Automatic review method and system for highway linear design drawings

Abstract

The present invention provides a method and system for automatically reviewing highway linear design drawings. A drawing preprocessing unit is used to obtain a review key element drawing containing key elements. A convolutional neural network is used to extract feature maps and vectorize them, and the identification type and element position identification box of each key element in the review key element drawing are output in parallel. A verification control module is input to verify the key element feature vector and obtain an adaptive review rule. The present invention is aimed at drawings containing multiple types of key elements and non-preset types of key elements, and the positions and spatial forms of each key element present highly dynamic changes, and can effectively improve the efficiency and accuracy of key element identification, analysis, and judgment.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and system for automatically reviewing highway linear design drawings. Background Art

[0002] Highway alignment design drawings are used to show the line shape, geometric structure, intersection structure and other contents in highway design. Highway alignment design drawings are required to meet the following characteristics: (1) Accuracy: The design drawings need to accurately reflect the alignment, size and location of the highway. (2) Standardization: The drawing must comply with relevant national or industry standards and specifications. (3) Dynamicity: During the highway design process, the drawings may need to be frequently adjusted and updated according to actual conditions.

[0003] The review of highway alignment design drawings mainly includes the following contents: (1) Whether the alignment design is reasonable and complies with relevant technical standards and specifications; for example, whether parameters such as curve radius, slope, superelevation, widening, etc. meet the requirements of the specifications. (2) Whether the design of the intersection structure is reasonable, such as the size and shape of the intersection. (3) The completeness and accuracy of the drawings, such as whether the dimensioning, symbol use, and text description are clear and accurate. (4) The consistency of the drawings, ensuring that the information between all related drawings is consistent and there is no conflict.

[0004] Due to the requirements of accuracy, dynamism and standardization of highway alignment design drawings, traditional manual review can no longer meet the requirements of review accuracy and efficiency. Therefore, the application of automatic review technology with the help of computer digitization is becoming more and more widespread. The automatic review of highway alignment design drawings usually includes the following steps: (1) Drawing preprocessing: converting the design drawings into a machine-readable format, such as converting PDF to image or vector graphics format; (2) Element recognition: using image recognition technology to extract various elements in the drawings as key elements of the review object, such as symbols representing lines, geometric structures, and intersection structures; among them, machine learning models can be used to identify key elements in the drawings, such as using convolutional neural networks (CNNs) to identify graphic elements. (3) Feature extraction of key elements, including: extracting design specifications, automatically confirming the design specifications and standards that should be followed in the drawings; extracting geometric parameters, extracting geometric parameters of the alignment design (such as curve radius, slope, etc.); extracting structural design parameters, extracting structural design parameters such as roadbed, bridges, tunnels, etc.; extracting intersection design parameters, extracting parameters such as the size and shape of the intersection design. (4) Rule matching and review: Use the pre-set rule base to match the extracted information with the rules and identify designs that do not meet the specifications. (5) Audit result output: Output the audit results, including non-compliance with specifications, possible problems, and recommended modifications; if necessary, the results of the automatic audit can be manually reviewed to ensure the accuracy and reliability of the audit.

[0005] At present, the main problems of automatic review of highway alignment design drawings are: there is often a high rate of non-recognition and missed recognition of key elements; misjudgment of key element types leads to the application of incorrect review rules; and large delays in operation time. In particular, some drawings have many element types, diverse and unclear element types, mixed positional relationships, and disordered and variable distribution rules. It is difficult to provide a good automatic review solution in the existing technology. Summary of the invention

[0006] The present invention provides a method and system for automatically reviewing highway linear design drawings. In the case where drawings contain multiple types and non-preset key elements, and the positions and forms of each element are diverse and irregular, the present invention can effectively improve the efficiency and accuracy of key element identification and type judgment of the above drawings during the automatic review process, and improve the reliability and operation efficiency of the above automatic review function.

[0007] The present invention provides a method for automatically reviewing highway linear design drawings, comprising the following steps:

[0008] Step S1, obtaining drawings as key elements for review and performing drawing preprocessing;

[0009] Step S2: for the audit key element drawings, a convolutional neural network is used to extract feature maps and vectorize them, and then tensor encoding of global features is performed, and self-attention interactive decoding is performed on the key element key element group and the global feature tensor to achieve parallel recognition of multiple key elements, and the identification type and element position identification box of each key element in the audit key element drawings are output in parallel;

[0010] Step S3, taking the vector representing the identification type and element position identification box of each key element in the audit key element drawing as the key element feature vector, inputting it into the verification control module, verifying the key element feature vector and obtaining the adapted audit rules.

[0011] Preferably, the step S2 specifically includes:

[0012] Step S2A, constructing a training set consisting of pairs of sample audit key element drawings and key element labels, and constructing an initialized convolutional neural network, inputting the sample audit key element drawings into the convolutional neural network for training, and extracting feature graphs of the sample audit key element drawings by the convolutional neural network and vectorizing them;

[0013] Step S2B, constructing and initializing an encoder for tensor encoding of global features and a decoder for performing self-attention interactive decoding on key elements, key element groups and global feature tensors; and inputting the feature map into the encoder to obtain tensor encoding of global features, and then performing self-attention interactive decoding on key elements, key element groups and global feature tensors;

[0014] Step S2C, constructing and initializing a key element type recognition model and an element position frame recognition model for parallel recognition of multiple key elements; then inputting the key element prediction decoding features of the key element key element group into the key element type recognition model and the element position frame recognition model, and outputting the recognition type and element position recognition frame of each key element in the review key element drawing in parallel;

[0015] Step S2D, reviewing the key element drawings of the samples of the training set, calculating the loss function based on the deviation between the sets of N key element identification types and element position identification frames output by the key element type identification model and the element position frame identification model and the key element labels of the training set, and obtaining an estimate of the model parameters;

[0016] Step S2E, after completing the training, for the audit key element drawings actually provided, the convolutional neural network is used to extract the feature map and vectorize it, and then the global feature tensor encoding is performed, and the key element group and the global feature tensor are self-attention interactively decoded to achieve parallel recognition of multiple key elements, and the identification type and element position identification box of each key element in the audit key element drawings are output in parallel.

[0017] Preferably, in step S2A, the training set consisting of pairs of sample audit key element drawings and key element labels is represented as:

[0018]

[0019] in is the total number of training samples in the training set, is the first Sample review key element drawings, is the drawing height of the drawing, is the drawing width of the drawing; is the training set Sample audit key elements drawings have the most The key element type and the key element position box positioning vector together constitute the key element label. is a pre-set integer that is significantly larger than the number of key elements that may exist in the sample audit key element drawings; , It is Sample review of key elements of the drawings The key element type of the key element, It is Sample review of key elements of the drawings A 4-dimensional vector consisting of the center point coordinates, frame height value, and frame width value of the key element position box of each key element.

[0020] Preferably, in step S2A, the initialized convolutional neural network is composed of a convolutional layer structure of a pre-trained ResNet model and a full convolutional network, and the convolutional layer structure of the pre-trained ResNet model is expressed as ,in It is the parameter tensor consisting of all the parameters of the convolutional layer of the pre-trained ResNet model. The key elements of the audit drawing for inputting the convolutional neural network; the full convolutional network is represented as , where the parameter tensor Initialize to ,and , Represents a low-resolution feature map generated by the convolutional neural network.

[0021] Preferably, in step S2A, the sample audit key element drawings are As a key element of the audit drawings , input the initialized convolutional neural network to obtain a low-resolution feature map The characteristic diagram To vectorize, use indivual The convolution kernel concatenated , the feature map The dimension is reduced to the number of channels New high-level feature map ,Right now

[0022]

[0023] symbol Representing images The convolution operation between the convolution kernel and the ; The feature map and positional encoding parameters Add, the position encoding parameter Initialize to , and then the sum of the two is calculated in the tensor Straighten the plane shape, thereby changing the shape , and transposed to The tensor of is used as the input of the encoder of the global feature tensor encoding, denoted as ,Right now

[0024] .

[0025] Preferably, in step S2B, the encoder and decoder are initialized The parameter tensor of , the encoder parameters and decoder Together they form the parameter tensor ; and, initialize the parameter tensor for the key element key element group in the decoder ; The vectorized Input the encoder of the self-attention mechanism and get Global feature encoding of the same shape , Right now

[0026] ,

[0027] , is the parameter tensor of the encoder; and the parameter tensor of the key element key element group Through the decoder of the cross attention mechanism, the predicted decoding features of N key elements in the key element group are obtained. , Right now

[0028] ,

[0029] is the N key elements predicting the decoding features The tensor composed of is the decoder parameter tensor.

[0030] Preferably, in step S2C, the identification model for each key element type is initialized. The parameter tensor in , and initialize the feature location box recognition model The parameter tensor in ; Predict the decoding features of N key elements Identify models by key element type and feature location frame recognition model , get the set of N identification types and element position identification boxes in the audit key element drawings , Right now .

[0031] Preferably, in step S2E, for the audit key element drawings actually provided, the identification type and element position identification box of each key element output in parallel are represented as follows:

[0032]

[0033] This is the first key element included in the audit drawing. A vector of probabilities that the key elements belong to all possible recognition categories, The The vector of the feature position identification box positioning of the key feature.

[0034] Preferably, in step S3, the verification control module uses a preset rule base, takes the extracted recognition type of the key element and the vector of the element position recognition box as a query vector, and matches it with the rule.

[0035] The present invention provides an automatic review system for highway linear design drawings, which is characterized by comprising:

[0036] A drawing pre-processing unit, used to obtain audit key element drawings containing key elements;

[0037] A key element identification module is used to extract and vectorize feature maps for the audit key element drawings using a convolutional neural network, and then perform tensor encoding of global features, perform self-attention interactive decoding on key element key element groups and global feature tensors, realize parallel identification of multiple key elements, and output the identification type and element position identification box of each key element in the audit key element drawings in parallel;

[0038] The verification control module is used to input the vectors representing the identification type and element position identification box of each key element in the audit key element drawing as key element feature vectors into the verification control module, verify the key element feature vectors and obtain adapted audit rules.

[0039] The present invention is aimed at the case where the drawings contain multiple types and non-preset types of key elements, and the positions and shapes of each element are diverse and irregular. The present invention can effectively improve the efficiency and accuracy of key element identification and type judgment of the above drawings during the automatic review process, thereby improving the reliability and operating efficiency of the above automatic review function. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The following is a brief introduction to the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0041] Figure 1 It is a general flow chart of a method for automatically reviewing highway linear design drawings provided by the present invention;

[0042] Figure 2 The present invention provides an overall structural block diagram of an automatic review system for highway linear design drawings. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the implementation of the present invention clearer, the technical solution in the embodiment of the present invention will be described in more detail below with reference to the accompanying drawings in the embodiment of the present invention.

[0044] It should be noted that: in the accompanying drawings, the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The described embodiments are part of the embodiments of the present invention, not all of the embodiments. In the absence of conflict, the embodiments in this application and the features in the embodiments can be combined with each other. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] Reference below Figure 1 A specific embodiment of a method for automatically reviewing highway linear design drawings disclosed in the present invention is described in detail.

[0046] like Figure 1 As shown, this embodiment mainly includes the following steps: S1. Using a drawing preprocessing unit, obtain an audit key element drawing containing key elements; S2. For the audit key element drawing, use a convolutional neural network to extract feature maps and vectorize them, and then perform tensor encoding of global features, perform self-attention interactive decoding on the key element key element group and the global feature tensor, realize parallel recognition of multiple key elements, and output the identification type and element position identification box of each key element in the audit key element drawing in parallel; S3. Use the vector representing the identification type and element position identification box of each key element in the audit key element drawing as the key element feature vector, input it into the verification control module, verify the key element feature vector and obtain the adapted audit rules.

[0047] In step S1 , the preprocessing of the drawing preprocessing unit includes: converting the design drawing into a machine-readable format, such as converting PDF into an image or vector graphics format.

[0048] In step S2, for the audit key element drawings, based on the fact that they contain multiple types of key elements and non-preset types of key elements, and the position and spatial form of each key element presents a highly dynamic change, this step implements global feature extraction, and then directly performs collective identification on the key elements contained therein, thereby simplifying the process and improving the efficiency of key element identification, avoiding repeated identification predictions and the generation of redundant identification boxes, reducing the amount of prior samples and training, and improving recognition accuracy.

[0049] Specifically, step S2 includes the following sub-steps.

[0050] Step S2A, construct a training set consisting of pairs of sample audit key element drawings and key element labels, and construct an initialized convolutional neural network, input the sample audit key element drawings into the convolutional neural network for training, and use the convolutional neural network to extract feature maps of the sample audit key element drawings and vectorize them.

[0051] The training set consisting of pairs of sample audit key element drawings and key element labels is represented as:

[0052]

[0053] in is the total number of training samples in the training set, is the first Sample review key element drawings, is the drawing height of the drawing, is the drawing width of the drawing; is the training set Sample audit key elements drawings have the most The key element type and the key element position box positioning vector together constitute the key element label. is a pre-set integer that is significantly larger than the number of key elements that may exist in the sample audit key element drawings; , It is Sample review of key elements of the drawings The key element type of the key element (when the key element is the scene background, the value is ), It is Sample review of key elements of the drawings A 4-dimensional vector consisting of the center point coordinates, frame height value, and frame width value of the key element position box of each key element.

[0054] The initialized convolutional neural network is composed of the convolutional layer structure of the pre-trained ResNet model and the full convolutional network. The convolutional layer structure of the pre-trained ResNet model is expressed as ,in It is a parameter tensor consisting of all parameters of the convolutional layer of the pre-trained ResNet model. The key elements of the audit drawing for inputting the convolutional neural network; the full convolutional network is represented as , where the parameter tensor Initialize to ,and , represents a low-resolution feature map generated by the convolutional neural network; as an optional value, , .

[0055] Then, the sample review key elements drawings As a key element of the audit drawings , input the initialized convolutional neural network to obtain a low-resolution feature map .

[0056] The feature map To vectorize, use indivual The convolution kernel concatenated , the feature map The dimension is reduced to the number of channels New high-level feature map ,Right now

[0057]

[0058] symbol Representing images The convolution operation between the convolution kernel and the ; The feature map and positional encoding parameters Add, the position encoding parameter Initialize to , and then the sum of the two is calculated in the tensor The plane shape is straightened, thereby changing the shape (expressed as reshape) , and transposed to The tensor of is used as the input of the encoder of the global feature tensor encoding, denoted as ,Right now

[0059]

[0060] Step S2B, construct and initialize an encoder for tensor encoding of global features and a decoder for performing self-attention interactive decoding on key elements, key element groups and global feature tensors; and input the feature map into the encoder to obtain tensor encoding of global features, and then perform self-attention interactive decoding on key elements, key element groups and global feature tensors.

[0061] Among them, initialization is used for encoder and decoder The parameter tensor of , the encoder parameters and decoder Together they form the parameter tensor . And, initialize the parameter tensor for the key element key element group in the decoder .

[0062] After vectorization, Input the encoder of the self-attention mechanism and get Global feature encoding of the same shape , Right now

[0063] ,

[0064] , is the parameter tensor of the encoder.

[0065] Will and the parameter tensor of the key element key element group Through the decoder of the cross attention mechanism, the predicted decoding features of N key elements in the key element group are obtained. , Right now

[0066] ,

[0067] is the N key elements predicting the decoding features The tensor composed of is the decoder parameter tensor.

[0068] Step S2C, constructs and initializes a key element type recognition model and an element position frame recognition model for parallel recognition of multiple key elements; then inputs the key element prediction decoding features of the key element key element group into the key element type recognition model and the element position frame recognition model, and outputs the recognition type and element position recognition frame of each key element in the audit key element drawing in parallel.

[0069] Initialize the recognition model for each key element type The parameter tensor in , and initialize the feature location box recognition model The parameter tensor in .

[0070] Predict the decoding features of N key elements Identify models by key element type and feature location frame recognition model The above model is composed of a feed forward neural network (FFN), which obtains a set of N identification types and element position identification boxes in the audit key element drawings. , Right now .

[0071] Step S2D, review the key element drawings for the samples of the training set, calculate the loss function based on the deviation of the set of N key element identification types and element position identification boxes output by the key element type identification model and the element position box identification model from the key element labels of the training set, and obtain an estimate of the model parameters.

[0072] Review the key element drawings output by optimizing the sample for the training set. The key element labels of the training set The loss function between them is used to obtain the parameter estimates of each model, which is expressed as: , , , , , , . Thus, the training process using the training set is completed.

[0073] Step S2E, after completing the training, for the audit key element drawings actually provided, the convolutional neural network is used to extract the feature map and vectorize it, and then the global feature tensor encoding is performed, and the key element group and the global feature tensor are self-attention interactively decoded to achieve parallel recognition of multiple key elements, and the identification type and element position identification box of each key element in the audit key element drawings are output in parallel.

[0074] For the audit key element drawings actually provided, the identification type and element position identification box of each key element output in parallel are expressed as follows:

[0075]

[0076] This is the first key element included in the audit drawing. A vector of probabilities that the key elements belong to all possible recognition categories, The The vector of the feature position identification box positioning of the key feature.

[0077] The above vectors represent the identification type and feature location identification box of each key element. , as the feature vector of the scene. In step S3, the above key element feature vector An input verification control module is used to obtain a response control amount adapted to the scenario for the key control elements.

[0078] In step S3, a preset rule library is used to take the extracted recognition type of the key element and the vector of the element position recognition box as a query vector and match it with the rule.

[0079] See also Figure 2 The present invention further provides a highway linear design drawing automatic review system, comprising:

[0080] A drawing preprocessing unit 201 is used to obtain a review key element drawing containing key elements;

[0081] The key element identification module 202 is used to extract and vectorize the feature map of the audit key element drawing using a convolutional neural network, and then perform tensor encoding of the global feature, perform self-attention interactive decoding on the key element key element group and the global feature tensor, realize parallel identification of multiple key elements, and output the identification type and element position identification box of each key element in the audit key element drawing in parallel;

[0082] The verification control module 203 is used to use the vector representing the identification type and element position identification box of each key element in the audit key element drawing as the key element feature vector as the query vector, use the pre-set rule library, and match with the rules.

[0083] The present invention is aimed at the case where the drawings contain multiple types and non-preset types of key elements, and the positions and shapes of each element are diverse and irregular. The present invention can effectively improve the efficiency and accuracy of key element identification and type judgment of the above drawings during the automatic review process, thereby improving the reliability and operating efficiency of the above automatic review function.

[0084] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for automatically reviewing highway linear design drawings, characterized in that: The following steps are involved: Step S1, using a drawing preprocessing unit to obtain a review key element drawing containing key elements; Step S2: for the review key element drawings, a convolutional neural network is used to extract feature maps and vectorize them, and then the feature maps are encoded to obtain tensor encoding of global features, and self-attention interactive decoding is performed on the key element key element group and the global feature tensor to achieve parallel recognition of multiple key elements, and the recognition type and element position recognition box of each key element in the review key element drawings are output in parallel; Step S3, taking the vector representing the identification type and element position identification box of each key element in the audit key element drawing as the key element feature vector, inputting it into the verification control module, verifying the key element feature vector and obtaining the adapted audit rules.

2. The method for automatically reviewing highway linear design drawings according to claim 1 is characterized in that: The step S2 specifically includes: Step S2A, constructing a training set consisting of pairs of sample audit key element drawings and key element labels, and constructing an initialized convolutional neural network, inputting the sample audit key element drawings into the convolutional neural network for training, and extracting feature graphs of the sample audit key element drawings by the convolutional neural network and vectorizing them; Step S2B, constructing and initializing an encoder for tensor encoding of global features and a decoder for performing self-attention interactive decoding on key elements, key element groups and global feature tensors; and inputting the feature map into the encoder to obtain tensor encoding of global features, and then performing self-attention interactive decoding on key elements, key element groups and global feature tensors; Step S2C, constructing and initializing a key element type recognition model and an element position frame recognition model for parallel recognition of multiple key elements; then inputting the key element prediction decoding features of the key element key element group into the key element type recognition model and the element position frame recognition model, and outputting the recognition type and element position recognition frame of each key element in the review key element drawing in parallel; Step S2D, reviewing the key element drawings of the samples of the training set, calculating the loss function based on the deviation between the sets of N key element identification types and element position identification frames output by the key element type identification model and the element position frame identification model and the key element labels of the training set, and obtaining an estimate of the model parameters; Step S2E, after completing the training, for the audit key element drawings actually provided, the convolutional neural network is used to extract the feature map and vectorize it, and then the global feature tensor encoding is performed, and the key element key element group and the global feature tensor are self-attention interactively decoded to achieve parallel recognition of multiple key elements, and the identification type and element position identification box of each key element in the audit key element drawings are output in parallel.

3. The method for automatically reviewing highway linear design drawings according to claim 2 is characterized in that: In step S2A, the training set consisting of pairs of sample audit key element drawings and key element labels is represented as: in is the total number of training samples in the training set, is the first Sample review of key element drawings, is the drawing height of the drawing, is the drawing width of the drawing; is the training set Sample review of key elements drawings with the most The key element label is composed of the key element type and the key element position box positioning vector. is a pre-set integer that is significantly larger than the number of key elements that may exist in the sample audit key element drawings; , It is Sample review of key elements in the drawings The key element type of the key element, It is Sample review of key elements in the drawings A 4-dimensional vector consisting of the center point coordinates, box height value, and box width value of the key element position box of a key element.

4. The method for automatically reviewing highway linear design drawings according to claim 3 is characterized in that: In step S2A, the initialized convolutional neural network is composed of the convolutional layer structure of the pre-trained ResNet model and the full convolutional network. The convolutional layer structure of the pre-trained ResNet model is expressed as ,in It is the parameter tensor consisting of all the parameters of the convolutional layer of the pre-trained ResNet model. The key element drawings for auditing the input convolutional neural network are shown as follows: , where the parameter tensor Initialize to ,and , Represents a low-resolution feature map generated by the convolutional neural network.

5. The method for automatically reviewing highway linear design drawings according to claim 4 is characterized in that: In step S2A, the sample review key element drawings Drawings as key elements for review , input the initialized convolutional neural network to obtain a low-resolution feature map The characteristic diagram To vectorize, use indivual The convolution kernel Spliced , the feature map The dimension is reduced to the number of channels New high-level feature map ,Right now symbol Representing images The convolution operation between the convolution kernel and the ; The feature map and positional encoding parameters Add, the position encoding parameter Initialize to , and then the sum of the two is calculated as The flat shape is straightened, thereby changing the shape , and transposed to The tensor of is used as the input of the encoder of the global feature tensor encoding, denoted as ,Right now 。 6. The method for automatically reviewing highway linear design drawings according to claim 5 is characterized in that: In step S2B, the parameter tensor for the key element key element group in the decoder is initialized ; After vectorization, Input the encoder of the self-attention mechanism and get Global feature encoding of the same shape , Right now , , is the parameter tensor of the encoder; and the parameter tensor of the key element key element group Through the decoder of the cross attention mechanism, the predicted decoding features of N key elements in the key element group are obtained. , Right now , is the N key element prediction decoding feature The tensor composed of is the decoder parameter tensor.

7. The method for automatically reviewing highway linear design drawings according to claim 6 is characterized in that: In step S2C, the identification model for each key element type is initialized. The parameter tensor in , and initialize the feature location box recognition model The parameter tensor in ; Predict the decoding features of N key elements Identify models by key element type and feature location frame recognition model , get a set of N identification types and element position identification boxes in the audit key element drawings ,Right now .

8. The method for automatically reviewing highway linear design drawings according to claim 7 is characterized in that: In step S2E, for the audit key element drawings actually provided, the identification type and element position identification box of each key element output in parallel are represented as follows: This is the first key element included in the audit drawing. A vector of probabilities that the key elements belong to all possible recognition categories, The The vector used to locate the feature position identification box of each key element.

9. The method for automatically reviewing highway linear design drawings according to claim 8 is characterized in that: In step S3, the verification control module performs a cyclic rolling optimization to obtain the expected key element feature vector of the expected scene feature. and the key element feature vector Proportional coefficient for predictive fitting Assign values ​​to generate the key element feature vector of the fitting y d : in, Indicates the prediction strength of the predictive fit, which is used to indicate the number of time periods ahead of the current scenario. For the predicted intensity Under the conditions k The output value of the fitting key element eigenvector of the period, For the k+ 1 -i The reference value of the expected key element eigenvector for the period, For the k+ 1 -i The actual key element eigenvector of the period; is the proportional coefficient of the fitting, by cyclic scrolling The value of , , Based on the three parameters after cyclic rolling optimization, the fitting key element characteristic vector used to calculate the response control amount is obtained. y d .

10. A highway linear design drawing automatic review system, characterized in that: include: A drawing pre-processing unit, used to obtain audit key element drawings containing key elements; A key element identification module is used to extract and vectorize feature maps for the review key element drawings using a convolutional neural network, and then encode the feature maps to obtain tensor encoding of global features, perform self-attention interactive decoding on the key element key element group and the global feature tensor, realize parallel recognition of multiple key elements, and output the recognition type and element position recognition box of each key element in the review key element drawings in parallel; The verification control module is used to input the vector representing the identification type and element position identification box of each key element in the audit key element drawing as the key element feature vector into the verification control module, verify the key element feature vector and obtain the adapted audit rules.

Images