A new energy vehicle automatic defrosting system and method

Abstract

The application relates to the field of automobile defrosting technology and discloses a new energy automobile automatic defrosting system and method, which acquires a first image and a second image of a windshield glass in response to adjacent engine-off instructions and start-up instructions of a target vehicle; performs edge detection on the first image and the second image respectively in response to a defrosting instruction of the target vehicle, and determines a frost profile in the second image; projects a uniform light curtain towards the windshield glass to acquire a third image; and segments the first image and the third image based on the frost profile to obtain a feature matrix; divides a feature area corresponding to the windshield glass of the feature matrix into M subareas of equal size, and determines the frost thickness of each subarea based on the feature matrix; and dynamically controls the defrosting angle and the defrosting power of N defrosting units based on the frost thickness of each subarea to reduce the defrosting time.

Description

Technical Field

[0001] This invention relates to the field of automotive defrosting technology, and more specifically, to an automatic defrosting system and method for new energy vehicles. Background Technology

[0002] In winter, low temperatures can cause frost to form on car windows after prolonged parking, requiring defrosting before driving. However, existing defrosting systems typically defrost the entire windshield evenly, resulting in slow defrosting times and longer waiting times for users.

[0003] Referring to the automatic defrosting method, apparatus, and system for pure electric vehicles disclosed in CN116442950A, the method includes receiving a defrosting signal and determining whether the conditions for plug-in defrosting are met; when the conditions for plug-in defrosting are met, determining whether the pre-operation of plug-in defrosting is completed; when the pre-operation of plug-in defrosting is completed, limiting the power of the air conditioner and water heater based on the current maximum allowable output power of the on-board charger, and performing automatic defrosting; and when the automatic defrosting exit condition is met, exiting automatic defrosting. This method achieves automatic defrosting, maximizes driving range, and improves user experience.

[0004] A comprehensive analysis of the existing technologies reveals that this technology controls the air conditioner and water heater to perform defrosting through the logical relationship between defrosting signals and defrosting conditions. However, this method does not consider the case of uneven frost thickness, such as differentiating between thicker and thinner frost layers for defrosting, thus causing a prolonged defrosting time. Summary of the Invention

[0005] This invention provides an automatic defrosting system and method for new energy vehicles, solving the technical problems mentioned in the background art.

[0006] In a first aspect, the present invention provides an automatic defrosting method for new energy vehicles, comprising:

[0007] Step 1: In response to the engine shutdown command and start command adjacent to the target vehicle, acquire the first image and the second image of the windshield respectively;

[0008] Step 2: In response to the defrosting command of the target vehicle, perform edge detection on the first image and the second image respectively to determine the frost outline in the second image;

[0009] Step 3: Project a uniform light curtain onto the windshield to obtain a third image; and segment the first and third images based on the frost outline to obtain a feature matrix;

[0010] Step 4: Divide the feature matrix into M equal-sized sub-regions corresponding to the feature area of ​​the windshield, and determine the frost thickness of each sub-region based on the feature matrix.

[0011] Step 5: Based on the frost thickness of each sub-region, dynamically control the defrosting angle and defrosting power of N defrosting units to reduce defrosting time.

[0012] Furthermore, step 2, determining the frost outline in the second image, includes:

[0013] Edge detection is performed on the first image and the second image to obtain a first edge matrix and a second edge matrix; wherein the first edge matrix and the second edge matrix are both Boolean matrices, with elements 1 and 0 representing edge features and non-edge features respectively;

[0014] The frosting outline is calculated as follows:

[0015] F sp =ReLU(F2-F1); where F sp F1 represents the frosting outline, ReLU represents the ReLU activation function, F1 represents the first edge matrix, and F2 represents the second edge matrix.

[0016] Furthermore, in step 3, the projection of a uniform light curtain onto the windshield is achieved based on the AR-HUD projection unit of the target vehicle.

[0017] Furthermore, the feature matrix in step 3 includes:

[0018] The frost outline is fitted to the third image, and the pixels in the third image located inside the frost outline are retained to obtain the first initial feature image;

[0019] The frost outline is fitted to the first image, and the pixels in the first image located inside the frost outline are retained to obtain the second initial feature image;

[0020] The first initial feature image and the second initial feature image are converted to grayscale respectively, and the difference is calculated to obtain the feature matrix.

[0021] Furthermore, step 4, which determines the frost thickness of each sub-region based on the feature matrix, includes:

[0022] Construct a frost thickness prediction model, including sample data and sample labels;

[0023] Within a historical time period, steps 1-3 are repeated K times to obtain K feature matrices.

[0024] Obtain the normalized element value of the element in the x-th row and y-th column of each feature matrix as sample data;

[0025] The element in the x-th row and y-th column of each feature matrix corresponds to the normalized frost thickness at the windshield position and is used as the sample label.

[0026] A frost thickness prediction model was trained based on sample data and sample labels.

[0027] The frost thickness prediction model is built on support vector machine and backpropagated through mean squared error loss function to update the hyperparameters of the frost thickness prediction model.

[0028] Within a preset time period, based on the frost thickness prediction model and feature matrix, the frost thickness at each corresponding position on the windshield is obtained.

[0029] Obtain the frost thickness at each location within the m-th sub-region, and calculate the average value to obtain the frost thickness of the m-th sub-region; where 1≤m≤M, and m is a positive integer.

[0030] Furthermore, step 5, which dynamically controls the defrosting angle and defrosting power of the N defrosting units, includes:

[0031] Step 61: Initialize and generate defrost individuals that meet the constraints; wherein, the defrost individual includes: the defrost angle G and defrost power P of N defrost units;

[0032] Step 62, the constraints include: G min ≤G≤G max P min ≤P≤P max Among them, G min and G max P represents the minimum defrost angle and the maximum defrost angle, respectively. min and P max These represent the minimum defrosting power and the maximum defrosting power, respectively.

[0033] Step 63, the steps to obtain the fitness value of the defrost individual are as follows:

[0034] Step 631: Defrost the windshield based on the defrosting individual and acquire a third image at fixed time intervals to obtain the frost thickness of M sub-regions at each moment;

[0035] Step 632: Sort the sub-regions based on the frost thickness at the initial moment to obtain the first sort;

[0036] Step 633: Calculate the difference in frost thickness between adjacent moments for each sub-region, and sort the M sub-regions based on the difference to obtain the second sort;

[0037] Step 634: Accumulate the difference between the positions of the M sub-regions in the first and second sorts respectively to obtain the fitness value of the defrosting individual;

[0038] Step 64: If the fitness value is less than or equal to the preset fitness threshold, then keep the defrosting individual defrosting the windshield; otherwise, update the defrosting individual based on gradient descent and repeat steps 61-63 until the fitness value is less than or equal to the preset fitness threshold.

[0039] Furthermore, step 5, which involves dynamically controlling the defrosting angle and defrosting power of the N defrosting units, also includes:

[0040] In response to the feature matrices corresponding to adjacent shutdown and start commands of multiple target vehicles, a standard database is constructed, and the defrosting individual corresponding to each feature matrix in the standard database is determined.

[0041] Within the target time period, obtain the target feature matrix;

[0042] Calculate the similarity between the target feature matrix and any feature matrix in the standard database, and match the target feature matrix with the feature matrix with the highest similarity.

[0043] The individual defrosting samples corresponding to the matched feature matrix are applied to the target time period for defrosting.

[0044] Furthermore, the similarity of the feature matrices includes:

[0045] Calculate the Euclidean distance between the target feature matrix and any feature matrix in the standard database;

[0046] The reciprocal of the Euclidean distance is used as the similarity.

[0047] Secondly, an automatic defrosting system for new energy vehicles, applied in any of the automatic defrosting methods for new energy vehicles, includes:

[0048] The first acquisition module is used to acquire the first image and the second image of the windshield in response to the engine shutdown command and start command adjacent to the target vehicle.

[0049] The second acquisition module is used to respond to the defrosting command of the target vehicle, perform edge detection on the first image and the second image respectively, and determine the frost outline in the second image;

[0050] The third acquisition module is used to project a uniform light curtain onto the windshield to acquire a third image; and to segment the first and third images based on the frost outline to obtain a feature matrix;

[0051] The data analysis module is used to divide the feature matrix corresponding to the feature area of ​​the windshield into M equal-sized sub-regions, and determine the frost thickness of each sub-region based on the feature matrix.

[0052] The glass defrosting module is used to dynamically control the defrosting angle and defrosting power of N defrosting units based on the frost thickness of each sub-region, so as to reduce defrosting time.

[0053] The beneficial effects of this invention are as follows: by integrating edge detection, light curtain projection, and feature matrix construction technologies of images during engine shutdown and startup, and based on the light refractive index of frost layers of different thicknesses, it achieves accurate identification of the thickness of frost layers on the windshield, and uses an intelligent model to adopt differentiated defrosting strategies for each sub-region, thereby achieving fast and accurate defrosting, significantly shortening defrosting time and reducing the user's defrosting waiting time. Attached Figure Description

[0054] Figure 1 This is a flowchart of an automatic defrosting method for new energy vehicles according to the present invention;

[0055] Figure 2 This is a block diagram of an automatic defrosting system for new energy vehicles according to the present invention. Detailed Implementation

[0056] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0057] like Figures 1-2 As shown, an automatic defrosting method for new energy vehicles includes:

[0058] Step 1: In response to the engine shutdown command and start command adjacent to the target vehicle, acquire the first image and the second image of the windshield respectively;

[0059] Step 2: In response to the defrosting command of the target vehicle, perform edge detection on the first image and the second image respectively to determine the frost outline in the second image;

[0060] Step 3: Project a uniform light curtain onto the windshield to obtain a third image; and segment the first and third images based on the frost outline to obtain a feature matrix;

[0061] Step 4: Divide the feature matrix into M equal-sized sub-regions corresponding to the feature area of ​​the windshield, and determine the frost thickness of each sub-region based on the feature matrix.

[0062] Step 5: Based on the frost thickness of each sub-region, dynamically control the defrosting angle and defrosting power of N defrosting units to reduce defrosting time.

[0063] It should be noted that the "Stop" command is a control signal issued when the vehicle's power system is turned off, used to confirm that the vehicle has stopped running. At this time, there is no frost on the windshield (corresponding to the first image). The "Start" command is a control signal issued when the vehicle's power system is turned on, used to confirm that the user is starting the vehicle. At this time, the vehicle may have frost on the windshield after a long period of inactivity (corresponding to the second image). The "Defrost" command indicates that after the user starts the vehicle, there is frost on the windshield as determined by visual inspection.

[0064] In one embodiment of the present invention, the first image, the second image, and the third image are all acquired using the vehicle's dashcam. The dashcam acquires images from a fixed angle and position; therefore, the first image, the second image, and the third image are the same size and comparable.

[0065] In one embodiment of the present invention, determining the frosting outline in the second image includes:

[0066] Edge detection is performed on the first image and the second image to obtain a first edge matrix and a second edge matrix; wherein the first edge matrix and the second edge matrix are both Boolean matrices, with elements 1 and 0 representing edge features and non-edge features respectively;

[0067] The frosting outline is calculated as follows:

[0068] F sp =ReLU(F2-F1); where F sp F1 represents the frosting outline, ReLU represents the ReLU activation function, F1 represents the first edge matrix, and F2 represents the second edge matrix.

[0069] In detail, if frost is present on the windshield, it indicates a large area of ​​obstruction on the windshield, allowing for the extraction of the corresponding crystal edge features. Therefore, unique edge features in the second image are selected and identified as frost contours.

[0070] In one embodiment of the present invention, edge detection includes, but is not limited to, the Canny edge detection algorithm. The Canny edge detection algorithm first performs Gaussian filtering on the image to remove noise, then calculates the gradient magnitude and direction, and obtains sharp edges through non-maximum suppression and double threshold detection.

[0071] In one embodiment of the invention, the ReLU activation function sets negative values ​​directly to zero, preserving positive difference edges (frost contours) that only appear in the second image.

[0072] Optionally, the obtained frosting contour matrix F sp Perform morphological operations (such as erosion, dilation, opening, or closing operations) to remove noise or broken edges, resulting in a more coherent and accurate frosting profile.

[0073] In one embodiment of the present invention, the projection of a uniform light curtain onto the windshield is achieved based on the AR-HUD projection unit of the target vehicle.

[0074] In one embodiment of the invention, the AR-HUD projection unit built into the target vehicle is used to project a uniform light screen onto the windshield. Specifically, new energy vehicles (such as the Lynk & Co 08) are typically equipped with an AR-HUD projection unit. This system is originally used to project navigation, vehicle information, or entertainment content onto the windshield in an augmented reality manner, thereby improving the driving experience and safety. When the vehicle is in cinema mode, the AR-HUD can project a preset uniform light screen onto the windshield.

[0075] In detail, by utilizing existing AR-HUD hardware, not only are the costs of system integration reduced, but the versatility of the vehicle's existing equipment is also fully utilized.

[0076] In one embodiment of the present invention, the feature matrix includes:

[0077] The frost outline is fitted to the third image, and the pixels in the third image located inside the frost outline are retained to obtain the first initial feature image;

[0078] The frost outline is fitted to the first image, and the pixels in the first image located inside the frost outline are retained to obtain the second initial feature image;

[0079] The first initial feature image and the second initial feature image are converted to grayscale respectively, and the difference is calculated to obtain the feature matrix.

[0080] In detail, the frost area of ​​the windshield is extracted based on the previously extracted frost contour, thereby obtaining the corresponding part of the frost area in the first and third images.

[0081] In one embodiment of the present invention, step 4, which involves determining the frost thickness of each sub-region based on the feature matrix, includes:

[0082] Construct a frost thickness prediction model, including sample data and sample labels;

[0083] Within a historical time period, steps 1-3 are repeated K times to obtain K feature matrices.

[0084] Obtain the normalized element value of the element in the x-th row and y-th column of each feature matrix as sample data;

[0085] The element in the x-th row and y-th column of each feature matrix corresponds to the normalized frost thickness at the windshield position and is used as the sample label.

[0086] A frost thickness prediction model was trained based on sample data and sample labels.

[0087] The frost thickness prediction model is built on support vector machine and backpropagated through mean squared error loss function to update the hyperparameters of the frost thickness prediction model.

[0088] Within a preset time period, based on the frost thickness prediction model and feature matrix, the frost thickness at each corresponding position on the windshield is obtained.

[0089] Obtain the frost thickness at each location within the m-th sub-region, and calculate the average value to obtain the frost thickness of the m-th sub-region; where 1≤m≤M, and m is a positive integer.

[0090] Because the thickness of frost is uneven across different areas after it forms, the shading rate varies across the frost-covered area of ​​the windshield. When a uniformly projected light curtain reaches the windshield, the different shading rates result in different grayscale values ​​after grayscale conversion. Consequently, a non-linear mapping relationship exists between grayscale values ​​and frost thickness.

[0091] In detail, image acquisition was performed multiple times within a historical time period (steps 1-3) to obtain feature matrices at K different time points. These matrices reflect the grayscale differences caused by varying frost occlusion rates under uniform light curtain illumination. For each feature matrix, the normalized grayscale value in the x-th row and y-th column was extracted as sample data. Simultaneously, the value was normalized using the actual frost thickness at a known location (obtainable through calibration or other measurement methods) and used as the sample label. This provides the model with a correspondence between input (grayscale information) and output (frost thickness).

[0092] In detail, a frost thickness prediction model is constructed using Support Vector Machines (SVM) based on sample data and sample labels. SVM has advantages in handling nonlinear mapping relationships; through kernel functions, it can map the original input space to a high-dimensional feature space, thereby better fitting the nonlinear relationship between gray values ​​and frost thickness. Mean squared error is used as the loss function to measure the error between the model's predicted values ​​and the true labels. The hyperparameters of the model are continuously adjusted through backpropagation, gradually reducing the prediction error and ultimately obtaining a high-precision thickness prediction model.

[0093] In detail, within a preset time period, the system uses a trained prediction model and real-time collected feature matrices to predict the frost thickness at each corresponding location on the windshield. The feature region corresponding to the windshield is divided into M equal-sized sub-regions. For each sub-region, the average of the predicted frost thickness at all locations within it is taken as the overall frost thickness index for that sub-region.

[0094] In one embodiment of the present invention, dynamically controlling the defrosting angle and defrosting power of N defrosting units includes:

[0095] Step 61: Initialize and generate defrost individuals that meet the constraints; wherein, the defrost individual includes: the defrost angle G and defrost power P of N defrost units;

[0096] Step 62, the constraints include: G min ≤G≤G max P min ≤P≤P max Among them, G min and G max P represents the minimum defrost angle and the maximum defrost angle, respectively. min and P max These represent the minimum defrosting power and the maximum defrosting power, respectively.

[0097] Step 63, the steps to obtain the fitness value of the defrost individual are as follows:

[0098] Step 631: Defrost the windshield based on the defrosting individual and acquire a third image at fixed time intervals to obtain the frost thickness of M sub-regions at each moment;

[0099] Step 632: Sort the sub-regions based on the frost thickness at the initial moment to obtain the first sort;

[0100] Step 633: Calculate the difference in frost thickness between adjacent moments for each sub-region, and sort the M sub-regions based on the difference to obtain the second sort;

[0101] Step 634: Accumulate the difference between the positions of the M sub-regions in the first and second sorts respectively to obtain the fitness value of the defrosting individual;

[0102] Step 64: If the fitness value is less than or equal to the preset fitness threshold, then keep the defrosting individual defrosting the windshield; otherwise, update the defrosting individual based on gradient descent and repeat steps 61-63 until the fitness value is less than or equal to the preset fitness threshold.

[0103] Optionally, the power of the N defrosting units in a conventional defrosting mode can be obtained, for example, if each power is Nd, then the total power is N×Nd. Furthermore, the total power of the N defrosting units in the defrosting method disclosed in this application is also N×Nd, as an optional constraint for step 62 in this application.

[0104] In one embodiment of the present invention, the frost thickness at each location on the windshield is obtained based on a frost prediction model. The difference in frost thickness at adjacent moments is used to represent the frost melting rate. For frost areas with greater thickness, the frost melting rate should be greater than that of frost areas with less thickness. Therefore, the first sort and the second sort represent the initial sorting of frost thickness and the sorting of melting rates for each sub-region, respectively.

[0105] In one embodiment of the present invention, the defrosting unit includes, but is not limited to, a warm air defrosting device or a resistance wire heating defrosting device. For example, a warm air defrosting device defrosts the windshield by adjusting the angle of the air outlet and the warm air power. A resistance wire heating defrosting device defrosts the windshield by heating it based on the direction and power of electromagnetic wave emission.

[0106] In detail, the system first uses a frost prediction model to obtain the frost thickness at each location on the windshield, and then compares the thickness changes at adjacent time points to obtain the frost melting rate. The system acquires the frost thickness at each point on the windshield in real time through continuous image acquisition and the prediction model. The difference in frost thickness at each location between consecutive time points is calculated as the frost melting rate at that location. The initial frost thickness is sorted (first sort), and the frost melting rate of each sub-region is sorted (second sort). Since thicker frost areas should theoretically have a higher melting rate, if the two sorts are inconsistent, it indicates that the defrosting effect has not met expectations, and the defrosting control strategy can be evaluated and optimized accordingly. Based on the above two sorts, the system calculates the fitness value of each sub-region, which is the sum of the differences between the initial thickness sort and the melting rate sort, as an evaluation index of the current defrosting scheme. When the fitness value exceeds a preset threshold, the system adjusts the parameters of the defrosting unit using algorithms such as gradient descent until the fitness value meets the requirements. This dynamic adjustment ensures that areas with thicker frost receive greater defrosting power and a better defrosting angle, thereby achieving differentiated defrosting and improving overall defrosting efficiency. Based on the actual measured frost thickness and melting rate, the system automatically selects and adjusts the operating mode and output parameters of the corresponding defrosting unit to ensure rapid frost removal.

[0107] In one embodiment of the present invention, dynamically controlling the defrosting angle and defrosting power of N defrosting units further includes:

[0108] In response to the feature matrices corresponding to adjacent shutdown and start commands of multiple target vehicles, a standard database is constructed, and the defrosting individual corresponding to each feature matrix in the standard database is determined.

[0109] Within the target time period, obtain the target feature matrix;

[0110] Calculate the similarity between the target feature matrix and any feature matrix in the standard database, and match the target feature matrix with the feature matrix with the highest similarity.

[0111] The individual defrosting samples corresponding to the matched feature matrix are applied to the target time period for defrosting.

[0112] In one embodiment of the present invention, a database is generated by saving historical feature matrices and corresponding defrosting individuals. Since the shape and thickness of frost on a vehicle exhibit self-similarity, within a target time period, the target feature matrix is ​​matched with the feature matrix in the database to quickly obtain the corresponding defrosting individuals. Based on these defrosting individuals, the defrosting unit is controlled to perform defrosting operations on the vehicle's windshield.

[0113] In one embodiment of the present invention, the similarity of the feature matrices includes:

[0114] Calculate the Euclidean distance between the target feature matrix and any feature matrix in the standard database;

[0115] The reciprocal of the Euclidean distance is used as the similarity.

[0116] The system not only adjusts the parameters of the defrosting units based on real-time acquired feature matrices, but also further improves the response speed and accuracy of defrosting control by constructing a historical standard database. Responding to adjacent shutdown and start commands from multiple target vehicles, the system acquires the corresponding feature matrices and combines them with the defrosting control parameters (i.e., defrosting individuals) at the corresponding time, storing this historical data in the database. Because the shape and thickness of frost on the vehicle's windshield have self-similarity, the feature matrices in the database can cover common frost patterns and provide a validated optimal defrosting solution for each pattern. Within a preset target time period, the system acquires new feature matrices. By calculating the similarity between the newly acquired target feature matrix and each feature matrix in the database (using the reciprocal of the Euclidean distance as the similarity index), the feature matrix with the highest similarity is selected. The matching standard feature matrix corresponds to a historically validated defrosting individual, i.e., the optimal combination of defrosting parameters. The system directly applies the matched defrosting individual to the defrosting control within the target time period, achieving rapid and precise removal of frost from the windshield by dynamically adjusting the angles and power of N defrosting units. This method leverages the self-similarity of historical data to achieve rapid parameter matching and response, reducing the complexity of real-time calculations and improving defrosting efficiency.

[0117] In one embodiment of the present invention, the system sets a similarity threshold. When the similarity between the target feature matrix and the most similar feature matrix in the database is lower than the threshold, the system will perform online optimization calculation to generate the corresponding target defrosting individual and store the defrosting individual and its feature matrix in the database; otherwise, the defrosting individual will be generated directly through similarity matching.

[0118] An automatic defrosting system for new energy vehicles, applied in any of the automatic defrosting methods for new energy vehicles, includes:

[0119] The first acquisition module is used to acquire the first image and the second image of the windshield in response to the engine shutdown command and start command adjacent to the target vehicle.

[0120] The second acquisition module is used to respond to the defrosting command of the target vehicle, perform edge detection on the first image and the second image respectively, and determine the frost outline in the second image;

[0121] The third acquisition module is used to project a uniform light curtain onto the windshield to acquire a third image; and to segment the first and third images based on the frost outline to obtain a feature matrix;

[0122] The data analysis module is used to divide the feature matrix corresponding to the feature area of ​​the windshield into M equal-sized sub-regions, and determine the frost thickness of each sub-region based on the feature matrix.

[0123] The glass defrosting module is used to dynamically control the defrosting angle and defrosting power of N defrosting units based on the frost thickness of each sub-region, so as to reduce defrosting time.

[0124] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A new energy vehicle automatic defrosting method, characterized in that, include: Step 1: In response to the engine shutdown command and start command adjacent to the target vehicle, acquire the first image and the second image of the windshield respectively; Step 2: In response to the defrosting command of the target vehicle, perform edge detection on the first image and the second image respectively to determine the frost outline in the second image; Step 3: Project a uniform light screen onto the windshield to obtain a third image; The first and third images are segmented based on the frost outline to obtain a feature matrix, including: The frost outline is fitted to the third image, and the pixels in the third image located inside the frost outline are retained to obtain the first initial feature image; The frost outline is fitted to the first image, and the pixels in the first image located inside the frost outline are retained to obtain the second initial feature image; The first initial feature image and the second initial feature image are converted to grayscale respectively, and the difference is calculated to obtain the feature matrix; Step 4: Divide the feature matrix corresponding to the feature area of ​​the windshield into M equal-sized sub-regions, and determine the frost thickness of each sub-region based on the feature matrix, including: Construct a frost thickness prediction model, including sample data and sample labels; Within a historical time period, steps 1-3 are repeated K times to obtain K feature matrices. Obtain the normalized element value of the element in the x-th row and y-th column of each feature matrix as sample data; The element in the x-th row and y-th column of each feature matrix corresponds to the normalized frost thickness at the windshield position and is used as the sample label. A frost thickness prediction model was trained based on sample data and sample labels. The frost thickness prediction model is built on support vector machine and backpropagated through mean squared error loss function to update the hyperparameters of the frost thickness prediction model. Within a preset time period, based on the frost thickness prediction model and feature matrix, the frost thickness at each corresponding position on the windshield is obtained. Obtain the frost thickness at each location within the m-th sub-region, and calculate the average value to obtain the frost thickness of the m-th sub-region; where 1≤m≤M, and m is a positive integer; Step 5: Based on the frost thickness of each sub-region, dynamically control the defrosting angle and defrosting power of N defrosting units to reduce defrosting time.

2. The automatic defrosting method for new energy vehicles according to claim 1, characterized in that, Step 2, determining the frost outline in the second image, includes: Edge detection is performed on the first image and the second image to obtain a first edge matrix and a second edge matrix; wherein the first edge matrix and the second edge matrix are both Boolean matrices, with elements 1 and 0 representing edge features and non-edge features respectively; The frosting outline is calculated as follows: ； in, Indicates the outline of frost. express Activation function Represents the first edge matrix. This represents the second edge matrix.

3. The automatic defrosting method for new energy vehicles according to claim 1, characterized in that, In step 3, projecting a uniform light curtain onto the windshield is achieved using the AR-HUD projection unit of the target vehicle.

4. The automatic defrosting method for new energy vehicles according to claim 1, characterized in that, Step 5, which dynamically controls the defrosting angle and defrosting power of the N defrosting units, includes: Step 61: Initialize and generate defrost individuals that meet the constraints; wherein, the defrost individual includes: the defrost angle G and defrost power P of N defrost units; Step 62, the constraints include: , ;in, and These represent the minimum defrost angle and the maximum defrost angle, respectively. and These represent the minimum defrosting power and the maximum defrosting power, respectively. Step 63, the steps to obtain the fitness value of the defrost individual are as follows: Step 631: Defrost the windshield based on the defrosting individual and acquire a third image at fixed time intervals to obtain the frost thickness of M sub-regions at each moment; Step 632: Sort the sub-regions based on the frost thickness at the initial moment to obtain the first sort; Step 633: Calculate the difference in frost thickness between adjacent moments for each sub-region, and sort the M sub-regions based on the difference to obtain the second sort; Step 634: Accumulate the difference between the positions of the M sub-regions in the first and second sorts respectively to obtain the fitness value of the defrosting individual; Step 64: If the fitness value is less than or equal to the preset fitness threshold, then keep the defrosting individual defrosting the windshield; otherwise, update the defrosting individual based on gradient descent and repeat steps 61-63 until the fitness value is less than or equal to the preset fitness threshold.

5. The automatic defrosting method for new energy vehicles according to claim 4, characterized in that, Step 5, which dynamically controls the defrosting angle and defrosting power of the N defrosting units, also includes: In response to the feature matrices corresponding to adjacent shutdown and start commands of multiple target vehicles, a standard database is constructed, and the defrosting individual corresponding to each feature matrix in the standard database is determined. Within the target time period, obtain the target feature matrix; Calculate the similarity between the target feature matrix and any feature matrix in the standard database, and match the target feature matrix with the feature matrix with the highest similarity. The individual defrosting samples corresponding to the matched feature matrix are applied to the target time period for defrosting.

6. The automatic defrosting method for new energy vehicles according to claim 5, characterized in that, The similarity of the feature matrices includes: Calculate the Euclidean distance between the target feature matrix and any feature matrix in the standard database; The reciprocal of the Euclidean distance is used as the similarity.

7. An automatic defrosting system for new energy vehicles, applied in the automatic defrosting method for new energy vehicles according to any one of claims 1-6, characterized in that, include: The first acquisition module is used to acquire the first image and the second image of the windshield in response to the engine shutdown command and start command adjacent to the target vehicle. The second acquisition module is used to respond to the defrosting command of the target vehicle, perform edge detection on the first image and the second image respectively, and determine the frost outline in the second image; The third acquisition module is used to project a uniform light curtain toward the windshield to obtain a third image; The first and third images are segmented based on the frost outline to obtain a feature matrix, including: The frost outline is fitted to the third image, and the pixels in the third image located inside the frost outline are retained to obtain the first initial feature image; The frost outline is fitted to the first image, and the pixels in the first image located inside the frost outline are retained to obtain the second initial feature image; The first initial feature image and the second initial feature image are converted to grayscale respectively, and the difference is calculated to obtain the feature matrix; The data analysis module is used to divide the feature matrix corresponding to the feature area of ​​the windshield into M equal-sized sub-regions, and determine the frost thickness of each sub-region based on the feature matrix, including: Construct a frost thickness prediction model, including sample data and sample labels; Within a historical time period, steps 1-3 are repeated K times to obtain K feature matrices. Obtain the normalized element value of the element in the x-th row and y-th column of each feature matrix as sample data; The element in the x-th row and y-th column of each feature matrix corresponds to the normalized frost thickness at the windshield position and is used as the sample label. A frost thickness prediction model was trained based on sample data and sample labels. The frost thickness prediction model is built on support vector machine and backpropagated through mean squared error loss function to update the hyperparameters of the frost thickness prediction model. Within a preset time period, based on the frost thickness prediction model and feature matrix, the frost thickness at each corresponding position on the windshield is obtained. Obtain the frost thickness at each location within the m-th sub-region, and calculate the average value to obtain the frost thickness of the m-th sub-region; where 1≤m≤M, and m is a positive integer; The glass defrosting module is used to dynamically control the defrosting angle and defrosting power of N defrosting units based on the frost thickness of each sub-region, so as to reduce defrosting time.

Images