An automatic deviation-correcting door panel intelligent cutting processing optimization method and optimization system

Abstract

The embodiment of the present invention discloses an automatic deviation-correcting intelligent cutting optimization method and system for door panels, which belongs to the field of industrial automation technology. The embodiment of the present invention performs point cloud recognition and mines image descriptors on the target door panel laser sensor detection information, covering multiple information such as size, contour and surface state, and can comprehensively and accurately grasp the initial state of the door panel. For multiple cutting task events, deviation correction discrimination and processing parameter mining are performed respectively to determine the automatic deviation correction execution strategy. This effectively improves the processing accuracy of door panels and can reduce errors caused by manual measurement and adjustment. It can adapt to the processing requirements of different door panel shapes, sizes and surface states, optimize the cutting process, reduce the scrap rate, and improve production efficiency. At the same time, it can reduce production costs and enhance the adaptability of door panel processing in different application scenarios.

Description

Technical Field

[0001] The embodiments of the present invention belong to the field of industrial automation technology, and specifically relate to an intelligent cutting processing optimization method and optimization system for door panels with automatic deviation correction. Background Art

[0002] Traditional door panel cutting methods face numerous challenges. Due to a lack of accurate detection and analysis of the initial door panel condition, insufficient precision often occurs during processing. For example, manual measurement of door panel size, shape, and surface condition is difficult to achieve comprehensive and accurate results. This results in an inability to properly adjust machining parameters based on the panel's actual condition during cutting, leading to processing deviations. This not only increases scrap rates and production costs, but also limits production efficiency. Furthermore, traditional processing methods struggle to adapt to diverse door panel processing tasks and complex shape and surface requirements, lacking flexibility to meet diverse door panel processing needs. Summary of the Invention

[0003] The embodiments of the present invention provide an automatic deviation-correcting intelligent cutting processing optimization method and optimization system for door panels, which can solve or partially solve the technical problems involved in the above-mentioned background technology.

[0004] An embodiment of the present invention provides an automatic deviation-correcting intelligent cutting processing optimization method for door panels, which is applied to an optimization system. The method includes: performing point cloud recognition on target laser sensor detection information of a target door panel to obtain initial point cloud state data corresponding to the target laser sensor detection information; performing an image descriptor mining operation on the initial point cloud state data to obtain a target point cloud image descriptor corresponding to the initial point cloud state data; wherein the target point cloud image descriptor includes at least a size descriptor, a contour shape descriptor, and a surface state descriptor; determining X cutting processing task events related to the target laser sensor detection information, performing a deviation correction discrimination operation on the target point cloud image descriptor according to each cutting processing task event, and obtaining a deviation correction discrimination recommendation feature corresponding to each cutting processing task event; X is a positive integer; performing a processing parameter mining operation on the initial point cloud state data to obtain a processing parameter feature corresponding to each cutting processing task event; and determining an automatic deviation correction execution strategy for the target door panel based on the deviation correction discrimination recommendation feature corresponding to each cutting processing task event and the processing parameter feature corresponding to each cutting processing task event.

[0005] An embodiment of the present invention provides an optimization system, comprising at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor performs the above-mentioned method.

[0006] An embodiment of the present invention provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above method are implemented.

[0007] The embodiment of the present invention performs point cloud recognition and mines image descriptors on the target door panel laser sensor detection information, covering multiple aspects of information such as size, contour and surface state, and can comprehensively and accurately grasp the initial state of the door panel. For multiple cutting task events, correction judgment and processing parameter mining are performed separately to determine the automatic correction execution strategy. This effectively improves the processing accuracy of door panels and can reduce errors caused by manual measurement and adjustment. It can adapt to the processing requirements of different door panel shapes, sizes and surface states, optimize the cutting process, reduce the scrap rate, improve production efficiency, and at the same time reduce production costs and enhance the adaptability of door panel processing in different application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A flow chart of an automatic deviation-correcting door panel intelligent cutting optimization method provided by an embodiment of the present invention;

[0009] Figure 2 A schematic structural diagram of an optimization system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0010] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort shall fall within the scope of protection of the embodiments of the present invention.

[0011] The terms "first," "second," and the like in the embodiments of the present invention are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present invention can be implemented in an order other than that illustrated or described herein, and that the objects distinguished by "first," "second," and the like are generally of the same type, and do not limit the number of objects; for example, the first object can be one or more. Furthermore, in the embodiments of the present invention, "and / or" refers to at least one of the connected objects, and the character " / " generally indicates that the objects connected are in an "or" relationship.

[0012] It should be noted that the formula parameters recorded in the subsequent content are formulas in LaTeX format (an existing formula editing format). Those skilled in the art can restore the corresponding original formula expression based on the formula parameter content recorded in the original technical solution.

[0013] Figure 1 A door panel intelligent cutting process optimization method with automatic deviation correction is shown, which is applied to an optimization system. The method includes the following steps 101 to 105.

[0014] Step 101: The optimization system performs point cloud recognition on target laser sensor detection information of the target door panel to obtain initial point cloud state data corresponding to the target laser sensor detection information;

[0015] Step 102: The optimization system performs an image descriptor mining operation on the initial point cloud state data to obtain a target point cloud image descriptor corresponding to the initial point cloud state data; wherein the target point cloud image descriptor includes at least a size descriptor, a contour shape descriptor, and a surface state descriptor;

[0016] Step 103: The optimization system determines X cutting task events related to the target laser sensor detection information, and performs a correction discrimination operation on the target point cloud image descriptor according to each cutting task event to obtain a correction discrimination recommendation feature corresponding to each cutting task event; X is a positive integer.

[0017] Step 104: The optimization system performs a processing parameter mining operation on the initial point cloud state data to obtain processing parameter features corresponding to each cutting task event;

[0018] Step 105: The optimization system determines the automatic deviation correction execution strategy of the target door panel based on the deviation correction discrimination suggestion feature corresponding to each cutting task event and the processing parameter feature corresponding to each cutting task event.

[0019] In an embodiment of the present invention, step 101 involves a technical solution for point cloud recognition processing of target laser sensor detection information of a target door panel.

[0020] During door panel processing optimization, laser sensing is a key step in obtaining information about the panel's initial state. The laser sensor transmits a laser beam to the target door panel surface and then receives the reflected laser signal. The distance between each point on the door panel surface and the sensor is determined based on the laser's time-of-flight (TOF) or triangulation principle. For example, using the TOF principle, if the laser propagation speed is c = 3 x 108 m / s and the round-trip time is t = 10-9 s, then the distance from the measurement point to the sensor is d = frac{ct}{2} = 0.15 m.

[0021] By moving the laser sensor along a certain scanning path (such as a serpentine scanning path) on the target door panel surface or using multiple groups of laser sensors to scan simultaneously, the distance information of a large number of discrete points on the door panel surface can be obtained. The information collection of these discrete points constitutes the target laser sensor detection information.

[0022] After receiving the target laser sensor detection information, the optimization system performs point cloud recognition. Point cloud recognition aims to convert this discrete distance information into point cloud data with spatial coordinate information, namely the initial point cloud state data. Each point in the point cloud data has three coordinate values ​​(x, y, z), where x and y represent the coordinate position of the point within the plane of the door panel, and z represents the vertical distance of the point relative to the sensor, that is, the height of the door panel surface at that point. For example, for a door panel measuring 1m*2m, laser scanning may generate point cloud data with tens of thousands of points. This point cloud data can accurately reflect the microscopic and macroscopic geometric characteristics of the door panel surface.

[0023] Next, step 102 enters into the image descriptor mining operation process of the initial point cloud state data.

[0024] An image descriptor is a mathematical representation that can concisely describe the features of an image or point cloud data. Mining image descriptors for initial point cloud data can transform complex point cloud data into feature vectors that are easier to analyze and process.

[0025] The size descriptor primarily reflects the overall size of the target door panel and the dimensions of key components. For example, for a rectangular door panel, its basic dimensions, such as length, width, and thickness, can be determined from point cloud data. For example, the difference between the maximum and minimum x-coordinates in the point cloud data is δx, the difference between the maximum and minimum y-coordinates is δy, and the average z-coordinate represents the thickness h of the door panel. These values ​​constitute part of the size descriptor.

[0026] In practical applications, it may also be necessary to consider the impact of door panel edge chamfers, holes, and other structures on the size. For example, if there is a circular hole with a diameter of d = 0.1m on the door panel, the space occupied by the hole must be accurately identified and deducted when calculating the size descriptor.

[0027] The contour descriptor is used to describe the external contour shape features of a door panel. A common method is to construct the door panel's contour curve based on point cloud data and then use a curve fitting algorithm (such as least squares curve fitting) to extract the contour descriptor. For example, for a door panel with curved edges, the coefficients a, b, and c in the quadratic curve equation y=ax²+bx+c, obtained by fitting the edge point cloud data, can be used as part of the contour descriptor.

[0028] Alternatively, the ShapeContext algorithm can be used to describe the point cloud contour shape. This algorithm calculates the relative position of each point on the point cloud contour relative to other points and constructs a shape description matrix. The element values ​​in this matrix reflect the shape characteristics of the door panel contour.

[0029] The surface state descriptor primarily focuses on information such as the flatness and roughness of the door panel surface. Surface flatness can be measured by calculating the variance sigma_{z}^{2} of the z coordinate in the point cloud data. A small sigma_{z}^{2} value indicates a relatively flat door panel surface; a small value indicates an uneven surface.

[0030] Regarding surface roughness, a point cloud-based roughness assessment algorithm can be used. For example, the standard deviation of the distances between adjacent points in the point cloud data can be calculated. For example, the average distance between adjacent points is overline{d}, and the standard deviation is sigma_{d}. A larger sigma_{d} indicates greater surface roughness. These calculated values ​​constitute the surface state descriptor. Through this mining operation, the target point cloud image descriptor corresponding to the initial point cloud state data is obtained. It comprehensively describes the characteristics of the target door panel, such as size, contour shape, and surface state.

[0031] Then, step 103 focuses on the correction judgment operation of the cutting task event.

[0032] The optimization system determines X cutting task events related to the target laser sensor detection information. These cutting task events can be determined based on the door panel's processing requirements. For example, edge cutting, surface milling, and hole processing can all be considered different cutting task events. For example, if X = 3, the three cutting task events are cutting the left edge of the door panel, milling the entire surface of the door panel by 0.5 mm, and processing the hole in the lower right corner of the door panel.

[0033] For each cutting task event, a correction is performed based on the target point cloud image descriptor. For example, for a door panel's left edge cutting task, if the door panel width indicated by the size descriptor is 0.2m wider than the design requirement, and the contour shape descriptor indicates an outward bulge on the left edge, the correction requires adjusting the cutting parameters to correct these deviations.

[0034] For the aforementioned door panel left edge cutting task event, the recommended features for correcting the deviation might include the required additional cutting depth h_{1} = 0.1m on the left edge, the coordinate adjustments for the start and end points of the cut (δx_{1}, δy_{1}), and so on. For the door panel surface milling task event, if the surface condition descriptor indicates surface unevenness, the recommended features for correcting the deviation might include the milling tool feed speed adjustment v_{1} = -0.05m / s (the negative sign indicates a reduced speed to improve milling accuracy), the milling depth fine-tuning δh_{2} = 0.02m, and so on. For the hole machining task event in the lower right corner of the door panel, if the size descriptor indicates a deviation in the hole position, the recommended features for correcting the deviation might include the adjustment for the hole machining center coordinates (δx_{3}, δy_{3}). Through this operation, the recommended features for correcting the deviation corresponding to each cutting task event are obtained.

[0035] Then, step 104 further turns to the processing parameter mining operation of the initial point cloud state data.

[0036] The machining parameter mining operation aims to determine the appropriate machining parameters for each cutting task event based on the initial point cloud state data. These machining parameters include tool speed, feed rate, cutting depth, cutting width, etc.

[0037] For the task event for cutting the left edge of the door panel, the tool's cutting depth is determined based on information such as the door panel thickness h from the size descriptor and the edge curvature from the contour shape descriptor. For example, if the door panel thickness h = 0.05m and the edge curvature is small, the tool's cutting depth can be set to h_{4} = 0.02m. Simultaneously, the tool's rotational speed n_{1} is determined based on the surface roughness from the surface state descriptor. If the surface roughness requirement is low, a higher tool speed, such as n_{1} = 5000 rpm, can be used to achieve a smoother cut surface.

[0038] For the entire door panel surface milling task event, the milling tool's feed rate v_{2} and milling depth h_{5} are determined based on the door panel's dimensions (determined by the size descriptor) and the surface condition descriptor. For larger door panels with good surface flatness, a higher feed rate (for example, v_{2} = 0.1 m / s and a milling depth h_{5} = 0.03 m) can be used to improve machining efficiency.

[0039] For the hole machining task event in the lower right corner of the door panel, the drilling tool's rotational speed n_{2} and feed rate v_{3} are determined based on the hole diameter d in the size descriptor and the shape features surrounding the hole in the contour shape descriptor. If the hole diameter d is 0.1m and the shape surrounding the hole is relatively regular, the drilling tool's rotational speed can be set to n_{2} = 3000 r / min and the feed rate v_{3} = 0.02m / s. Through this processing parameter mining operation, the processing parameter features corresponding to each cutting task event are obtained.

[0040] Finally, step 105 is used to determine the automatic deviation correction execution strategy for the target door panel.

[0041] The optimization system determines the automatic correction strategy for the target door panel based on the correction recommendation feature and processing parameter features corresponding to each cutting task event. For the left edge cutting task event, the additional cutting depth h_{1}=0.1m in the correction recommendation feature is combined with the tool cutting depth h_{4}=0.02m in the processing parameter feature. If the correction process requires initial cutting according to the cutting depth h_{4} in the processing parameter feature, then the remaining deviation is adjusted based on the correction recommendation feature.

[0042] For the entire door panel surface milling task, the milling tool feed rate adjustment v_{1} = -0.05 m / s in the deviation correction recommendation feature is combined with the feed rate v_{2} = 0.1 m / s in the machining parameter feature to obtain the actual feed rate v = v_{2} + v_{1} = 0.05 m / s. Simultaneously, the milling depth fine-tuning δh_{2} = 0.02 m in the deviation correction recommendation feature is combined with the milling depth h_{5} = 0.03 m in the machining parameter feature to determine the final milling depth h = h_{5} + δh_{2} = 0.05 m.

[0043] For the hole machining task event in the lower right corner of the door panel, the adjustment amount (δx_{3}, δy_{3}) for the hole machining center coordinates in the correction discrimination suggestion feature is combined with the drilling tool speed n_{2} and feed rate v_{3} in the machining parameter feature. While adjusting the hole machining center coordinates, the hole is machined at the determined speed and feed rate. This comprehensive processing of each cutting task event determines a complete automatic correction execution strategy for the target door panel, enabling precise correction of door panel deviations during the cutting process, improving the quality and precision of the door panel processing.

[0044] In order to facilitate a clear and complete understanding of the above technical solution, the size descriptor, contour shape descriptor and surface state descriptor are further introduced below.

[0045] 1. Size Descriptor (Size Feature)

[0046] (1) Basic size information

[0047] Linear dimensions: For a rectangular door panel, a coordinate system is established with one corner of the panel as the origin. The difference between the maximum (x_{max}) and minimum (x_{min}) values ​​of the point cloud data (x) coordinate represents the panel's length (L_x) in the (x) direction. The difference between the maximum (y_{max}) and minimum (y_{min}) values ​​of the (y) coordinate represents the panel's width (L_y) in the (y) direction. The average (overline{z}) value of the (z) coordinate represents the panel's thickness (h).

[0048] For example, after laser scanning and point cloud recognition, we obtain (x_{max}=1.2m), (x_{min}=0m), (y_{max}=2.1m), (y_{min}=0m), and (overline{z}=0.05m). The linear dimension feature vector of this part of the dimension descriptor can be expressed as ([L_x, L_y, h]=[1.2m, 2.1m, 0.05m]).

[0049] Hole and Depression Dimensions: If a door panel has holes or depressions, their dimensions need to be described. For example, if there is a circular hole in the door panel, the point cloud data can be used to identify the point cloud of the hole's edge. The hole's diameter (d) and the coordinates of the circle's center ((x_0, y_0)) can be calculated.

[0050] For example, if the diameter of a hole is identified (d=0.1m) and the coordinates of the center are ((0.5m, 1.0m)), then this part of the feature vector can be expanded to ([L_x, L_y, h, d, x_0, y_0]=[1.2m, 2.1m, 0.05m, 0.1m, 0.5m, 1.0m]).

[0051] (2) Relative size ratio

[0052] Aspect Ratio: Calculate the door panel's aspect ratio (r = L_x / L_y). In the example above, (r = 1.2 / 2.1 ≈ 0.571). This aspect ratio can also be used as part of the size descriptor, and the feature vector is further expanded to ([L_x, L_y, h, d, x_0, y_0, r] = [1.2m, 2.1m, 0.05m, 0.1m, 0.5m, 1.0m, 0.571]).

[0053] Hole-to-door panel size ratio: Calculate the ratio (p = d / L_y) of the hole diameter to the minimum door panel size (e.g., width (L_y)). Here, p = 0.1 / 2.1 ≈ 0.048. The final size descriptor feature vector can be expressed as ([L_x, L_y, h, d, x_0, y_0, r, p] = [1.2m, 2.1m, 0.05m, 0.1m, 0.5m, 1.0m, 0.571, 0.048]).

[0054] 2. Contour shape descriptor (contour shape feature)

[0055] (1) Features based on curve fitting

[0056] Edge Contour Curve Fitting: For the edge contour of a door panel, we use polynomial curve fitting. For example, fitting the point cloud data for a door panel edge yields a quadratic polynomial (y = ax^{2} + bx + c). For example, the fitting yields (a = 0.01), (b = -0.1), and (c = 0.2). These three coefficients can be used as part of a contour shape descriptor to represent the shape characteristics of this edge.

[0057] For the four edges of the door panel, we can fit them separately to obtain four sets of coefficients. For example, if the fitting coefficients of the four edges are ((a_1, b_1, c_1)), ((a_2, b_2, c_2)), ((a_3, b_3, c_3)), and ((a_4, b_4, c_4)), then the feature vector of the shape descriptor for this part of the contour can be expressed as ([a_1, b_1, c_1, a_2, b_2, c_2, a_3, b_3, c_3, a_4, b_4, c_4] = [0.01, -0.1, 0.2, -0.02, 0.05, 0.15, 0.015, -0.08, 0.18, -0.01, -0.06, 0.16]).

[0058] Corner features: If a door panel has corners, you can also describe features such as the curvature of the corner. For example, for a right-angled corner, you can calculate the radius of curvature (R). For example, if the radius of curvature of a corner is R = 0.05m, then the feature vector can be further expanded to ([a_1, b_1, c_1, a_2, b_2, c_2, a_3, b_3, c_3, a_4, b_4, c_4, R] = [0.01, -0.1, 0.2, -0.02, 0.05, 0.15, 0.015, -0.08, 0.18, -0.01, -0.06, 0.16, 0.05m]).

[0059] (2) Features based on shape context

[0060] Constructing the shape context description matrix: Taking the point cloud data of the door panel's outer contour as an example, select (n) points on the contour (e.g., (n = 100)). For each point (P_i), calculate its relative position relative to the other (n-1) points. Divide the entire contour into several sectors (e.g., (m = 8) sectors), and calculate the distribution of each point within each sector.

[0061] This constructs an (n*m) shape description matrix (S). For example, for a point (P_{10}), the number of points in the 8 fan-shaped areas may be distributed as ([5, 3, 2, 1, 0, 1, 2, 3]). The element values ​​in the entire shape description matrix (S) constitute part of the contour shape descriptor.

[0062] Eigenvector representation: Arrange the elements of the shape description matrix (S) in a specific order to form a eigenvector. For example, expanding the matrix (S) row by row yields a eigenvector (V_s) of length (n*m=100*8=800). This eigenvector (V_s) is part of the contour shape descriptor based on shape context and can be combined with the curve-fitting eigenvector to form a complete contour shape descriptor.

[0063] 3. Surface state descriptor (surface state features)

[0064] (1) Surface flatness characteristics

[0065] Features based on (z) coordinate variance: Calculate the variance (sigma_{z}^{2}) of the (z) coordinates in the point cloud data to measure surface flatness. For example, the calculated value is (sigma_{z}^{2} = 0.001m^{2}). This value can be directly used as a feature of the surface state descriptor to indicate the degree of surface flatness. Local flatness features: Divide the door panel surface into several small areas (for example, (k=10*10) small square areas) and calculate the (z) coordinate variance (sigma_{z_{ij}}^{2}) for each small area ((i=1, 2, cdots, 10), (j=1, 2, cdots, 10)). Arrange the variance values ​​of these small areas into a vector ([sigma_{z_{11}}^{2}, sigma_{z_{12}}^{2}, cdots, sigma_{z_{10, 10}}^{2}]) in a certain order, which can describe the local differences in surface flatness in more detail;

[0066] (2) Surface roughness characteristics

[0067] Adjacent point distance standard deviation feature: Calculate the standard deviation (sigma_{d}) of the distances between adjacent points in the point cloud data. For example, after calculation, we get (sigma_{d} = 0.002m). This value is used as a feature of surface roughness to indicate the roughness of the surface. Roughness feature based on height difference: Calculate the average value of the absolute value of the (z) coordinate height difference between adjacent points (overline{δz}). For example, we get (overline{δz} = 0.0005m). This value can also be used as a feature of surface roughness, together with the standard deviation of the adjacent point distances, to more comprehensively describe the surface state. The final surface state descriptor feature vector can be expressed as ([sigma_{z}^{2}, sigma_{d}, overline{δz}] = [0.001m^{2}, 0.002m, 0.0005m]). If local flatness features are considered, the corresponding vector can also be added.

[0068] The embodiment of the present invention performs point cloud recognition and mines image descriptors on the target door panel laser sensor detection information, covering multiple aspects of information such as size, contour and surface state, and can comprehensively and accurately grasp the initial state of the door panel. For multiple cutting task events, correction judgment and processing parameter mining are performed separately to determine the automatic correction execution strategy. This effectively improves the processing accuracy of door panels and can reduce errors caused by manual measurement and adjustment. It can adapt to the processing requirements of different door panel shapes, sizes and surface states, optimize the cutting process, reduce the scrap rate, improve production efficiency, and at the same time reduce production costs and enhance the adaptability of door panel processing in different application scenarios.

[0069] In a possible technical solution, the target laser sensor detection information of the target door panel is subjected to point cloud recognition to obtain initial point cloud state data corresponding to the target laser sensor detection information, including: performing sensor signal preprocessing on the target laser sensor detection information to obtain the initial laser sensor signal of the target laser sensor detection information; and performing point cloud coding recognition on the initial laser sensor signal to obtain the initial point cloud state data of the initial laser sensor signal.

[0070] Among them, the sensor signal preprocessing of the target laser sensor detection information to obtain the initial laser sensor signal of the target laser sensor detection information includes: when the target laser sensor detection information of the target door panel is obtained, obtaining a target deviation analysis algorithm for performing deviation analysis on the target door panel; the target deviation analysis algorithm includes a first genetic iterative branch; determining the sensor element characteristics related to the target laser sensor detection information through the first genetic iterative branch, and extracting the sensor detection input reflection signal of the target laser sensor detection information based on the sensor element characteristics; performing signal decoding on the sensor detection input reflection signal through the first genetic iterative branch to obtain a signal decoding result corresponding to the sensor detection input reflection signal; performing position distribution feature mapping on the signal decoding result corresponding to the sensor detection input reflection signal through the first genetic iterative branch to obtain the initial laser sensor signal of the target laser sensor detection information.

[0071] Furthermore, the first genetic iterative branch includes an excitation module; the sensor detection input reflection signal includes Y sensor detection input reflection signals, and the Y sensor detection input reflection signals include a target sensor detection input reflection signal, where Y is a positive integer; based on this, the sensor detection input reflection signal is decoded by the first genetic iterative branch to obtain a signal decoding result corresponding to the sensor detection input reflection signal, including: determining a characteristic coordinate system corresponding to the target sensor detection input reflection signal based on the sensor element characteristics to which the target sensor detection input reflection signal belongs; the characteristic coordinate system includes multiple feature projection variables, each feature projection variable has a corresponding input reflection signal mapping set; determining the input reflection signal mapping set to which the target sensor detection input reflection signal belongs through the excitation module, and determining the feature projection variable corresponding to the input reflection signal mapping set to which the target sensor detection input reflection signal belongs as the signal decoding result corresponding to the target sensor detection input reflection signal.

[0072] When performing point cloud recognition on the target laser sensor detection information of the target door panel, the sensor signal must first be preprocessed to obtain an initial laser sensor signal. During this process, a target deviation analysis algorithm for target door panel deviation analysis is obtained, which includes a first genetic iteration branch.

[0073] The first genetic iteration branch is used to determine sensor element features related to the target laser sensor detection information. Sensor element features here refer to those related to factors that affect the accuracy of laser sensor detection results, such as the features corresponding to factors such as the laser sensor's wavelength, emission angle, and receiving sensitivity.

[0074] For example, the wavelength of a laser sensor is (lambda=650nm), the emission angle is (theta=30^{circ}), and the receiving sensitivity is (S=0.8) (the sensitivity here refers to the ability of the sensor to receive the reflected signal and accurately convert it into an electrical signal). The characteristics corresponding to these parameters are part of the sensing element characteristics.

[0075] Based on these sensing element characteristics, the sensor detection input and reflection signals can be extracted from the target laser sensor detection information. For example, for a specific laser sensor device, when laser light is emitted from the door panel surface and reflected back, the aforementioned sensing element characteristics can accurately separate the sensor detection input and reflection signals from the received mixed signal.

[0076] The excitation module in the first genetic iteration branch plays a key role in the signal decoding process. Since there are (Y) sensor detection input reflection signals (for example, (Y = 5)), the corresponding feature coordinate system for the target sensor detection input reflection signal must be determined based on the characteristics of the sensor element to which it belongs.

[0077] The characteristic coordinate system consists of multiple characteristic projection variables, each of which has a corresponding input and reflection signal mapping set. For example, the characteristic coordinate system has three characteristic projection variables (x_1), (x_2), and (x_3), and the corresponding input and reflection signal mapping sets for each variable are (M_1), (M_2), and (M_3).

[0078] The excitation module analyzes the characteristics of the target sensor's incoming reflection signal to determine the incoming reflection signal mapping set to which it belongs. For example, if certain characteristics of the target sensor's incoming reflection signal match those in (M_2), then (x_2) is determined to be the signal decoding result corresponding to the target sensor's incoming reflection signal. This process is similar to classifying a signal into a specific region (represented by a feature projection variable) based on its characteristics in a multidimensional space.

[0079] After obtaining the signal decoding result corresponding to the sensor detection input reflection signal, the signal decoding result is subjected to position distribution feature mapping through the first genetic iteration branch. For example, if the signal decoding result is (x_2), the position distribution feature mapping may involve converting the signal into a signal related to the door panel surface position based on the features represented by (x_2) and a pre-established relationship model between door panel surface position and signal features. For example, a mapping function (f(x_2)), where (f) may be a complex function based on the door panel geometry and the laser scanning path, is used to convert (x_2) into (z=f(x_2)), where (z) is the signal value related to the position of a point on the door panel surface. By performing this processing on all sensor detection input reflection signals, the initial laser sensor signal of the target laser sensor detection information is ultimately obtained. This signal contains pre-processed and converted information related to the door panel surface, providing the basis for subsequent point cloud code recognition.

[0080] After obtaining the initial laser sensor signal, point cloud coding and recognition are performed to generate initial point cloud state data. Point cloud coding and recognition is the process of converting the information in the initial laser sensor signal into point cloud data. For example, the initial laser sensor signal contains a series of discrete signal values ​​(z_1, z_2, cdots, z_n). Using a specific coding algorithm (C), these signal values ​​can be converted into point coordinates (x_i, y_i, z_i) in the point cloud data. This coding algorithm (C) may be based on the laser sensor's scanning pattern and coordinate transformation relationship. For example, if the laser sensor scans the door panel surface in a row-column scanning pattern, (x_i) and (y_i) can be calculated based on the scanned row and column numbers and the laser sensor's resolution. (z_i) is the signal value related to the door panel surface height after the previous processing. Through this point cloud coding and recognition process, initial point cloud state data is obtained. This data accurately describes the target door panel surface geometry in the form of a point cloud, including the position and relative height of each point on the door panel surface.

[0081] Next, image descriptor mining is performed on the initial point cloud state data. Dimensional descriptors can be mined by calculating the range of the point cloud data along each coordinate axis. For example, for the x-axis, the maximum (x_{max}) and minimum (x_{min}) values ​​of the x-coordinate in the point cloud data are calculated. The x-axis dimension of the door panel is then calculated as L_x = x_{max} - x_{min}. For example, if x_{max} = 1.2m and x_{min} = 0m, then L_x = 1.2m.

[0082] Contour shape descriptors can be mined using curve fitting algorithms, such as the least squares method. For example, for the point cloud data of a door panel edge, the edge point cloud data can be fitted with a quadratic function (y = ax^{2} + bx + c), and the coefficients (a), (b), and (c) can be calculated using the least squares method. For example, by calculating the edge point cloud data ((x_1, y_1), (x_2, y_2), cdots, (x_m, y_m)), we obtain (a = 0.01), (b = -0.1), and (c = 0.2). These coefficients constitute part of the contour shape descriptor.

[0083] To mine surface state descriptors, we can calculate the variance (sigma_{z}^{2}) of the z coordinates in the point cloud data. For example, if there are n points in the point cloud data, with z coordinates (z_1, z_2, cdots, z_n), the variance formula is (sigma_{z}^{2}=frac{1}{n}sum_{i=1}^{n}(z_i - overline{z})^2), where (overline{z}=frac{1}{n}sum_{i=1}^{n}z_i). If the calculated value (sigma_{z}^{2}=0.001m^{2}) reflects the surface flatness and serves as part of the surface state descriptor. Through these operations, we obtain the target point cloud image descriptor, which contains information on dimensions, contour shape, and surface state.

[0084] Next, determine the (X) cutting processing task events related to the target laser sensor detection information (for example, (X=3), respectively, door panel edge cutting, surface milling, and hole processing).

[0085] For each cutting task event, a correction and discrimination operation is performed based on the target point cloud image descriptor. For example, for door panel edge cutting, if the door panel width indicated by the size descriptor is wider than the design requirement (δL = 0.1m), and the contour shape descriptor indicates an outward bulge at the edge, then the correction feature recommendation might include the depth (h_1) required for additional cutting at the edge. A computational model can be established based on factors such as the door panel material and the tool's cutting capability, such as (h_1 = frac{δL}{k}), where (k) is a coefficient determined empirically or experimentally. For example, if (k = 5), then (h_1 = 0.02m).

[0086] At the same time, processing parameter mining operations are performed on the initial point cloud state data. For the door panel edge cutting task event, the tool cutting depth is determined based on the door panel thickness (h = 0.05m) in the size descriptor and the edge curvature in the contour shape descriptor. For example, according to an empirical formula (h_{4} = htimesalpha), where (alpha) is a coefficient determined based on the door panel material and tool type, for example (alpha = 0.4), then (h_{4} = 0.02m). At the same time, the tool rotation speed (n_1) is determined based on the surface roughness in the surface state descriptor. If the surface roughness requirement is low, in order to obtain a smoother cutting surface, an empirical formula for the relationship between rotation speed and roughness can be used (n_1 = frac{R_{0}}{R}timesn_{0}), where (R_{0}) is the desired roughness, (R) is the current roughness, and (n_{0}) is the base rotation speed. For example, (R_{0}=0.0005m), (R=0.001m), (n_{0}=3000r / min), then (n_1=6000r / min).

[0087] Through such operations, the correction judgment suggestion features and processing parameter features corresponding to each cutting task event are obtained.

[0088] Finally, the automatic correction strategy for the target door panel is determined based on the correction discrimination recommendation feature and processing parameter feature corresponding to each cutting task event. For example, for the door panel edge cutting task event, the additional cutting depth (h_1 = 0.02m) in the correction discrimination recommendation feature is combined with the tool cutting depth (h_{4} = 0.02m) in the processing parameter feature. If, during the correction process, preliminary cutting is required according to the cutting depth (h_{4}) in the processing parameter feature, the remaining deviation is then adjusted based on the correction discrimination recommendation feature. For the door panel surface milling task event, the milling tool feed speed adjustment (v_1 = -0.05m / s) in the correction discrimination recommendation feature is combined with the feed speed (v_{2} = 0.1m / s) in the processing parameter feature to obtain the actual feed speed (v = v_{2} + v_1 = 0.05m / s). The milling depth fine-tuning value (δh_{2} = 0.02m) in the correction discrimination recommendation feature is combined with the milling depth (h_{5} = 0.03m) in the machining parameter feature to determine the final milling depth (h = h_{5} + δh_{2} = 0.05m). This comprehensive processing of each cutting task event determines a complete automatic correction execution strategy for the target door panel, enabling precise correction of door panel deviations during the cutting process, improving the quality and precision of door panel processing.

[0089] This design, firstly, through a unique sensor signal preprocessing process, utilizes the first genetic iteration branch of the target deviation analysis algorithm to accurately extract and process the sensor detection input and reflection signals, thereby improving the accuracy and effectiveness of laser sensor detection information. For example, during the signal decoding process, the characteristic coordinate system and signal decoding results are determined based on the characteristics of the sensor elements, enabling more precise signal analysis. Secondly, the initial point cloud state data obtained by point cloud coding recognition accurately reflects the surface geometry of the door panel. Furthermore, image descriptor mining comprehensively captures various aspects of information such as the door panel's size, contour shape, and surface condition, providing a detailed basis for subsequent processing operations. During the cutting task event processing, through deviation correction and processing parameter mining operations, the processing strategy can be precisely adjusted for different tasks. For example, the appropriate parameters such as cutting depth, rotation speed, and feed rate can be determined based on the actual deviation and surface condition of the door panel. This effectively improves the door panel processing accuracy and reduces the scrap rate, while also increasing production efficiency and reducing production costs, thereby enhancing the intelligence and precision of the door panel processing process.

[0090] In a preferred embodiment, the target point cloud image descriptor is obtained by performing an image descriptor mining operation on the initial point cloud state data by a point cloud description mining module in a first genetic iterative branch; the first genetic iterative branch belongs to a target deviation analysis algorithm for performing deviation analysis on the target door panel; the method further includes a training step of the first genetic iterative branch: obtaining a first algorithm debugging sample; the first priori debugging sample included in the first algorithm debugging sample is a priori debugging sample that does not carry a deviation label; obtaining Z debugging items for the first original genetic iterative branch; Z is a positive integer; processing the first priori debugging sample through the first original genetic iterative branch to obtain a point cloud image sample descriptor associated with each debugging item, and determining Z training error data corresponding to the Z debugging items based on the point cloud image sample descriptor; one debugging item corresponds to one training error data; determining first target training error data based on the Z training error data, improving the algorithm configuration weight of the first original genetic iterative branch based on the first target training error data, and determining the improved first original genetic iterative branch as the first genetic iterative branch.

[0091] In this embodiment, the target point cloud image descriptor is derived by mining the initial point cloud state data using the point cloud description mining module in the first genetic iteration branch of the target deviation analysis algorithm. The first genetic iteration branch plays a key role in the entire deviation analysis algorithm, using its own processing mechanism to mine the target point cloud image descriptor, which contains rich information, from the initial point cloud state data.

[0092] The target point cloud image descriptor here encompasses multiple aspects of information, including size descriptors, contour shape descriptors, and surface state descriptors. Taking the size descriptor as an example, it can accurately reflect dimensional information such as the size of a door panel. For example, it can determine the length, width, and thickness of a door panel by calculating the extreme value difference of coordinates in the point cloud data. For example, if the maximum (x) coordinate in the point cloud data is (x_{max}=1.2m) and the minimum is (x_{min}=0m), then the dimension in the (x) direction is (L_x=x_{max}-x_{min}=1.2m). The contour shape descriptor may describe the edge shape of a door panel through algorithms such as curve fitting, and the surface state descriptor can be used to represent information such as the flatness and roughness of the door panel surface.

[0093] Furthermore, the training steps of the first genetic iteration branch are introduced as follows.

[0094] The first algorithm debugging sample is the fundamental data source for training the first genetic iteration branch. The first algorithm debugging sample includes the first a priori debugging samples, which are unannotated deviation-free. These a priori debugging samples can be the initial point cloud state data of a large number of different door panels. This data was collected under normal production or testing conditions, without any pre-labeled deviations. For example, 100 sets of initial point cloud state data of different door panels were collected as the first a priori debugging samples.

[0095] For the first original genetic iteration branch, (Z) debugging items are set (for example, (Z=5)). These debugging items can cover different aspects of point cloud image descriptor mining, such as debugging items for the accuracy of size descriptor mining, debugging items for the accuracy of contour shape descriptor mining, debugging items for the accuracy of surface state descriptor mining, and debugging items for overall mining efficiency.

[0096] The first prior debugging sample is processed through the first original genetic iterative branch to obtain the point cloud image sample descriptor associated with each debugging item, and the (Z) training error data corresponding to the (Z) debugging items are determined based on the point cloud image sample descriptor.

[0097] For each first-prior debugging example, the first original genetic iteration branch processes it. For example, let's say the actual dimensions of the door panel corresponding to this example are (L_{x_0} = 1.2m), (L_{y_0} = 2.0m), and (h_0 = 0.05m) (length, width, thickness). The first original genetic iteration branch processes this example and obtains the dimension descriptor values ​​of (L_{x_1}), (L_{y_1}), and (h_1).

[0098] For the Size Descriptor Mining Accuracy debugging project, you can calculate the training error data using the error formula (e_{size} = sqrt{(L_{x_1} - L_{x_0})^2 + (L_{y_1} - L_{y_0})^2 + (h_1 - h_0)^2}). Here, (e_{size}) represents the training error data for the Size Descriptor Mining Accuracy debugging project. This formula measures the error by calculating the Euclidean distance between the mined size value and the actual size value.

[0099] For debugging projects testing the accuracy of contour descriptor mining, curve fitting error can be used to measure this. For example, the actual door panel edge contour conforms to a quadratic function (y=a_0x^{2}+b_0x+c_0), and the quadratic function corresponding to the contour descriptor mined by the first original genetic iteration branch is (y=a_1x^{2}+b_1x+c_1). The training error data (e_{contour}) for this debugging project is calculated using the curve fitting error formula (e_{contour}=int_{x_1}^{x_2}[(a_0x^{2}+b_0x+c_0)-(a_1x^{2}+b_1x+c_1)]^2dx) (where (x_1) and (x_2) are the range of values ​​of the door panel edge contour on the (x) axis).

[0100] For the debugging project of surface state descriptor mining accuracy, if we take surface flatness as an example, the actual surface flatness is represented by the (z) coordinate variance (sigma_{z_0}^{2}), and the mined surface flatness is (sigma_{z_1}^{2}), then the training error data (e_{surface}=vertsigma_{z_1}^{2}-sigma_{z_0}^{2}vert).

[0101] For debugging projects that focus on overall mining efficiency, we can calculate the training error data (e_{efficiency} = vertt_1 - t_0vert) based on the difference between the time (t_1) it takes the first original genetic iteration branch to process a priori debugging example and the preset standard time (t_0). In this way, we can obtain corresponding training error data for each debugging project.

[0102] A first target training error data is determined based on the (Z) training error data. The algorithm configuration weights of the first original genetic iteration branch are improved based on the first target training error data, and the improved first original genetic iteration branch is determined as the first genetic iteration branch. The first target training error data can be determined by weighted summation. For example, the training error data corresponding to (Z=5) debugging items are (e_1), (e_2), (e_3), (e_4), and (e_5), and the corresponding weights are (w_1), (w_2), (w_3), (w_4), and (w_5), respectively. (These weights can be pre-set based on the importance of each debugging item, for example, (w_1=0.2), (w_2=0.2), (w_3=0.2), (w_4=0.2), and (w_5=0.2). Then the first target training error data is: (E=sum_{i=1}^{Z}w_ie_i=w_1e_1+w_2e_2+w_3e_3+w_4e_4+w_5e_5).

[0103] The algorithm configuration weights of the first original genetic iteration branch are improved based on the first target training error data (E). For example, for a key parameter (alpha) in the first original genetic iteration branch, its initial weight is (w_{alpha}), and the improvement formula can be (w_{alpha}^{new}=w_{alpha}-etaE) (where (eta) is the learning rate, for example, (eta=0.01)). By making such improvements to the weights of each key parameter, an improved first original genetic iteration branch is obtained, which is designated as the first genetic iteration branch. This trained first genetic iteration branch can more accurately and efficiently obtain the target point cloud image descriptor when mining image descriptors on the initial point cloud state data.

[0104] This design, firstly, by specifically training the first genetic iteration branch, it is possible to improve the accuracy of its mining of target point cloud image descriptors. For example, in dimension descriptor mining, the trained branch can more accurately capture door panel dimensions, reducing deviations from actual dimensions. The same applies to contour shape and surface state descriptor mining, improving the accuracy of the description of various door panel features. Secondly, by calculating training error data for multiple debugging projects and determining the target training error data through weighted summation, the algorithm configuration weights are improved, making the training process comprehensive and reasonable. This helps optimize the performance of the first genetic iteration branch, ensuring good adaptability and stability when processing different initial point cloud state data. Finally, this training method helps improve the accuracy of deviation analysis throughout the door panel processing process, providing a more reliable basis for subsequent cutting tasks, helping to improve door panel processing quality, reduce processing errors caused by inaccurate descriptor mining, and ultimately improve production efficiency and reduce costs.

[0105] In actual application, Z debugging projects include different sub-projects. The following four examples are used to introduce them.

[0106] Example 1: The Z debugging projects include a signal noise addition decoding project; the first a priori debugging sample is processed by the first original genetic iterative branch to obtain a point cloud image sample descriptor associated with each debugging project, and Z training error data corresponding to the Z debugging projects are determined based on the point cloud image sample descriptor, including: performing sensor signal preprocessing on the first a priori debugging sample through the first original genetic iterative branch to obtain a first initial laser sensor signal sample of the first a priori debugging sample; the first initial laser sensor signal sample includes signal decoding result samples under multiple distribution labels; from the multiple distribution labels corresponding to the first initial laser sensor signal sample, arbitrarily determining a first distribution label to be subjected to signal noise addition; and performing signal noise addition processing on the signal decoding result sample under the first distribution label in the first initial laser sensor signal sample based on the signal noise addition information. to obtain a first noisy sample; perform point cloud coding recognition on the first noisy sample through the first original genetic iterative branch to obtain a first initial point cloud state data sample of the first noisy sample, perform an image descriptor mining operation on the first initial point cloud state data sample to obtain a first point cloud image sample descriptor corresponding to the first initial point cloud state data sample; the first point cloud image sample descriptor belongs to the point cloud image sample descriptor; perform a noise decoding operation on the first point cloud image sample descriptor through a decoding module related to the signal noise decoding project to obtain noisy signal decoding prediction information corresponding to the signal noise information in the first noisy sample; determine the training error data corresponding to the signal noise decoding project based on the signal decoding result sample under the first distribution label in the first initial laser sensor signal sample and the noisy signal decoding prediction information.

[0107] Example 2: The Z debugging projects include a signal change decoding project; the first a priori debugging sample is processed by the first original genetic iterative branch to obtain a point cloud image sample descriptor associated with each debugging project, and Z training error data corresponding to the Z debugging projects are determined based on the point cloud image sample descriptor, including: performing sensor signal preprocessing on the first a priori debugging sample through the first original genetic iterative branch to obtain a first initial laser sensor signal sample of the first a priori debugging sample; the first initial laser sensor signal sample includes signal decoding result samples under multiple distribution labels; from the multiple distribution labels corresponding to the first initial laser sensor signal sample, arbitrarily determining a second distribution label to be subjected to signal change; and performing signal change processing on the signal decoding result sample under the second distribution label in the first initial laser sensor signal sample based on the signal change information. , obtaining a second noisy sample; the signal change information is different from the signal decoding result sample under the second distribution label; performing point cloud coding recognition on the second noisy sample through the first original genetic iterative branch to obtain a second initial point cloud state data sample of the second noisy sample, performing an image descriptor mining operation on the second initial point cloud state data sample to obtain a second point cloud image sample descriptor corresponding to the second initial point cloud state data sample; the second point cloud image sample descriptor belongs to the point cloud image sample descriptor; performing a signal change decoding operation on the second point cloud image sample descriptor through a decoding module related to the signal change decoding project to obtain signal change decoding information corresponding to the multiple distribution labels; determining the training error data corresponding to the signal change decoding project based on the second distribution label and the signal change decoding information corresponding to the multiple distribution labels.

[0108] Example 3: The Z debugging projects include an adversarial training project; the first a priori debugging sample is processed through the first original genetic iterative branch to obtain a point cloud image sample descriptor associated with each debugging project, and Z training error data corresponding to the Z debugging projects are determined based on the point cloud image sample descriptor, including: performing adversarial processing on the sensor detection input reflection signal of the first a priori debugging sample to obtain an adversarial debugging sample, performing sensor signal preprocessing on the adversarial debugging sample to obtain a laser sensor adversarial signal of the adversarial debugging sample; performing sensor signal preprocessing on the first a priori debugging sample to obtain a first initial laser sensor signal sample of the first a priori debugging sample; performing point cloud coding recognition on the laser sensor adversarial signal through the first original genetic iterative branch to obtain a third initial point cloud state of the laser sensor adversarial signal. Data sample, performing an image descriptor mining operation on the third initial point cloud state data sample to obtain a third point cloud image sample descriptor corresponding to the third initial point cloud state data sample; the third point cloud image sample descriptor belongs to the point cloud image sample descriptor; performing point cloud coding recognition on the first initial laser sensor signal sample through the first original genetic iterative branch to obtain a fourth initial point cloud state data sample of the first initial laser sensor signal sample, performing an image descriptor mining operation on the fourth initial point cloud state data sample to obtain a fourth point cloud image sample descriptor corresponding to the fourth initial point cloud state data sample; the fourth point cloud image sample descriptor belongs to the point cloud image sample descriptor; based on the third point cloud image sample descriptor and the fourth point cloud image sample descriptor, determining the training error data corresponding to the adversarial training item.

[0109] Example 4: The Z debugging projects include a delayed signal decoding project; the first a priori debugging sample is processed through the first original genetic iterative branch to obtain a point cloud image sample descriptor associated with each debugging project, and Z training error data corresponding to the Z debugging projects are determined based on the point cloud image sample descriptor, including: performing sensor signal preprocessing on the first a priori debugging sample to obtain a first initial laser sensor signal sample of the first a priori debugging sample; performing point cloud coding recognition on the first initial laser sensor signal sample through the first original genetic iterative branch to obtain a fourth initial point cloud state data sample of the first initial laser sensor signal sample, and performing an image descriptor mining operation on the fourth initial point cloud state data sample. A fourth point cloud image sample descriptor corresponding to the fourth initial point cloud state data sample is obtained; the fourth point cloud image sample descriptor belongs to the point cloud image sample descriptor; according to the decoding module related to the delayed signal decoding project, a deviation state decision is performed on the fourth point cloud image sample descriptor to obtain a deviation state decision feature corresponding to the first priori debugging sample; a deep convolutional network related to the first original genetic iterative branch is obtained, and a deviation state decision is performed on the first priori debugging sample according to the deep convolutional network to obtain a deviation correction discrimination training feature corresponding to the first priori debugging sample; based on the deviation state decision feature and the deviation correction discrimination training feature, the training error data corresponding to the delayed signal decoding project is determined.

[0110] 1. Example 1: Signal-noise decoding project

[0111] 1) Signal processing flow

[0112] In the signal-noise decoding project, the sensor signal of the first a priori debugging sample is first preprocessed using the first original genetic iteration branch. This process is similar to the sensor signal preprocessing of target laser sensor detection information mentioned previously. For example, during the sensor signal preprocessing process, the original signal is adjusted based on relevant parameters in the target deviation analysis algorithm (such as certain characteristic parameters of the laser sensor). After this processing, the first initial laser sensor signal sample of the first a priori debugging sample is obtained. This sample contains signal decoding result samples under multiple distribution labels, which may be related to different signal characteristics or sources. From the multiple distribution labels corresponding to the first initial laser sensor signal sample, the first distribution label to be subjected to signal noise addition is arbitrarily determined. For example, if there are three distribution labels (A), (B), and (C), (A) is determined to be the first distribution label. Then, based on the signal noise addition information, the signal decoding result sample under (A) in the first initial laser sensor signal sample is subjected to signal noise addition. For example, the signal noise addition information specifies that the noise addition amplitude is (10%) of the original signal amplitude (the amplitude here can be the voltage amplitude of the signal or other values ​​representing the signal strength), and random noise is added to perform noise addition to obtain the first noisy sample;

[0113] 2) Point cloud data generation and image descriptor mining

[0114] Next, the first noisy sample is subjected to point cloud coding recognition through the first original genetic iterative branch. In this process, according to the point cloud coding recognition principle mentioned above, the noisy signal is converted into point cloud data to obtain the first initial point cloud state data sample of the first noisy sample. Then, an image descriptor mining operation is performed on the first initial point cloud state data sample to obtain the first point cloud image sample descriptor corresponding to the first initial point cloud state data sample. In the process of mining image descriptors, for example, for the size descriptor, the door panel-related dimensions are determined according to the method of calculating the extreme value difference of the point cloud data coordinates mentioned above; for the contour shape descriptor, a curve fitting algorithm is used to determine the contour shape; for the surface state descriptor, the (z) coordinate variance is calculated to determine the surface state;

[0115] 3) Noise Decoding Operation and Training Error Data Determination

[0116] A decoding module associated with the signal-noise decoding project performs a noise decoding operation on the first point cloud image sample descriptor to obtain noisy signal decoding prediction information corresponding to the signal-noise information in the first noisy sample. For example, this decoding module may be based on a pre-trained model or algorithm, and it predicts relevant features of the noisy signal based on information in the first point cloud image sample descriptor to obtain the noisy signal decoding prediction information.

[0117] Finally, the training error data corresponding to the signal-noise decoding project is determined based on the signal decoding result sample under the first distribution label in the first initial laser sensor signal sample and the noisy signal decoding prediction information. For example, if a key parameter value in the signal decoding result sample is (x_0) and the corresponding parameter value in the noisy signal decoding prediction information is (x_1), the error formula (e = vertx_1 - x_0vert) can be used to calculate the training error data (e). This error data reflects the accuracy of the decoding under the condition of signal noise.

[0118] 2. Example 2: Signal Alternation Decoding Project

[0119] 1) Initial signal processing and signal iteration

[0120] Similar to Example 1, the sensor signal of the first prior debugging sample is first preprocessed through the first original genetic iterative branch to obtain a first initial laser sensor signal sample, which contains signal decoding result samples under multiple distribution labels. From these distribution labels, arbitrarily determine the second distribution label to be subjected to signal iteration. For example, (B) is determined as the second distribution label among multiple distribution labels. Signal iteration processing is performed on the signal decoding result sample under (B) in the first initial laser sensor signal sample based on the signal iteration information. The signal iteration information here is different from the original signal decoding result sample under (B). For example, the signal iteration information is to change a certain frequency component of the original signal. For example, the frequency of the original signal is (f_0=100Hz), and the signal iteration information changes it to (f_1=120Hz), thereby obtaining the second noise-added sample;

[0121] 2) Subsequent processing flow

[0122] Then, the second noisy sample is subjected to point cloud coding recognition through the first original genetic iterative branch to obtain the second initial point cloud state data sample of the second noisy sample. Then, the second initial point cloud state data sample is subjected to image descriptor mining operation to obtain the second point cloud image sample descriptor corresponding to the second initial point cloud state data sample. In the image descriptor mining process, the size descriptor, contour shape descriptor and surface state descriptor, etc. are determined in the same manner according to the established method. Through the decoding module related to the signal change decoding project, the second point cloud image sample descriptor is subjected to signal change decoding operation to obtain signal change decoding information corresponding to multiple distribution labels. For example, this decoding module may calculate the relevant information of each distribution label after the signal change based on certain features in the second point cloud image sample descriptor and the pre-set signal change rules to obtain signal change decoding information;

[0123] 3) Determination of training error data

[0124] Based on the second distribution label and the signal iteration decoding information corresponding to the multiple distribution labels, training error data corresponding to the signal iteration decoding project is determined. For example, for the second distribution label (B), if the original eigenvalue is (y_0), and the corresponding eigenvalue calculated based on the signal iteration decoding information is (y_1), a similar error formula (e' = verty_1 - y_0vert) can be used to calculate the training error data (e'). This error data reflects the accuracy of decoding under signal iteration.

[0125] Example 3: Adversarial Training Project

[0126] 1) Adversarial Processing and Signal Generation

[0127] In the adversarial training project, adversarial processing is first performed on the sensor detection input reflection signal of the first prior debugging example to obtain an adversarial debugging example. This adversarial processing can introduce an adversarial sample generation mechanism, such as through the relevant algorithms of the Generative Adversarial Network (GAN), to make the adversarial debugging example adversarial to the original first prior debugging example. The sensor signal of the adversarial debugging example is preprocessed to obtain the laser sensor adversarial signal of the adversarial debugging example. At the same time, the sensor signal of the first prior debugging example is preprocessed to obtain the first initial laser sensor signal sample of the first prior debugging example.

[0128] 2) Point cloud data and image descriptor mining

[0129] The laser sensor countermeasure signal is subjected to point cloud coding recognition through the first original genetic iterative branch to obtain a third initial point cloud state data sample of the laser sensor countermeasure signal, and then an image descriptor mining operation is performed on the third initial point cloud state data sample to obtain a third point cloud image sample descriptor corresponding to the third initial point cloud state data sample. Similarly, the first initial laser sensor signal sample is subjected to point cloud coding recognition through the first original genetic iterative branch to obtain a fourth initial point cloud state data sample of the first initial laser sensor signal sample, and then an image descriptor mining operation is performed on the fourth initial point cloud state data sample to obtain a fourth point cloud image sample descriptor corresponding to the fourth initial point cloud state data sample. During the mining process, various descriptor information is accurately obtained according to the established method;

[0130] 3) Determination of training error data

[0131] Based on the third and fourth point cloud image sample descriptors, training error data corresponding to the adversarial training project is determined. For example, if a comprehensive eigenvalue in the third point cloud image sample descriptor is (z_0) (this comprehensive eigenvalue can be calculated by combining multiple descriptors such as size, shape, and surface condition, for example, through weighted summation), and the corresponding comprehensive eigenvalue in the fourth point cloud image sample descriptor is (z_1), the error formula (e'' = sqrt{(z_1 - z_0)^2}) can be used to calculate the training error data (e''). This error data reflects the degree of difference between the two in the adversarial training scenario.

[0132] Example 4: Delayed Signal Decoding Project

[0133] 1) Initial processing and image descriptor mining

[0134] First, the sensor signal of the first a priori debugging sample is preprocessed to obtain the first initial laser sensor signal sample of the first a priori debugging sample. Then, point cloud coding recognition is performed on the first initial laser sensor signal sample through the first original genetic iteration branch to obtain a fourth initial point cloud state data sample of the first initial laser sensor signal sample. An image descriptor mining operation is performed on the fourth initial point cloud state data sample to obtain a fourth point cloud image sample descriptor corresponding to the fourth initial point cloud state data sample. In the image descriptor mining process, various descriptor information is accurately obtained according to standard methods;

[0135] 2) Deviation state decision and training error data determination

[0136] Based on the decoding module associated with the delayed signal decoding project, a deviation state decision is performed on the fourth point cloud image sample descriptor to obtain a deviation state decision feature corresponding to the first prior debugging sample. For example, this decoding module may determine whether a door panel has deviation and the approximate type of deviation based on certain features in the fourth point cloud image sample descriptor to obtain a deviation state decision feature.

[0137] At the same time, a deep convolutional network associated with the first original genetic iteration branch is obtained, and a deviation state decision is made for the first prior debugging example based on the deep convolutional network, thereby obtaining a deviation-correction discriminant training feature corresponding to the first prior debugging example. This deep convolutional network may be pre-trained and can accurately determine the deviation state based on the input prior debugging example and output the deviation-correction discriminant training feature.

[0138] Finally, the training error data corresponding to the delayed signal decoding project is determined based on the deviation state decision feature and the correction discrimination training feature. For example, if a key indicator value in the deviation state decision feature is (a_0) and the corresponding key indicator value in the correction discrimination training feature is (a_1), the error formula (e''' = verta_1 - a_0vert) can be used to calculate the training error data (e'''). This error data reflects the degree of difference between the two in the delayed signal decoding case.

[0139] The above embodiment performs comprehensive and targeted training on the first original genetic iteration branch by setting up different sub-projects within the (Z) debugging projects, such as a signal noise decoding project, a signal iteration decoding project, an adversarial training project, and a delayed signal decoding project. Through different processing methods (such as noise addition, iteration, adversarial training, delay, etc.) and corresponding error calculation methods, the impact of various practical situations on the genetic iteration branch can be fully considered. This helps to improve the robustness and accuracy of the genetic iteration branch, enabling it to more accurately perform image descriptor mining operations when processing different initial point cloud state data, thereby providing a more reliable data foundation for subsequent door panel processing-related operations, reducing processing errors caused by inaccurate data processing, and improving the quality and efficiency of door panel processing.

[0140] Under an optional design idea, the target point cloud image descriptor is obtained by performing an image descriptor mining operation on the initial point cloud state data by the point cloud description mining module in the first genetic iterative branch; the first genetic iterative branch belongs to a target deviation analysis algorithm for performing deviation analysis on the target door panel; the target deviation analysis algorithm also includes a second genetic iterative branch after the point cloud description mining module; the correction discrimination operation is performed on the target point cloud image descriptor according to each cutting processing task event to obtain the correction discrimination recommendation feature corresponding to each cutting processing task event, including: determining the second genetic iterative branch from the target deviation analysis algorithm; the second genetic iterative branch includes the spatial attention model corresponding to each cutting processing task event; through the spatial attention model corresponding to each cutting processing task event, the target point cloud image descriptor is subjected to a correction discrimination operation to obtain the correction discrimination recommendation feature corresponding to each cutting processing task event; a spatial attention model is used to determine the correction discrimination recommendation feature corresponding to a cutting processing task event.

[0141] In addition, the method further includes: obtaining a second algorithm debugging sample; the second a priori debugging sample included in the second algorithm debugging sample is set with a task annotation; the task annotation includes X task classification feature values ​​corresponding to the X cutting task events, one cutting task event corresponds to one task classification feature value, and the X task classification feature values ​​are determined based on the cutting task event to which the second a priori debugging sample belongs; obtaining a first genetic iteration branch that has completed debugging, performing point cloud recognition on the second a priori debugging sample through the first genetic iteration branch to obtain a fifth initial point cloud state data sample of the second a priori debugging sample, performing an image descriptor mining operation on the fifth initial point cloud state data sample, and obtaining a fifth point cloud image sample descriptor corresponding to the fifth initial point cloud state data sample. Obtaining X original spatial attention models corresponding to the X cutting task events, and performing task matching analysis on the fifth point cloud image sample descriptors based on the X models to obtain X task matching analysis viewpoints corresponding to the X original spatial attention models; one cutting task event corresponds to one original spatial attention model, and one original spatial attention model is used to determine one task matching analysis viewpoint; determining second target training error data based on the X task matching analysis viewpoints and the X task classification feature values, improving algorithm configuration weights of the X original spatial attention models based on the second target training error data, determining the improved X original spatial attention models as the X spatial attention models, and determining the second genetic iteration branch based on the X spatial attention models.

[0142] In the above embodiment, the target point cloud image descriptor is obtained by mining the initial point cloud state data using the point cloud description mining module in the first genetic iteration branch. In this process, the first genetic iteration branch is part of the target deviation analysis algorithm, providing an accurate point cloud image descriptor for subsequent door panel deviation analysis.

[0143] For example, in mining size descriptors, for the initial point cloud state data, the size of the door panel is determined by calculating the difference between the maximum and minimum coordinate values ​​of the point cloud in each coordinate axis direction. For example, if the point cloud data has the maximum coordinate value (x_{max}=1.2m) and the minimum coordinate value (x_{min}=0m) in the (x) direction, then the size of the door panel in the (x) direction is (L_x=x_{max}-x_{min}=1.2m). For contour shape descriptors, a curve fitting algorithm may be used, such as fitting a quadratic curve (y=ax^{2}+bx+c) to the point cloud data of the door panel edge, and calculating the coefficients (a), (b), and (c) using the least squares method. For surface state descriptors, the variance (sigma_{z}^{2}) of the (z) coordinate in the point cloud data is calculated to represent the surface flatness. For example, there are (n) points in the point cloud data, and the (z) coordinates are (z_1, z_2, cdots, z_n). According to the variance formula:

[0144] (sigma_{z}^{2}=frac{1}{n}sum_{i=1}^{n}(z_i-overline{z})^2), where (overline{z}=frac{1}{n}sum_{i=1}^{n}z_i).

[0145] Next, a second genetic iteration branch is determined from the target deviation analysis algorithm, in which each cutting task event corresponds to a spatial attention model. These spatial attention models play a key role in the deviation correction discriminant operation.

[0146] For example, in the case of a door panel edge cutting task, the corresponding spatial attention model focuses on the parts of the target point cloud image descriptor related to edge cutting. For example, if the contour shape descriptor in the target point cloud image descriptor indicates an outward bulge at the door panel edge, the spatial attention model analyzes this deformation based on predefined rules and algorithms.

[0147] For the spatial attention model corresponding to each cutting task event, the way it performs correction and discrimination operations on the target point cloud image descriptor is as follows. For example, for a certain cutting task event, the spatial attention model constructs a feature vector (mathbf{V}=[v_1, v_2, cdots, v_m]) based on information such as the door panel's size descriptor, contour shape descriptor, and surface state descriptor. This feature vector contains all the point cloud image descriptor information related to the task event. Then, the spatial attention model uses a discriminant function (f(mathbf{V})) to determine the correction discrimination recommendation feature. For example, in the door panel edge cutting task, if the discriminant function determines the depth (h) that the door panel edge needs to be additionally cut based on the feature vector, this (h) is part of the correction discrimination recommendation feature.

[0148] Furthermore, the training steps of the second genetic iteration branch are as follows.

[0149] Obtain the second algorithm debugging sample and related preparations: Obtain the second algorithm debugging sample, in which the second prior debugging sample sets task annotations. These task annotations include (X) task classification feature values ​​corresponding to (X) cutting processing task events. For example, (X=3), the three cutting processing task events are door panel edge cutting, surface milling, and hole processing, and the corresponding task classification feature values ​​are (C_1), (C_2), and (C_3). These task classification feature values ​​are determined based on the cutting processing task event to which the second prior debugging sample belongs. For example, for the prior debugging sample of the door panel edge cutting task event, its task classification feature value (C_1) may contain some preset parameters or indicators related to edge cutting.

[0150] Use the first genetic iteration branch to process the a priori debugging example: Obtain the debugged first genetic iteration branch and use it to perform point cloud recognition on the second a priori debugging example to obtain the fifth initial point cloud state data sample of the second a priori debugging example. Then perform image descriptor mining on this sample to obtain the fifth point cloud image sample descriptor corresponding to the fifth initial point cloud state data sample. This process is similar to the process of mining the target point cloud image descriptor by the first genetic iteration branch.

[0151] Task Matching Analysis and Training Error Data Determination: Obtain (X) original spatial attention models corresponding to (X) cutting task events. Based on these models, perform task matching analysis on the fifth point cloud image sample descriptor, obtaining (X) task matching analysis opinions corresponding to the (X) original spatial attention models. For example, for the original spatial attention model corresponding to the door panel edge cutting task, it performs task matching analysis based on the information in the fifth point cloud image sample descriptor, obtaining a task matching analysis opinion (P_1). This opinion may include information such as an assessment of the degree of match between the door panel edge cutting task and the fifth point cloud image sample descriptor, as well as any potential deviations.

[0152] The second target training error data is determined based on (X) task matching analysis viewpoints and (X) task classification feature values. For example, for a cutting task event, a key indicator value (k_1) in the task matching analysis viewpoint corresponds to a task classification feature value (C_1). The error value corresponding to this task event can be calculated using the error formula (e=vertk_1-C_1vert). For all (X) task events, these error values ​​are weighted and summed (for example, each task event has a weight (w_i)) to obtain the second target training error data (E=sum_{i=1}^{X}w_ie_i).

[0153] Improve the spatial attention model to determine the second genetic iteration branch: Based on the second target training error data, improve the algorithm configuration weights of the (X) original spatial attention models. For example, for a parameter (alpha) in an original spatial attention model, its improvement formula can be (alpha_{new}=alpha-etaE) (where (eta) is the learning rate). Through this improvement, the (X) original spatial attention models are improved to determine (X) spatial attention models, and the second genetic iteration branch is determined based on these spatial attention models.

[0154] Therefore, firstly, by mining the target point cloud image descriptor through the first genetic iteration branch, the size, contour shape and surface state of the door panel can be fully and accurately obtained, providing a rich data basis for subsequent correction and discrimination operations. Secondly, the spatial attention model in the second genetic iteration branch performs correction and discrimination operations for each cutting task event, making the correction and discrimination more targeted and accurate. Furthermore, in the training process of the second genetic iteration branch, by utilizing the second algorithm debugging samples, task annotations and task matching analysis, the algorithm configuration weights of the spatial attention model can be optimized, and the accuracy and effectiveness of the spatial attention model can be improved, thereby further improving the correction and discrimination capabilities of the entire door panel processing process, reducing processing errors, and improving processing quality and efficiency.

[0155] It can be understood that, based on the above content, in an extended embodiment, the initial point cloud state data is obtained by performing point cloud coding recognition on the initial laser sensor signal by the coding module in the first genetic iterative branch; the initial laser sensor signal is obtained by performing sensor signal preprocessing on the target laser sensor detection information by the excitation module in the first genetic iterative branch; the first genetic iterative branch belongs to a target deviation analysis algorithm for performing deviation analysis on the target door panel; the target deviation analysis algorithm also includes a third genetic iterative branch after the coding module; the processing parameter mining operation is performed on the initial point cloud state data to obtain the processing parameter characteristics corresponding to each cutting processing task event, including: determining from the target deviation analysis algorithm A third genetic iterative branch is defined; a processing parameter mining operation is performed on the initial point cloud state data through the third genetic iterative branch to obtain the processing parameter features corresponding to each cutting processing task event; the third genetic iterative branch includes a backbone module and a fully connected module; the processing parameter mining operation is performed on the initial point cloud state data through the third genetic iterative branch to obtain the processing parameter features corresponding to each cutting processing task event, including: performing a linear image descriptor mining operation on the initial point cloud state data through the backbone module to obtain a linear point cloud image descriptor corresponding to the initial point cloud state data; processing the linear point cloud image descriptor through the fully connected module to obtain the processing parameter features corresponding to each cutting processing task event.

[0156] In the above-described embodiment, the initial point cloud state data is obtained through a series of processes. First, the excitation module in the first genetic iteration branch preprocesses the target laser sensor detection information to obtain an initial laser sensor signal. During this process, the excitation module operates according to the relevant rules of the target deviation analysis algorithm. For example, for signals of different frequency bands or signals of different intensity regions in the target laser sensor detection information, the excitation module adjusts them according to pre-set algorithm parameters. For example, if the signal intensity of a certain frequency band (f_1) in the target laser sensor detection information is (I_1), the excitation module processes the signal intensity of this frequency band according to the parameter (a_1) in the algorithm (for example, (a_1) is a coefficient related to signal amplification or attenuation), changing it to (I_1'=a_1I_1). After processing the entire target laser sensor detection information, the initial laser sensor signal is obtained.

[0157] Next, the encoding module in the first genetic iteration branch performs point cloud coding recognition on the initial laser sensor signal, thereby obtaining the initial point cloud state data. The point cloud coding recognition process converts the information in the initial laser sensor signal into point cloud data. For example, the initial laser sensor signal contains a series of discrete signal values ​​(s_1, s_2, cdots, s_n). The encoding module converts these signal values ​​into point coordinates ((x_i, y_i, z_i)) in the point cloud data according to certain coding rules, such as the mapping relationship between signal values ​​and spatial coordinates (for example, a mapping function (M) such that the spatial coordinates ((x_i, y_i, z_i) = M(s_i))). These point coordinates constitute the initial point cloud state data.

[0158] In addition, the structure and function of the third genetic iteration branch are introduced as follows.

[0159] The target deviation analysis algorithm also includes a third genetic iteration branch after the encoding module. The third genetic iteration branch consists of a backbone module and a fully connected module, which plays a key role in the machining parameter mining operation.

[0160] Linear Image Descriptor Mining Operation - Backbone Module: The backbone module performs linear image descriptor mining on the initial point cloud state data to obtain a linear point cloud image descriptor corresponding to the initial point cloud state data. This linear image descriptor mining operation may involve extracting certain linear features from the initial point cloud state data. For example, for the coordinate information in the point cloud data, the linear distance relationship between adjacent points can be calculated. For example, if there are two adjacent points in the point cloud data (P_1(x_1, y_1, z_1)) and (P_2(x_2, y_2, z_2)), the linear distance between them is d = sqrt{(x_2-x_1)^2+(y_2-y_1)^2+(z_2-z_1)^2}). By calculating this linear distance relationship for all adjacent points or points within a specific area, a portion of the linear point cloud image descriptor can be constructed. In addition, we can also calculate the linear distribution characteristics of point cloud data in a certain direction, such as the distribution density of points along the (x) axis (rho_x). For example, if there are (m) points in a specific region (R), and the length of the region (R) along the (x) axis is (L_x), then (rho_x = frac{m}{L_x}). These linear features together constitute the linear point cloud image descriptor.

[0161] Machining Parameter Feature Determination - Fully Connected Module: The fully connected module processes the linear point cloud image descriptor to obtain machining parameter features corresponding to each cutting task event. For example, for the cutting task event of a door panel edge, the fully connected module determines the tool's cutting depth (h) based on the information in the linear point cloud image descriptor. For example, the fully connected module contains a mapping function (F) that maps the feature vectors in the linear point cloud image descriptor (mathbf{V}=[v_1, v_2, cdots, v_n]) (where (vi) are the previously calculated eigenvalues ​​of the linear point cloud image descriptor, such as (d) and (rho_x)) to the cutting depth (h), i.e., (h=F(mathbf{V})). This mapping function (F) can be a complex function trained based on extensive experimental data or empirical formulas. For the tool's rotational speed (n), another mapping function (G) may exist, such that (n=G(mathbf{V})). Similarly, for other cutting processing task events, such as surface milling and hole processing, the fully connected module will also determine the corresponding processing parameter characteristics based on the linear point cloud image descriptor, such as the feed speed of the milling tool and the drilling speed of hole processing.

[0162] A specific example and algorithmic details of machining parameter mining operations: For example, consider a door panel edge cutting task. For example, there's an empirical relationship between the linear distance feature (d) in the linear point cloud image descriptor and the cutting depth (h). This relationship is based on the empirical formula (h = k_1d + k_2), where (k_1) and (k_2) are coefficients determined based on factors such as the door panel material and tool type. For example, for a certain wooden door panel material and a specific tool, (k_1 = 0.1) and (k_2 = 0.02) are used. If the linear distance between adjacent points (d = 0.3m) is calculated, the cutting depth is calculated using the formula (h = 0.1 times 0.3 + 0.02 = 0.05m).

[0163] The tool speed (n), for example, is related to the point cloud's point distribution density (rho_x) along the (x) axis, according to an empirical formula (n = k_3rho_x + k_4). If (rho_x = 5) points / meter (here, the number of points per meter), (k_3 = 100), and (k_4 = 3000), then the tool speed is (n = 100 times 5 + 3000 = 3500 rpm).

[0164] In surface milling tasks, for example, the feed rate (v) of the milling tool is related to multiple features in the linear point cloud image descriptor. A multivariate linear regression model can be constructed (v = k5v1 + k6v2 + cdots + kmvm + km+1), where (v1, v2, cdots, vm) are the different eigenvalues ​​in the linear point cloud image descriptor and (k5, k6, cdots, km, km+1) are the corresponding coefficients. By determining these coefficients through regression analysis of a large amount of experimental data, the milling tool feed rate can be accurately calculated based on the linear point cloud image descriptor.

[0165] The drilling speed (s) is similarly determined in hole machining tasks. For example, based on specific features in the linear point cloud image descriptor (such as the point cloud distribution uniformity index (u) in a certain direction), the formula (s = k_{n+1}u + k_{n+2}) is used. By determining the coefficients (k_{n+1}) and (k_{n+2}), the drilling speed is calculated based on the value (u).

[0166] In this way, first, through the excitation module and encoding module in the first genetic iterative branch, the target laser sensor detection information is gradually converted into initial point cloud state data. This processing method makes the data conversion process more accurate and can effectively extract information on the door panel surface. Secondly, the setting of the third genetic iterative branch, especially the collaborative work of the backbone module and the fully connected module, plays an important role in the processing parameter mining operation. The linear point cloud image descriptor mined by the backbone module provides an effective input for the fully connected module, enabling the fully connected module to accurately determine the processing parameter characteristics corresponding to each cutting task event based on the actual state of the door panel. This method of determining processing parameters based on point cloud data mining is more accurate than traditional empirical estimation, and can improve the accuracy of door panel cutting, reduce processing errors, and improve processing quality. It also helps to improve processing efficiency and reduce production costs.

[0167] In some other scalable technical solutions, the X cutting task events include a target cutting task event; the method further includes: obtaining a target deviation analysis algorithm for performing deviation analysis on a target door panel, and obtaining an original lightweight algorithm for collaborative debugging based on the target deviation analysis algorithm; obtaining a fourth algorithm debugging sample related to the target cutting task event; the fourth algorithm debugging sample includes a fourth a priori debugging sample; processing the fourth a priori debugging sample through the target deviation analysis algorithm to obtain an automatic deviation correction execution strategy corresponding to the fourth a priori debugging sample, and converting the fourth a priori debugging sample into an automatic deviation correction execution strategy corresponding to the fourth a priori debugging sample. The automatic correction execution strategy corresponding to the debugging sample is determined as the perturbation training annotation used to debug the original lightweight algorithm; the fourth prior debugging sample is processed by the original lightweight algorithm to obtain the collaborative correction execution strategy corresponding to the fourth prior debugging sample; the algorithm weights of the original lightweight algorithm are improved according to the collaborative correction execution strategy and the perturbation training annotation, and the improved original lightweight algorithm is determined as the target lightweight algorithm of the target deviation analysis algorithm; the target lightweight algorithm is used to perform correction analysis processing on the target laser sensor detection information under the target cutting processing task event.

[0168] In this embodiment, a target deviation analysis algorithm for performing deviation analysis on a target door panel is first obtained, along with an original lightweight algorithm that is collaboratively debugged based on this target deviation analysis algorithm. The target deviation analysis algorithm, already discussed in previous embodiments, is a key algorithm for analyzing the various states of the target door panel to achieve accurate machining. The original lightweight algorithm is a relatively streamlined and efficient algorithm designed to achieve similar functionality at a lower computational cost.

[0169] Next, a fourth algorithm debugging sample related to the target cutting task event is obtained, including a fourth a priori debugging sample. This fourth a priori debugging sample is an example of the state data of the target door panel under specific conditions. For example, for a door panel edge cutting task event (e.g., the target cutting task event), the fourth a priori debugging sample can be an example of door panel point cloud data with specific size, shape deviation, and surface condition.

[0170] The fourth a priori debugging example is then processed using the target deviation analysis algorithm to determine the corresponding automatic deviation correction execution strategy. This process involves various modules in the target deviation analysis algorithm, such as the previously mentioned genetic iteration branch, which analyzes the point cloud data of the a priori debugging example, determines the deviation of the door panel, and then formulates an automatic deviation correction execution strategy based on the requirements of the processing task event.

[0171] For example, in a door panel edge cutting task, if the panel edge has an outward bulge, the target deviation analysis algorithm uses the panel's size descriptor (e.g., if the panel width is wider than the standard width (0.1m)), contour shape descriptor (information such as the curvature of the bulge), and surface condition descriptor (surface flatness) to develop an automatic correction strategy. This strategy may include parameters such as cutting depth, starting point, and ending point. This automatic correction strategy is used as a perturbation training annotation for debugging the original lightweight algorithm.

[0172] Next, the fourth prior debugging example is processed by the original lightweight algorithm to obtain the collaborative correction execution strategy corresponding to the fourth prior debugging example. Although the original lightweight algorithm is relatively simple, it can also analyze the prior debugging example according to its own algorithm logic and generate a correction execution strategy. For example, the original lightweight algorithm may be based on some simplified size and shape analysis rules. For the same door panel edge cutting task event, it may generate a collaborative correction execution strategy based on the approximate width of the door panel and the exemplary shape characteristics of the edge. The parameters such as cutting depth and cutting path in this strategy may be different from the automatic correction execution strategy generated by the target deviation analysis algorithm.

[0173] The algorithm weights of the original lightweight algorithm are improved based on the collaborative error correction execution strategy and perturbation training annotations. For example, if the cutting depth in the collaborative error correction execution strategy is (h_1) and the cutting depth in the perturbation training annotation is (h_2), the error formula (e=verth_1-h_2vert) can be used to calculate the difference in cutting depth between the two parameters. Similar methods can be used to calculate the difference in other parameters, such as the coordinates of the starting and ending points of the cutting path. These differences are combined, for example through a weighted summation (for example, the weights of each parameter are (w_1, w_2, cdots)), to obtain a total error value (E=sum_{i}w_ie_i).

[0174] Based on this total error value (E), the algorithm weights of the original lightweight algorithm are improved. For example, for a key weight (alpha) in the original lightweight algorithm, the improved formula can be (alpha_{new}=alpha-etaE), where (eta) is the learning rate. Based on this formula, the weights in the original lightweight algorithm are adjusted, and the improved original lightweight algorithm is determined as the target lightweight algorithm of the target deviation analysis algorithm. This target lightweight algorithm can be used to perform deviation correction analysis on target laser sensor detection information under target cutting task events. Due to the coordinated debugging with the target deviation analysis algorithm, it can perform more accurate deviation correction analysis while maintaining the lightweight characteristics.

[0175] It can be understood that when the target deviation analysis algorithm processes the fourth prior debugging example, for example, the point cloud data of the fourth prior debugging example is represented as (P = {p_1, p_2, cdots, p_n}), where (p_i = (x_i, y_i, z_i)) represents a point in the point cloud data. The size descriptor mining module in the target deviation analysis algorithm calculates the length (L_x), width (L_y), and thickness (h) of the door panel, such as (L_x = max(x_i) - min(x_i)), (L_y = max(y_i) - min(y_i)), and (h = frac{1}{n}sum_{i=1}^{n}z_i). The contour shape descriptor mining module fits the point cloud data of the door panel edge using a curve fitting algorithm, obtaining the edge curve equation (y = ax^{2} + bx + c). The shape deviation is determined based on the curve parameters (a), (b), and (c) and the difference between the curve and the standard shape.

[0176] Based on this descriptor information, the target deviation analysis algorithm develops an automatic correction strategy. For example, the cutting depth (h_d) is determined based on the door panel thickness (h) and the degree of shape deviation (δs) (δs can be quantified by comparing it with the standard shape). The formula (h_d = k_1h + k_2δs) is used, where (k_1) and (k_2) are coefficients determined by factors such as the door panel material and tool type. For example, if (h = 0.05m), (δs = 0.02), (k_1 = 0.2), and (k_2 = 0.1), then (h_d = 0.2 times 0.05 + 0.1 times 0.02 = 0.012m).

[0177] The original lightweight algorithm might adopt a more adaptive approach when handling the fourth prior debugging example. For example, it might determine the cutting depth (h_{d1}) based solely on the door panel's width (L_y) and a preset deviation coefficient (k_d), where h_{d1} = k_dL_y. For example, if L_y = 1.2m and k_d = 0.01, then h_{d1} = 0.01 times 1.2 = 0.012m. In this case, while the cutting depths are the same for both methods, they may differ in other situations. Through the collaborative debugging process described above, the algorithm weights of the original lightweight algorithm can be continuously improved, bringing it closer to the accuracy of the target deviation analysis algorithm.

[0178] Applying the above embodiment, first, the target deviation analysis algorithm is used to provide perturbation training annotations for the original lightweight algorithm, so that the debugging of the lightweight algorithm has an accurate reference standard. Secondly, the collaborative debugging process can utilize the accuracy of the target deviation analysis algorithm and the high efficiency of the original lightweight algorithm to improve the accuracy of its correction analysis while maintaining the computational efficiency of the lightweight algorithm. For example, through specific algorithm weight improvements, the lightweight algorithm is made more accurate in determining key parameters such as cutting depth and cutting path. This helps in practical applications, especially in scenarios where computing resources are limited but accurate correction analysis is required, to quickly and accurately perform correction analysis on the target door panel, thereby improving processing quality and efficiency, while also reducing the cost of algorithm development and application.

[0179] In some further expandable technical solutions, the automatic correction execution strategy of the target door panel is determined based on the correction discrimination recommendation features corresponding to each cutting processing task event and the processing parameter features corresponding to each cutting processing task event, including: adjusting the correction discrimination recommendation features corresponding to each cutting processing task event through the processing parameter features corresponding to each cutting processing task event to obtain the correction discrimination adjustment features corresponding to each cutting processing task event; a feature enhancement factor corresponding to a cutting processing task event is used to adjust the correction discrimination recommendation features corresponding to the corresponding cutting processing task event; and determining the automatic correction execution strategy of the target door panel based on the correction discrimination adjustment features corresponding to each cutting processing task event.

[0180] In the above embodiment, the adjustment process of the correction discrimination suggestion feature is as follows: when determining the automatic correction execution strategy for the target door panel based on the correction discrimination suggestion feature and the processing parameter feature corresponding to each cutting task event, the correction discrimination suggestion feature must first be adjusted. Here, the corresponding correction discrimination suggestion feature is adjusted based on the processing parameter feature corresponding to each cutting task event to obtain the correction discrimination adjustment feature corresponding to each cutting task event.

[0181] For each cutting task event, a feature enhancement factor is used to adjust the correction discrimination feature recommendations. For example, in the door panel edge cutting task event, the correction discrimination recommendation features include the depth of the additional cut required (h_{adv}), the adjustment amount (δx_{start}) and (δy_{start}) of the cutting starting point coordinates ((x_{start}, y_{start})), etc.

[0182] For example, the tool cutting capability-related parameters in the machining parameter features corresponding to this cutting task event are (C_{tool}) (e.g., the tool's maximum cutting depth (h_{max}), cutting accuracy (p_{tool}), etc.), and there is a feature enhancement factor (k) corresponding to this task event. Regarding the adjustment of the depth (h_{adv}) requiring additional cutting, if the original value of (h_{adv}) is (0.03m), based on the tool's maximum cutting depth in the machining parameter features (h_{max} = 0.05m), and the feature enhancement factor (k = 0.8) (where the value of (k) is determined based on factors such as the tool's compatibility with the door panel material and machining accuracy requirements), the adjusted cutting depth is (h_{adj} = k times sh_{adv} = 0.8 times 0.03 = 0.024m).

[0183] Regarding the adjustment of the cutting starting point coordinate adjustment values ​​(δx_{start}) and (δy_{start}), for example, in the machining parameter feature for cutting accuracy (p_{tool} = 0.001m), if the original value of (δx_{start}) is (0.002m), the feature enhancement factor (k) may be adjusted to (0.5) based on the cutting accuracy requirements (indicating that the starting point coordinate adjustment value is halved while taking cutting accuracy into account). The adjusted value is (δx_{start}^{adj} = k times δx_{start} = 0.5 times 0.002 = 0.001m). In this way, the correction discrimination recommendation features in each cutting task event are comprehensively adjusted to obtain the correction discrimination adjustment features.

[0184] The following steps determine the automatic correction execution strategy: Based on the correction discrimination and adjustment features corresponding to each cutting task event, the automatic correction execution strategy for the target door panel is determined. The entire door panel processing process may include multiple cutting tasks, such as edge cutting, surface milling, and hole processing.

[0185] Take the two cutting tasks of door panel edge cutting and surface milling as examples. For the door panel edge cutting task, the adjusted features for correction discrimination include the adjusted cutting depth (h_{adj}) and the coordinates of the cutting starting point ((x_{start}^{adj}, y_{start}^{adj})). For the surface milling task, for example, its correction discrimination features include the milling depth (h_{milling}^{adj}) and the feed rate of the milling tool (v_{feed}^{adj}).

[0186] When determining the automatic deflection correction execution strategy, the order and mutual influence of these task events must be comprehensively considered. For example, if a door panel's edge cutting is performed first, followed by surface milling, the edge cutting results will affect the initial state of the surface milling. For example, after edge cutting, the actual thickness of the door panel becomes (h_{actual}) (calculated based on the deflection correction discrimination and adjustment features of edge cutting). During surface milling, the milling depth (h_{milling}^{adj}) may need to be further adjusted based on (h_{actual}). For example, there is a formula for adjusting the milling depth based on the door panel thickness: (h_{milling}^{adj}=h_{milling}^{0}+k_{h}(h_{standard}-h_{actual})), where (h_{milling}^{0}) is the initial milling depth, (h_{standard}) is the standard door panel thickness, and (k_{h}) is a coefficient determined based on the door panel material and processing requirements.

[0187] For example, if (h_{milling}^{0}=0.02m), (h_{standard}=0.05m), (h_{actual}=0.04m), and (k_{h}=0.5), then (h_{milling}^{adj}=0.02+0.5times(0.05-0.04)=0.025m). By comprehensively considering the correction discrimination and adjustment features corresponding to each cutting task event, an automatic correction execution strategy for the target door panel is determined. This strategy includes the specific execution parameters for each task event, ensuring accurate correction operations throughout the door panel processing process and improving processing accuracy.

[0188] It's understandable that adjusting the deviation correction feature may involve the combined influence of multiple processing parameter features on a single deviation correction feature. For example, the adjustment amount of the hole machining center coordinates ((δx_{hole}, δy_{hole})) in the deviation correction feature for a door panel hole machining task is affected not only by the tool positioning accuracy (p_{loc}) (as a processing parameter feature), but also by other processing parameter features such as the hole machining order (if there are multiple hole machining tasks) and the required distance between adjacent holes (d_{hole-hole}).

[0189] For example, there is an adjustment formula that takes multiple factors into account: (δx_{hole}^{adj}=k_{1}δx_{hole}+k_{2}(p_{loc}-p_{required})+k_{3}(d_{hole-hole}-d_{standard})); where (k_{1}), (k_{2}), and (k_{3}) are coefficients determined based on processing requirements and experience, (p_{required}) is the expected tool positioning accuracy, and (d_{standard}) is the standard distance between adjacent holes. If (δx_{hole}=0.003m), (p_{loc}=0.002m), (p_{required}=0.001m), (d_{hole-hole}=0.1m), (d_{standard}=0.08m), (k_{1}=0.6), (k_{2}=0.2), (k_{3}=0.2), then (δx_{hole}^{adj}=0.6times0.003+0.2times(0.002-0.001)+0.2times(0.01)=0.0026m).

[0190] This adjustment method that comprehensively considers multiple processing parameter characteristics can adjust the correction judgment suggestion characteristics more comprehensively and accurately, thereby obtaining more reasonable correction judgment adjustment characteristics and further improving the accuracy of the automatic correction execution strategy.

[0191] It's worth noting that the synergistic relationship between different cutting tasks is crucial when determining the automatic correction execution strategy. In addition to the aforementioned sequential relationship, there are also some constraints between parallel tasks. For example, in the task of simultaneously cutting the edges of a door panel, although these are relatively independent cutting tasks, the cutting depths (h_{left}^{adj}) and (h_{right}^{adj}) on both sides must maintain a certain balance due to the overall structural stability requirements of the door panel.

[0192] For example, there's a balance equation: (verth_{left}^{adj} - h_{right}^{adj}vertleqk_{balance}h_{total}), where (h_{total}) is the total thickness of the door panel, and (k_{balance}) is a coefficient determined by the required structural strength. If (h_{total} = 0.05m) and (k_{balance} = 0.1), when the cutting depth on one side (h_{left}^{adj} = 0.02m), the cutting depth on the other side (h_{right}^{adj}) can range from (0.015m to h_{right}^{adj} ≤ 0.025m). This approach, by considering the synergistic relationships between different cutting tasks, determines the final automatic deviation correction strategy to ensure stability and accuracy throughout the entire door panel machining process.

[0193] With this design, first, the correction discrimination recommendation features are adjusted through the processing parameter features, taking into account multiple factors such as tools and processing accuracy, so that the correction discrimination features are more in line with the actual processing conditions and the accuracy of correction is improved. For example, the cutting depth is adjusted according to the cutting capacity of the tool to avoid unreasonable processing operations. Secondly, when determining the automatic correction execution strategy, the sequence, mutual influence and synergy between different cutting processing task events are comprehensively considered, which can effectively improve the stability and accuracy of the entire door panel processing process. Whether it is the order between task events that affects the milling depth, or the mutual constraint relationship between parallel tasks, both can be reasonably handled, thereby improving the door panel processing quality and reducing processing errors.

[0194] Further, Figure 2 FIG. 2 is a schematic diagram of a structure of an optimization system 200 provided in an embodiment of the present invention. Figure 2 The optimization system 200 shown includes a processor 210, which can call and run a computer program from a memory to implement the method in the embodiment of the present invention. Figure 2 As shown, the optimization system 200 may further include a memory 230. The processor 210 may call and run a computer program from the memory 230 to implement the method in the embodiment of the present invention. The memory 230 may be a separate device independent of the processor 210 or may be integrated into the processor 210. Optionally, as Figure 2As shown, optimization system 200 may also include a transceiver 220. Processor 210 may control transceiver 220 to interact with other devices. Specifically, it may send information or data to other devices, or receive information or data sent by other devices. Optionally, optimization system 200 may implement the corresponding processes of the storage engine, or components within the storage engine (such as a processing module), or devices equipped with the storage engine, as described in various methods of the embodiments of the present invention. For the sake of brevity, these processes are not further described here. It should be understood that the processor in the embodiments of the present invention may be an integrated circuit chip with signal processing capabilities.

[0195] It is understood that the memory in the embodiments of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. It should be noted that the memory of the systems and methods described herein is intended to include but is not limited to suitable types of memory.

[0196] Based on the above, a readable storage medium is provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above method are implemented.

[0197] The above describes an embodiment of the present invention in conjunction with the accompanying drawings, but the embodiment of the present invention 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 embodiment of the present invention, ordinary technicians in this field can also make many forms without departing from the purpose of the embodiment of the present invention and the scope of protection of the embodiment of the present invention, all of which are protected by the embodiment of the present invention.

Claims

1. An automatic deviation-correcting door panel intelligent cutting optimization method, characterized in that: The method is applied to an optimization system, and the method includes: performing point cloud recognition on target laser sensor detection information of a target door panel to obtain initial point cloud state data corresponding to the target laser sensor detection information; performing image descriptor mining operation on the initial point cloud state data to obtain a target point cloud image descriptor corresponding to the initial point cloud state data; wherein the target point cloud image descriptor at least includes a size descriptor, a contour shape descriptor, and a surface state descriptor; determining X cutting processing task events related to the target laser sensor detection information, and performing a correction discrimination operation on the target point cloud image descriptor according to each cutting processing task event to obtain a correction discrimination recommendation feature corresponding to each cutting processing task event; X is a positive integer; the target point cloud image descriptor is obtained by performing an image descriptor mining operation on the initial point cloud state data by a point cloud description mining module in a first genetic iteration branch; the first genetic iteration branch belongs to a method for performing correction discrimination on the target point cloud image descriptor. A target deviation analysis algorithm for deviation analysis of door panels; the target deviation analysis algorithm also includes a second genetic iteration branch after the point cloud description mining module; the second genetic iteration branch is determined from the target deviation analysis algorithm; the second genetic iteration branch includes a spatial attention model corresponding to each cutting processing task event; through the spatial attention model corresponding to each cutting processing task event, the target point cloud image descriptor is subjected to a correction discrimination operation to obtain the correction discrimination recommendation feature corresponding to each cutting processing task event; a spatial attention model is used to determine the correction discrimination recommendation feature corresponding to a cutting processing task event; the initial point cloud state data is subjected to a processing parameter mining operation to obtain the processing parameter feature corresponding to each cutting processing task event; according to the correction discrimination recommendation feature corresponding to each cutting processing task event and the processing parameter feature corresponding to each cutting processing task event, the automatic correction execution strategy of the target door panel is determined.

2. The method according to claim 1, characterized in that The point cloud recognition of the target laser sensor detection information of the target door panel is performed to obtain the initial point cloud state data corresponding to the target laser sensor detection information, including: performing sensor signal preprocessing on the target laser sensor detection information to obtain the initial laser sensor signal of the target laser sensor detection information; performing point cloud coding recognition on the initial laser sensor signal to obtain the initial point cloud state data of the initial laser sensor signal; the sensor signal preprocessing on the target laser sensor detection information to obtain the initial laser sensor signal of the target laser sensor detection information includes: when the target laser sensor detection information of the target door panel is obtained, obtaining a target deviation analysis algorithm for performing deviation analysis on the target door panel; the target deviation analysis algorithm includes a first genetic iteration branch; determining the sensor element feature related to the target laser sensor detection information through the first genetic iteration branch, and extracting the sensor detection input reflection signal of the target laser sensor detection information according to the sensor element feature; performing signal decoding on the sensor detection input reflection signal through the first genetic iteration branch to obtain the signal decoding result corresponding to the sensor detection input reflection signal ; Performing position distribution feature mapping on the signal decoding result corresponding to the sensor detection input reflection signal through the first genetic iteration branch to obtain the initial laser sensor signal of the target laser sensor detection information; the first genetic iteration branch includes an excitation module; the sensor detection input reflection signal includes Y sensor detection input reflection signals, the Y sensor detection input reflection signals include a target sensor detection input reflection signal, and Y is a positive integer; performing signal decoding on the sensor detection input reflection signal through the first genetic iteration branch to obtain the signal decoding result corresponding to the sensor detection input reflection signal includes: determining the characteristic coordinate system corresponding to the target sensor detection input reflection signal according to the sensor element characteristics to which the target sensor detection input reflection signal belongs; the characteristic coordinate system includes multiple characteristic projection variables, each characteristic projection variable has a corresponding input reflection signal mapping set; determining the input reflection signal mapping set to which the target sensor detection input reflection signal belongs through the excitation module, and determining the characteristic projection variable corresponding to the input reflection signal mapping set to which the target sensor detection input reflection signal belongs as the signal decoding result corresponding to the target sensor detection input reflection signal.

3. The method according to claim 1, characterized in that The target point cloud image descriptor is obtained by performing an image descriptor mining operation on the initial point cloud state data by a point cloud description mining module in a first genetic iteration branch; the first genetic iteration branch belongs to a target deviation analysis algorithm for performing deviation analysis on the target door panel; the method also includes: obtaining a first algorithm debugging sample; the first a priori debugging sample included in the first algorithm debugging sample is a priori debugging sample that does not carry a deviation annotation; obtaining Z debugging items for the first original genetic iteration branch; Z is a positive integer; processing the first a priori debugging sample through the first original genetic iteration branch to obtain a point cloud image sample descriptor associated with each debugging item, and determining Z training error data corresponding to the Z debugging items based on the point cloud image sample descriptor; one debugging item corresponds to one training error data; determining first target training error data based on the Z training error data, improving the algorithm configuration weight of the first original genetic iteration branch based on the first target training error data, and determining the improved first original genetic iteration branch as the first genetic iteration branch.

4. The method according to claim 3, characterized in that The Z debugging projects include a signal noise decoding project; the first a priori debugging sample is processed through the first original genetic iteration branch to obtain a point cloud image sample descriptor associated with each debugging project, and Z training error data corresponding to the Z debugging projects are determined according to the point cloud image sample descriptor, including: sensor signal preprocessing is performed on the first a priori debugging sample through the first original genetic iteration branch to obtain a first initial laser sensor signal sample of the first a priori debugging sample; the first initial laser sensor signal sample includes a signal decoding result sample under a plurality of distribution labels; a first distribution label to be subjected to signal noise addition is arbitrarily determined from a plurality of distribution labels corresponding to the first initial laser sensor signal sample; signal noise addition is performed on the signal decoding result sample under the first distribution label in the first initial laser sensor signal sample according to the signal noise addition information, A first noisy sample is obtained; point cloud coding recognition is performed on the first noisy sample through the first original genetic iterative branch to obtain a first initial point cloud state data sample of the first noisy sample, and an image descriptor mining operation is performed on the first initial point cloud state data sample to obtain a first point cloud image sample descriptor corresponding to the first initial point cloud state data sample; the first point cloud image sample descriptor belongs to the point cloud image sample descriptor; a noise decoding operation is performed on the first point cloud image sample descriptor through a decoding module related to the signal noise decoding project to obtain the noise signal decoding prediction information corresponding to the signal noise information in the first noisy sample; based on the signal decoding result sample under the first distribution label in the first initial laser sensor signal sample and the noisy signal decoding prediction information, the training error data corresponding to the signal noise decoding project is determined.

5. The method according to claim 3, characterized in that The Z debugging projects include a signal iteration decoding project; the first a priori debugging sample is processed through the first original genetic iteration branch to obtain a point cloud image sample descriptor associated with each debugging project, and Z training error data corresponding to the Z debugging projects are determined according to the point cloud image sample descriptor, including: sensor signal preprocessing is performed on the first a priori debugging sample through the first original genetic iteration branch to obtain a first initial laser sensor signal sample of the first a priori debugging sample; the first initial laser sensor signal sample includes signal decoding result samples under multiple distribution labels; from the multiple distribution labels corresponding to the first initial laser sensor signal sample, a second distribution label to be subjected to signal iteration is arbitrarily determined; according to the signal iteration information, a signal decoding result sample under the second distribution label in the first initial laser sensor signal sample is subjected to signal iteration processing to obtain to the second noisy sample; the signal change information is different from the signal decoding result sample under the second distribution label; the second noisy sample is subjected to point cloud coding recognition through the first original genetic iterative branch to obtain the second initial point cloud state data sample of the second noisy sample, and the second initial point cloud state data sample is subjected to image descriptor mining operation to obtain the second point cloud image sample descriptor corresponding to the second initial point cloud state data sample; the second point cloud image sample descriptor belongs to the point cloud image sample descriptor; the second point cloud image sample descriptor is subjected to signal change decoding operation through the decoding module related to the signal change decoding project to obtain the signal change decoding information corresponding to the multiple distribution labels; the training error data corresponding to the signal change decoding project is determined according to the signal change decoding information corresponding to the second distribution label and the multiple distribution labels.

6. The method according to claim 3, characterized in that The Z debugging projects include adversarial training projects; the first a priori debugging sample is processed through the first original genetic iteration branch to obtain a point cloud image sample descriptor associated with each debugging project, and Z training error data corresponding to the Z debugging projects are determined according to the point cloud image sample descriptor, including: adversarial processing of the sensor detection input reflection signal of the first a priori debugging sample to obtain an adversarial debugging sample, sensor signal preprocessing of the adversarial debugging sample to obtain a laser sensor adversarial signal of the adversarial debugging sample; sensor signal preprocessing of the first a priori debugging sample to obtain a first initial laser sensor signal sample of the first a priori debugging sample; point cloud coding recognition of the laser sensor adversarial signal through the first original genetic iteration branch to obtain a third initial point cloud state data sample of the laser sensor adversarial signal. For example, an image descriptor mining operation is performed on the third initial point cloud state data sample to obtain a third point cloud image sample descriptor corresponding to the third initial point cloud state data sample; the third point cloud image sample descriptor belongs to the point cloud image sample descriptor; point cloud coding recognition is performed on the first initial laser sensor signal sample through the first original genetic iterative branch to obtain a fourth initial point cloud state data sample of the first initial laser sensor signal sample, and an image descriptor mining operation is performed on the fourth initial point cloud state data sample to obtain a fourth point cloud image sample descriptor corresponding to the fourth initial point cloud state data sample; the fourth point cloud image sample descriptor belongs to the point cloud image sample descriptor; based on the third point cloud image sample descriptor and the fourth point cloud image sample descriptor, the training error data corresponding to the adversarial training item is determined.

7. The method according to claim 3, characterized in that The Z debugging projects include a delayed signal decoding project; the first a priori debugging sample is processed through the first original genetic iteration branch to obtain a point cloud image sample descriptor associated with each debugging project, and Z training error data corresponding to the Z debugging projects are determined according to the point cloud image sample descriptor, including: sensor signal preprocessing of the first a priori debugging sample to obtain a first initial laser sensor signal sample of the first a priori debugging sample; point cloud coding recognition of the first initial laser sensor signal sample is performed through the first original genetic iteration branch to obtain a fourth initial point cloud state data sample of the first initial laser sensor signal sample, and image descriptor mining operation is performed on the fourth initial point cloud state data sample. A fourth point cloud image sample descriptor corresponding to the fourth initial point cloud state data sample is obtained; the fourth point cloud image sample descriptor belongs to the point cloud image sample descriptor; according to the decoding module related to the delayed signal decoding project, a deviation state decision is made on the fourth point cloud image sample descriptor to obtain a deviation state decision feature corresponding to the first priori debugging sample; a deep convolutional network related to the first original genetic iterative branch is obtained, and a deviation state decision is made on the first priori debugging sample according to the deep convolutional network to obtain a deviation correction discrimination training feature corresponding to the first priori debugging sample; according to the deviation state decision feature and the deviation correction discrimination training feature, the training error data corresponding to the delayed signal decoding project is determined.

8. The method according to claim 1, characterized in that The method further includes: obtaining a second algorithm debugging sample; the second a priori debugging sample included in the second algorithm debugging sample is provided with a task annotation; the task annotation includes X task classification feature values ​​corresponding to the X cutting task events, one cutting task event corresponds to one task classification feature value, and the X task classification feature values ​​are determined according to the cutting task event to which the second a priori debugging sample belongs; obtaining a first genetic iteration branch that completes debugging, performing point cloud recognition on the second a priori debugging sample through the first genetic iteration branch, obtaining a fifth initial point cloud state data sample of the second a priori debugging sample, performing an image descriptor mining operation on the fifth initial point cloud state data sample, and obtaining a fifth point cloud image sample descriptor corresponding to the fifth initial point cloud state data sample; X original spatial attention models corresponding to the X cutting task events are obtained, and task matching analysis is performed on the fifth point cloud image sample descriptors according to the X models to obtain X task matching analysis viewpoints corresponding to the X original spatial attention models; one cutting task event corresponds to one original spatial attention model, and one original spatial attention model is used to determine one task matching analysis viewpoint; second target training error data is determined according to the X task matching analysis viewpoints and the X task classification feature values, and algorithm configuration weights of the X original spatial attention models are improved according to the second target training error data, and the improved X original spatial attention models are determined as X spatial attention models, and the second genetic iteration branch is determined according to the X spatial attention models.

9. An optimization system, characterized in that: The method comprises at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method according to any one of claims 1 to 8.

Images