A corrosion risk identification module is used to fit the standard data with the set of corrosion characteristic maps of the battery pack to obtain a fitting result, and output the corrosion risk score of the battery pack according to the fitting result, and the range of the corrosion risk score is 0 to 100;

[0010] The response measure generation module is used to generate response measures according to the corrosion risk score.

[0011] Furthermore, the image data of the battery pack is corrected, including:

[0012] Step S201, using the gray level co-occurrence matrix method and the local binary pattern method to identify the large particle occluder area in the image data, and fill it with adjacent pixels;

[0013] Step S202, identifying a small particle occluder region in the image data based on the consistency of neighboring pixels, and filling it with neighboring pixels;

[0014] Step S203: Process the image data processed in the above steps using a histogram equalization method to obtain standard data.

[0015] Furthermore, the specific steps of step S201 include:

[0016] Step S2011, converting the image data into a grayscale image and performing Gaussian blur processing;

[0017] Step S2012, using the gray level co-occurrence matrix method to extract the energy feature, contrast feature, entropy feature and correlation feature of the gray image, and concatenating the energy feature, contrast feature, entropy feature and correlation feature to obtain a first texture feature vector; wherein the calculation formula of the energy feature is: , where EN represents energy, Represents the position in the normalized co-occurrence matrix generated using the gray-level co-occurrence matrix method The element value at , i and j represent the row index and column index of the normalized co-occurrence matrix element; the calculation formula of the contrast feature is: , where CR represents contrast; the calculation formula of entropy feature is: , where ENT represents entropy value and log represents logarithmic function; the calculation formula of correlation feature is:

[0018] , where CO represents the linear correlation of grayscale;

[0019] Step S2013, using a local binary pattern method to analyze the grayscale image to obtain a second texture feature vector;

[0020] Step S2014, concatenating the first texture feature vector and the second texture feature vector to obtain a combined texture feature vector, inputting the combined texture feature vector into an SVM classifier, and outputting the region of the large particle occluder in the grayscale image;

[0021] Step S2015, calculating the average grayscale value of pixels adjacent to the boundary of the large particle occluder region, and filling the large particle occluder region with the average grayscale value.

[0022] Furthermore, the specific steps of step S202 include:

[0023] Step S2021, for each pixel in the grayscale image except for the boundary pixels, the grayscale values ​​are compared with the eight neighboring pixels around it in turn, the grayscale difference is calculated, and the pixels whose grayscale difference is greater than a first preset threshold are marked as different pixels of the pixel;

[0024] Step S2022, when the number of different pixels of the pixel is less than a second preset threshold, the pixel and the different pixels are regarded as small particle occluders;

[0025] Step S2023: For pixels marked as small particle occluders, the original grayscale value is replaced by the average grayscale value of adjacent pixels.

[0026] Furthermore, battery pack corrosion image samples covering various material types are collected, including battery pack surface images with light corrosion, moderate corrosion and severe corrosion. The battery pack corrosion image samples are corrected to generate corrosion feature maps of different materials with different corrosion degrees, and a "material-corrosion feature map" database is constructed. Using the material type of the current battery pack as an index, a set of corrosion feature maps with different corrosion degrees corresponding to the material type is extracted from the "material-corrosion feature map" database, wherein the set of corrosion feature maps includes: a corrosion feature map with a light corrosion degree, a corrosion feature map with a moderate corrosion degree and a corrosion feature map with a severe corrosion degree.

[0027] Furthermore, a generative corrosion feature map is obtained by constructing a generative adversarial network model, and the generative corrosion feature map is added to the "material-corrosion feature map" database, wherein the generative adversarial network model includes a first generator and a first discriminator, the first generator is used to input a random noise vector and output a false corrosion feature map, the first discriminator is used to input the generated false corrosion feature map and the real corrosion feature map obtained by collecting battery pack corrosion image samples and correcting them, and outputs the probability that the generated false corrosion feature map is discriminated as a real corrosion feature map; the generative adversarial network model is trained by maximizing the value of the loss function of the first discriminator and minimizing the value of the loss function of the first generator, and finally a generative corrosion feature map is obtained; the loss function of the first generator is: ,in represents the loss value of the first generator, E represents the expected value, represents a random noise vector, obtained from the noise distribution, represents the output of the first generator, It represents the prediction probability of the first discriminator that the false corrosion feature map generated by the first generator is a real corrosion feature map; the loss function of the first discriminator is:

[0028] ,in represents the loss value of the first discriminator, represents the probability distribution of receiving the true corrosion feature map from the first discriminator, Represents the predicted probability of the first discriminator that the input image is a real corrosion feature map.

[0029] Furthermore, the standard data is fitted with the set of corrosion characteristic diagrams of the battery pack to obtain similarity values ​​between the standard data and the corrosion characteristic diagrams of different corrosion degrees. The specific steps of fitting include:

[0030] Step S301, calculating grayscale histograms for the standard data and the corrosion feature maps of different corrosion degrees in the corrosion feature map set respectively;

[0031] Step S302, normalizing the grayscale histogram to obtain the probability of each grayscale value;

[0032] Step S303, calculating the KL divergence values ​​of the standard data and the corrosion feature maps of different corrosion degrees in the corrosion feature map set according to the normalized grayscale histogram, when the KL divergence value is close to 0, it indicates that the standard data and the corrosion feature map are highly similar;

[0033] Step S304, using an exponential decay conversion formula to convert the KL divergence value into a similarity value, the exponential decay conversion formula is: , where S represents the similarity value between the standard data and the corrosion characteristic diagrams of different corrosion degrees, represents the KL divergence value of the standard data and the corrosion feature map, and e represents a natural constant.

[0034] Furthermore, the corrosion risk score of the battery pack is calculated based on the similarity values ​​of the standard data and the corrosion characteristic diagrams of different corrosion degrees. The calculation formula of the corrosion risk score is:

[0035] , where RS represents the corrosion risk score of the battery pack, , and The weights of the rust feature maps for mild, moderate and severe rust levels are represented respectively, and the sum of the three weights is 1. , and Represents the similarity values ​​between the standard data and the rust characteristic diagrams for mild, moderate and severe rust degrees respectively.

[0036] The beneficial effects of the present invention are as follows: the present invention generates a virtual corrosion feature map by generating an adversarial network model, significantly expands the corrosion feature sample database, enhances the adaptability of the system to complex corrosion scenarios, and provides strong data support for corrosion feature analysis;

[0037] The present invention constructs a "material-corrosion feature map" database and uses KL divergence as a similarity calculation method to achieve feature quantification and classification analysis of different materials and corrosion degrees, providing an accurate data basis for the scientific assessment of corrosion risks.

[0038] The present invention calculates the corrosion risk score of the battery pack based on the similarity values ​​of the standard data of the battery pack and the corrosion characteristic diagrams of different corrosion degrees, divides the corrosion risk score into multiple risk levels, and takes corresponding response measures based on the risk levels, thus realizing an intelligent integrated process from corrosion monitoring to decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a module schematic diagram of a battery monitoring system for a pure electric unmanned ship of the present invention;

[0040] Figure 2 is a flow chart of the present invention for performing correction processing on the image data of the battery pack;

[0041] Figure 3 It is a flow chart of the present invention for fitting standard data with a set of corrosion characteristic diagrams of a battery pack.

[0042] In the figure: an image acquisition module 101, a correction processing module 102, a corrosion feature map acquisition module 103, a corrosion risk identification module 104 and a response measure generation module 105. DETAILED DESCRIPTION

[0043] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0044] like Figure 1-Figure 3 As shown, a battery monitoring system for a pure electric unmanned ship includes:

[0045] An image acquisition module 101 is used to acquire image data of six surfaces of a battery pack and a material type of the battery pack when replacing the battery of an unmanned boat;

[0046] A correction processing module 102, which is used to perform correction processing on the image data of the battery pack to obtain standard data;

[0047] A corrosion feature map acquisition module 103 is used to acquire battery pack image samples of different material types and different corrosion degrees, establish a "material-corrosion feature map" database, match the material type of the battery pack with the "material-corrosion feature map" database, and obtain a set of corrosion feature maps of different corrosion degrees corresponding to the battery pack;

[0048] A corrosion risk identification module 104 is used to fit the standard data with the set of corrosion characteristic maps of the battery pack to obtain a fitting result, and output a corrosion risk score of the battery pack according to the fitting result, where the range of the corrosion risk score is 0 to 100;

[0049] The response measure generation module 105 is used to generate response measures according to the corrosion risk score.

[0050] In one embodiment of the present invention, when replacing the battery of an unmanned boat, an automated robotic arm is used in combination with a high-resolution camera to collect image data of the six surfaces of the battery pack to ensure standardization and consistency of image acquisition each time the battery is replaced; and the material type is obtained by reading the data label on the battery pack.

[0051] In one embodiment of the present invention, Figure 2 As shown, the image data of the battery pack is corrected and processed, including:

[0052] Step S201, using the gray level co-occurrence matrix method and the local binary pattern method to identify the large particle occluder area in the image data, and fill it with adjacent pixels;

[0053] Step S202, identifying a small particle occluder region in the image data based on the consistency of neighboring pixels, and filling it with neighboring pixels;

[0054] Step S203: The image data processed in the above steps is processed using a histogram equalization method to remove interference from light and obtain standard data.

[0055] In one embodiment of the present invention, the specific steps of step S201 include:

[0056] Step S2011, converting the image data into a grayscale image and performing Gaussian blur processing to reduce noise;

[0057] Step S2012, using the gray level co-occurrence matrix method to extract the energy feature, contrast feature, entropy feature and correlation feature of the gray level image, and concatenating the energy feature, contrast feature, entropy feature and correlation feature to obtain a first texture feature vector;

[0058] Specifically, the grayscale image is divided into 16×16 pixel blocks, and the grayscale values ​​are divided into 8 levels from 0 to 255, each level occupies 32 grayscale values. For each pixel block, an 8×8 initial co-occurrence matrix is ​​preset based on the grayscale level, and the initial element values ​​of the initial co-occurrence matrix are all 0; for each pair of adjacent pixels in the grayscale image, such as pixel A and pixel B, if the grayscale value of pixel A belongs to the i-th level and the grayscale value of pixel B belongs to the j-th level, then the corresponding grayscale value in the co-occurrence matrix is The value at the position is increased by 1, indicating that the pixel pair with gray levels i and j co-appears once; the entire gray image is traversed to obtain an updated update co-occurrence matrix; each element value in the update co-occurrence matrix is ​​divided by the total number of adjacent pixel pairs to obtain a normalized co-occurrence matrix, and each element of the normalized co-occurrence matrix represents the probability of a pair of gray levels appearing adjacently in the gray image; energy features, contrast features, entropy features and correlation features are extracted from the normalized co-occurrence matrix, and the energy features, contrast features, entropy features and correlation features are spliced ​​to obtain the first texture feature vector, where the calculation formula of the energy feature is: , where EN represents energy, Represents the position in the normalized co-occurrence matrix The element value of , i and j represent the row index and column index of the normalized co-occurrence matrix element. The energy feature is used to describe the uniformity of the grayscale distribution of the grayscale image. The larger the value, the more uniform the texture, that is, the greater the probability that the pixel block is a rusted area. The calculation formula of the contrast feature is: , where CR represents contrast, It represents the square difference between gray levels. The greater the contrast, the greater the gray difference and the rougher the texture of the pixel block. The calculation formula of entropy feature is: , where log represents the logarithmic function, ENT represents the entropy value, and the higher the entropy value, the more complex the texture of the pixel block; the calculation formula of the correlation feature is: , where CO represents the linear correlation of grayscale. The higher the correlation, the smoother the grayscale changes of adjacent pixels, that is, the finer the texture of the pixel block;

[0059] Step S2013, using a local binary pattern method to analyze the grayscale image to obtain a second texture feature vector;

[0060] Specifically, the local binary pattern (LBP) method of a 3×3 window is used. The LBP value of each pixel is determined by the eight surrounding neighbors, generating an 8-bit binary number corresponding to the range of 0 to 255. The grayscale image is divided into 16×16 pixel blocks. For each pixel block, a histogram array with a length of 256 is initialized, and the initial value of each position in the histogram array is 0. For each pixel in each pixel block except the boundary pixels, the LBP value is calculated. The calculation method is as follows: within the 3×3 window, taking the center pixel of the window as the reference, the grayscale values ​​of the eight neighbor pixels are compared with the grayscale value of the center pixel of the window in turn. When the neighbor image When the gray value of the pixel is greater than the gray value of the center pixel of the window, it is marked as 1, otherwise it is marked as 0, and the obtained 8 binary bits are combined into an 8-bit binary number, converted into a decimal number, and the LBP value of the center pixel of the window is obtained. The range of the LBP value is 0 to 255; add 1 to the position of the histogram array corresponding to the LBP value of the center pixel. For example, if the LBP value of a certain pixel is 78, then add 1 to the value of the 78th position of the histogram array; after calculating the LBP value of each pixel, divide the value of each position of the histogram array by the number of pixels in the pixel block except the boundary pixels to obtain the second texture feature vector;

[0061] Step S2014, concatenating the first texture feature vector and the second texture feature vector to obtain a combined texture feature vector, inputting the combined texture feature vector into an SVM classifier, and outputting the region of the grayscale image that is a large particle occluder;

[0062] Specifically, in step S2012, a first texture feature vector can be obtained for each pixel block, and in step S2013, a second texture feature vector can be obtained for each pixel block, the first texture feature vector of each pixel block is concatenated with the second texture feature vector to obtain a third texture feature vector, and then the third texture feature vectors of all pixel blocks are concatenated to obtain a combined texture feature vector;

[0063] Step S2015, calculating the average grayscale value of pixels adjacent to the boundary of the large particle occluder region, and filling the large particle occluder region with the average grayscale value.

[0064] In one embodiment of the present invention, the specific steps of step S202 include:

[0065] Step S2021, for each pixel in the grayscale image except the boundary pixels, the grayscale values ​​are compared with the eight neighboring pixels around it in turn, the grayscale difference is calculated, and the pixels whose grayscale difference is greater than a first preset threshold are marked as different pixels of the pixel. Preferably, the first preset threshold is set to 15;

[0066] Step S2022, when the number of different pixels of the pixel is less than a second preset threshold, the pixel and the different pixels are regarded as small particle occluders, preferably, the second preset threshold is set to 5;

[0067] Step S2023: For pixels marked as small particle occluders, the original grayscale value is replaced by the average grayscale value of adjacent pixels.

[0068] In one embodiment of the present invention, battery pack corrosion image samples covering multiple material types are collected, including battery pack surface images with light corrosion, moderate corrosion, and severe corrosion. The battery pack corrosion image samples are corrected to generate corrosion feature maps of different materials with different corrosion degrees, and a "material-corrosion feature map" database is constructed. The material type of the current battery pack is used as an index to extract a set of corrosion feature maps with different corrosion degrees corresponding to the material type from the "material-corrosion feature map" database. For example, if the material type of the current battery pack is aluminum alloy, the "aluminum alloy-corrosion feature map" set is returned, wherein the corrosion feature map set includes: a corrosion feature map with a light corrosion degree, a corrosion feature map with a moderate corrosion degree, and a corrosion feature map with a severe corrosion degree.

[0069] In one embodiment of the present invention, a generative corrosion feature map is obtained by constructing a generative adversarial network model, and the generative corrosion feature map is added to the "material-corrosion feature map" database, wherein the generative adversarial network model includes a first generator and a first discriminator, the first generator is used to input a random noise vector, and output a false corrosion feature map, the first discriminator is used to input the generated false corrosion feature map and the real corrosion feature map obtained by collecting battery pack corrosion image samples and correcting them, and outputs the probability that the generated false corrosion feature map is discriminated as a real corrosion feature map; the generative adversarial network model is trained by maximizing the value of the loss function of the first discriminator and minimizing the value of the loss function of the first generator, and finally a generative corrosion feature map is obtained; the loss function of the first generator is: ,in represents the loss value of the first generator, E represents the expected value, represents a random noise vector, obtained from the noise distribution, represents the output of the first generator, It represents the prediction probability of the first discriminator that the false corrosion feature map generated by the first generator is a real corrosion feature map; the loss function of the first discriminator is:

[0070] ,in represents the loss value of the first discriminator, represents the probability distribution of receiving the true corrosion feature map from the first discriminator, Represents the predicted probability of the first discriminator that the input image is a real corrosion feature map.

[0071] In one embodiment of the present invention, the standard data is fitted with the set of corrosion characteristic diagrams of the battery pack to obtain similarity values ​​between the standard data and the corrosion characteristic diagrams of different corrosion degrees, such as Figure 3 As shown, the specific steps of fitting include:

[0072] Step S301, calculating grayscale histograms for the standard data and the corrosion feature maps of different corrosion degrees in the corrosion feature map set, and counting the number of occurrences of each grayscale value;

[0073] Step S302, normalize the grayscale histogram to obtain the probability of each grayscale value, wherein the normalized calculation formula is: , represents the probability of gray value i, Represents the number of pixels with gray value i, Indicates the total number of pixels;

[0074] Step S303, calculating the KL divergence values ​​of the standard data and the corrosion feature maps of different corrosion degrees in the corrosion feature map set according to the normalized grayscale histogram. When the KL divergence value is close to 0, it means that the standard data and the corrosion feature map are highly similar. The calculation formula of the KL divergence is: , represents the KL divergence value of the standard data and the corrosion feature map, represents the probability of gray value i in the standard data, Represents the probability of gray value i in the rust feature map;

[0075] KL divergence can effectively capture the subtle differences between the grayscale distribution of images. It provides a more scientific and accurate way to quantify similarity by measuring the distribution difference and information loss between the standard data and the rust feature map.

[0076] Step S304, using an exponential decay conversion formula to convert the KL divergence value into a similarity value, the exponential decay conversion formula is: , where S represents the similarity value between the standard data and the corrosion characteristic diagrams of different corrosion degrees, and e represents the natural constant.

[0077] In one embodiment of the present invention, the corrosion risk score of the battery pack is calculated based on the similarity values ​​of the standard data and the corrosion characteristic graphs of different corrosion degrees. The calculation formula of the corrosion risk score is:

[0078] , where RS represents the corrosion risk score of the battery pack, , and The weights of the rust feature maps for mild, moderate and severe rust levels are represented respectively, and the sum of the three weights is 1. , and Represents the similarity values ​​between the standard data and the rust characteristic diagrams for mild, moderate and severe rust degrees respectively.

[0079] In an embodiment of the present invention, the corrosion risk score of the battery pack is divided into four corrosion risk levels, and corresponding response measures are taken according to different corrosion risk levels. A corrosion risk score between 0 and 25 is a low risk level, indicating that the battery pack is slightly corroded or shows no signs of corrosion. No special measures are required at this time. A corrosion risk score between 25 and 50 is a medium-low risk level, indicating that the battery pack begins to show early signs of corrosion and the degree of corrosion is relatively light. The corresponding response measures are to perform local cleaning of the rusted area to remove the surface oxide layer and rust spots. A corrosion risk score between 50 and 75 is a medium-high risk level, indicating that the battery pack has obvious corrosion problems and may affect the structure or performance of the battery pack. In this case, the rusted area needs to be thoroughly cleaned and checked to see if the corrosion has damaged the structural integrity and performance of the battery pack. At the same time, the battery pack should be checked to see if there is any performance degradation or safety hazard caused by corrosion. Based on the inspection results, it is determined whether the battery pack unit needs to be replaced. A corrosion risk score between 75 and 100 is a high risk level, indicating that the battery pack is severely corroded and there are safety hazards. In this case, the battery pack needs to be stopped immediately and replaced to avoid electrical failures or safety accidents caused by corrosion.

[0080] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A battery monitoring system for a pure electric unmanned ship, characterized in that: include: An image acquisition module, which is used to obtain image data of six surfaces of the battery pack and the material type of the battery pack when replacing the battery of the unmanned boat; A correction processing module, which is used to perform correction processing on the image data of the battery pack to obtain standard data; The correction process includes: step S201, using the gray level co-occurrence matrix method and the local binary pattern method to identify the large particle occluder area in the image data, and filling it with adjacent pixels, as follows: Step S2011, converting the image data into a grayscale image and performing Gaussian blur processing; Step S2012, using the gray level co-occurrence matrix method to extract the energy feature, contrast feature, entropy feature and correlation feature of the gray image, and concatenating the energy feature, contrast feature, entropy feature and correlation feature to obtain a first texture feature vector; wherein the calculation formula of the energy feature is: , where EN represents energy, Represents the position in the normalized co-occurrence matrix generated using the gray-level co-occurrence matrix method The element value at , i and j represent the row index and column index of the normalized co-occurrence matrix element; the calculation formula of the contrast feature is: , where CR represents contrast; the calculation formula of entropy feature is: , where ENT represents entropy value and log represents logarithmic function; the calculation formula of correlation feature is: , where CO represents the linear correlation of grayscale; Step S2013, using a local binary pattern method to analyze the grayscale image to obtain a second texture feature vector; Step S2014, concatenating the first texture feature vector and the second texture feature vector to obtain a combined texture feature vector, inputting the combined texture feature vector into an SVM classifier, and outputting the region of the grayscale image that is a large particle occluder; Step S2015, calculating the average grayscale value of pixels adjacent to the boundary of the large particle occluder region, and filling the average grayscale value into the large particle occluder region; A corrosion feature map acquisition module is used to acquire battery pack image samples of different material types and different corrosion degrees, establish a "material-corrosion feature map" database, match the material type of the battery pack with the "material-corrosion feature map" database, and obtain a set of corrosion feature maps of different corrosion degrees corresponding to the battery pack; A corrosion risk identification module is used to fit the standard data with the set of corrosion characteristic maps of the battery pack to obtain a fitting result, and output the corrosion risk score of the battery pack according to the fitting result, and the range of the corrosion risk score is 0 to 100; The fitting result represents the similarity value between the standard data and the corrosion characteristic diagrams of different corrosion degrees. The specific steps of fitting include: Step S301, calculating grayscale histograms for the standard data and the corrosion feature maps of different corrosion degrees in the corrosion feature map set respectively; Step S302, normalizing the grayscale histogram to obtain the probability of each grayscale value; Step S303, calculating the KL divergence values ​​of the standard data and the corrosion feature maps of different corrosion degrees in the corrosion feature map set according to the normalized grayscale histogram, when the KL divergence value is close to 0, it indicates that the standard data and the corrosion feature map are highly similar; Step S304, using an exponential decay conversion formula to convert the KL divergence value into a similarity value, the exponential decay conversion formula is: , where S represents the similarity value between the standard data and the corrosion characteristic diagrams of different corrosion degrees, represents the KL divergence value of the standard data and the corrosion characteristic map, and e represents the natural constant; The calculation formula for the corrosion risk score is: , where RS represents the corrosion risk score of the battery pack, , and The weights of the rust feature maps for mild, moderate and severe rust levels are represented respectively, and the sum of the three weights is 1. , and Represents the similarity values ​​between the standard data and the rust characteristic diagrams of light, moderate and heavy rust degrees respectively; The response measure generation module is used to generate response measures according to the corrosion risk score.

2. A battery monitoring system for a pure electric unmanned ship according to claim 1, characterized in that: Correction processing is performed on the image data of the battery pack, and also includes: Step S202, identifying a small particle occluder region in the image data based on the consistency of neighboring pixels, and filling it with neighboring pixels; Step S203: Process the image data processed in the above steps using a histogram equalization method to obtain standard data.

3. A battery monitoring system for a pure electric unmanned ship according to claim 2, characterized in that: The specific steps of step S202 include: Step S2021, for each pixel in the grayscale image except for the boundary pixels, the grayscale values ​​are compared with the eight neighboring pixels around it in turn, the grayscale difference is calculated, and the pixels whose grayscale difference is greater than a first preset threshold are marked as different pixels of the pixel; Step S2022, when the number of different pixels of the pixel is less than a second preset threshold, the pixel and the different pixels are regarded as small particle occluders; Step S2023: For pixels marked as small particle occluders, the original grayscale value is replaced by the average grayscale value of adjacent pixels.

4. A battery monitoring system for a pure electric unmanned ship according to claim 1, characterized in that: Battery pack corrosion image samples covering various material types are collected, including battery pack surface images with light, moderate and severe corrosion. The battery pack corrosion image samples are corrected to generate corrosion feature maps of different materials with different corrosion degrees, and a "material-corrosion feature map" database is constructed. The material type of the current battery pack is used as an index to extract a set of corrosion feature maps with different corrosion degrees corresponding to the material type from the "material-corrosion feature map" database, where the corrosion feature map set includes: corrosion feature maps with light corrosion degrees, corrosion feature maps with moderate corrosion degrees and corrosion feature maps with severe corrosion degrees.

5. A battery monitoring system for a pure electric unmanned ship according to claim 4, characterized in that: A generative corrosion feature map is obtained by constructing a generative adversarial network model, and the generative corrosion feature map is added to the "material-corrosion feature map" database, wherein the generative adversarial network model includes a first generator and a first discriminator, the first generator is used to input a random noise vector and output a false corrosion feature map, the first discriminator is used to input the generated false corrosion feature map and the real corrosion feature map obtained by collecting battery pack corrosion image samples and correcting them, and outputs the probability that the generated false corrosion feature map is discriminated as a real corrosion feature map; the generative adversarial network model is trained by maximizing the value of the loss function of the first discriminator and minimizing the value of the loss function of the first generator, and finally a generative corrosion feature map is obtained; the loss function of the first generator is: ,in represents the loss value of the first generator, E represents the expected value, represents a random noise vector, obtained from the noise distribution, represents the output of the first generator, It represents the prediction probability of the first discriminator that the false corrosion feature map generated by the first generator is a real corrosion feature map; the loss function of the first discriminator is: ,in represents the loss value of the first discriminator, represents the probability distribution of receiving the true corrosion feature map from the first discriminator, Represents the predicted probability of the first discriminator that the input image is a real corrosion feature map.

Images