Method and system for identifying and collecting drawing information of nuclear power valve

Abstract

The invention provides a nuclear power valve drawing information identification and acquisition method and system, and the method comprises the steps: obtaining a target valve drawing, and carrying out the preprocessing of geometric requirements and formats of the target valve drawing, and obtaining a first target valve drawing; pre-identifying the first target valve drawing information by using a CRNN valve drawing information identification model, and sequentially performing drawing image information identification removal, drawing pixel channel merging, drawing information filtering processing and threshold value processing after identification is completed to obtain a second target valve drawing; identifying information of a to-be-identified area of the second target valve drawing by using a CRNN valve drawing information identification model; adopting a CRNN valve drawing information identification model to extract and collect information of an identification result in the step 3, and classifying and integrating the collected information; and performing correctness evaluation on the classified and integrated drawings, and then operating correct and effective information into a valve drawing information database.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of nuclear power valve design, and particularly relates to a nuclear power valve drawing information recognition and collection method and system. BACKGROUND

[0002] Nuclear power valves are widely used in reactor primary loop systems, and there are currently many problems such as a large number of design drawings submitted by suppliers, heavy review tasks, and the like, which result in the information involved being unable to be saved in a timely manner.

[0003] Moreover, nuclear power valves involve many types, such as gate valves, stop valves, ball valves, butterfly valves, check valves, safety valves, and the like, and the design information of different valve drawing types is different, and at the same time, the design drawings of different nuclear power project suppliers differ greatly in information expression style, text font, legend annotation, and the like.

[0004] Valves are important components in nuclear power pipeline systems, and they can be key equipment when the system is executing functions, so the design optimization of nuclear power valves restricts the design optimization of nuclear power system functions. Important design parameter information in nuclear power valve drawings should be properly saved, otherwise it will cause waste of valve design data resources.

[0005] It can be seen that the information contained in the nuclear power valve drawings can be roughly divided into: assembly drawing information, design requirement information, material specification table information, technical parameter information, detailed design detail information, and other information. Among them, the assembly drawing and detailed design detail information belong to image information, and the rest belong to parameter information. Parameter information is important information for nuclear power valve design, and can be further divided into text information (such as design requirements, standards and specifications), numerical information (such as component quantity), and character information (such as material standard name).

[0006] The existing drawing information collection system or method can recognize the design parameter information in the drawings, but still has the following problems or deficiencies:

[0007] 1. The existing scheme mainly targets industrial valve drawings, lacks pertinence and applicability for nuclear power valve drawing information parameter recognition, and is difficult to practically solve the engineering problems of nuclear power valve equipment, and cannot adapt to the special requirements of nuclear power valve drawing parameter information recognition.

[0008] 2. The existing scheme only focuses on using novel and advanced drawing parameter recognition methods for model building to improve the model recognition capability, but in fact, the model needs to be trained according to the drawings, so the drawings involved in the model training will affect the actual capability of the model. It can be seen that the existing scheme lacks method-based self-learning and self-enhancement, and the actual capability of the model cannot be fundamentally updated, enhanced and iteratively upgraded.

[0009] Three, the existing scheme is mainly aimed at the collection and identification process of drawing information, and stops at this process, lacks the classification, storage and management of data information, and cannot truly achieve the effective use of valve design parameter resources, resulting in the waste of drawing parameter data resources, and the important information of the drawing is not judged, and the effectiveness and reliability of the drawing information cannot be evaluated;

[0010] Four, the machine learning model used in the existing scheme is relatively old and outdated, and its calculation analysis and identification judgment ability is not excellent, and it lacks effectiveness for the key elements of nuclear power valve drawing recognition. The traditional OCR (Optical Character Recognition) recognition method is mostly to cut the text into single characters and then recognize the characters, and the recognition ability for indefinite length text sequence is poor, so the practicability for the image sequence recognition problem is not strong.

[0011] Therefore, the present application provides a nuclear power valve drawing information recognition and collection method and system to solve one or more of the above technical problems. SUMMARY

[0012] The technical problem to be solved by the present application is to overcome the defects of low accuracy and utilization rate of valve drawing parameter recognition in the prior art, and to provide a nuclear power valve drawing information recognition and collection method and system.

[0013] The present application solves the above technical problems by the following technical solutions:

[0014] The present application provides a nuclear power valve drawing information recognition and collection method by the first aspect, comprising:

[0015] Step 1, obtaining a target valve drawing, and pre-processing the target valve drawing in terms of geometric requirements and format to obtain a first target valve drawing;

[0016] Step 2, using a valve drawing information recognition model of CRNN to pre-identify the first target valve drawing information, and sequentially performing drawing image information recognition removal, drawing pixel channel merging, drawing information filtering and threshold processing after identification to obtain a second target valve drawing;

[0017] Step 3, using the valve drawing information recognition model of CRNN to identify the information of the to-be-identified region of the second target valve drawing;

[0018] Step 4, using the valve drawing information recognition model of CRNN to extract and collect the information of the identification result in step 3, and classifying and integrating the collected information;

[0019] Step 5, the classified integrated drawing is evaluated for correctness, and then correct and effective information is operated into the valve drawing information database; wherein the valve drawing information database is associated with the valve drawing information recognition model of the CRNN.

[0020] According to one embodiment of the present application, the step 1 comprises:

[0021] Step 11, the target valve drawing is read by using imread() of OpenCV;

[0022] Step 12, the format of the target valve drawing is checked and judged, and the format of the target valve drawing which is not in JPG format is converted into JPG format;

[0023] Step 13, the drawing angle check based on pixel point coordinates and the drawing pixel height transformation are completed by pixel operation, and the first target valve drawing is obtained.

[0024] According to one embodiment of the present application, the step 2 comprises:

[0025] Step 21, the first target valve drawing is received, and pixel check and key valve recognition are performed on the first target valve drawing information, and the third target valve drawing without abnormal pixel points is obtained;

[0026] Step 22, image information recognition is performed on the third target valve drawing;

[0027] Step 23, after the image recognition is completed, the third target valve drawing file is merged in three directions by using the merge() method of OpenCV, and the fourth target valve drawing is obtained;

[0028] Step 24, the fourth target valve drawing after merging is subjected to bilateral filtering processing;

[0029] Step 25, the drawing after filtering processing is subjected to drawing pixel binary threshold processing, so that the pixel information in the drawing is highlighted.

[0030] According to one embodiment of the present application, in the step 21, by acquiring the coordinate pixels of the first target valve drawing, the abnormal pixel point area between 0-255 is introduced into the valve drawing information recognition model based on the CRNN for pre-recognition of key information.

[0031] If there is no abnormal pixel point, the method proceeds to step 22; if there is an abnormal pixel point, key information is identified by using the valve drawing information recognition model of the CRNN, error information is fed back to the valve drawing information database for the drawing that fails in key information identification and belongs to a key valve, and for the drawing that fails in key information identification and does not belong to a key valve, corresponding key information is invalidated.

[0032] According to an embodiment of the present application, in step 22, the valve drawing image recognition model of the CNN is used to remove image information that does not belong to the image information to be identified and collected in the third target valve drawing; wherein the valve assembly drawing and detailed design detail image information in the third target valve drawing is image information that does not belong to the image information to be identified and collected.

[0033] According to an embodiment of the present application, in step 3, the valve drawing information recognition model of the CRNN includes a CRNN convolution layer, a CRNN cycle layer and a CRNN transcription layer, the CRNN convolution layer is used to complete the extraction of text features of the information to be identified, the CRNN cycle layer is used to extract sequence features and generate a probability feature sequence, and the CRNN transcription layer is used to predict the probability feature sequence and convert it into a character sequence.

[0034] According to an embodiment of the present application, step 4 includes:

[0035] Step 41, receiving the identified result information in step 3, using an end-to-end identification mode of the valve drawing information recognition model of the CRNN to extract drawing information;

[0036] Step 42, classifying and integrating the extracted information.

[0037] According to an embodiment of the present application, step 5 includes:

[0038] Step 51, receiving the classified and integrated drawing information in step 4, comparing the drawing information with a design key information library composed of valve design files, finding out error information through comparison and correcting the error information;

[0039] Step 52, correcting the correct information to a valve drawing information database; wherein the valve drawing information database is associated with the valve drawing information recognition model of the CRNN.

[0040] According to a computer program product of the second aspect of the present application, the computer program is executed by a processor to realize the steps executed by the computer in the method of the first aspect.

[0041] According to a computer readable storage medium of a third aspect of the present application, a computer program is stored, and the program is executed by a processor to realize the steps executed by a computer in the method of the first aspect.

[0042] According to a collection system of a fourth aspect of the present application, a memory is capable of storing instructions executable by a processor, and the processor is capable of executing the instructions to realize the steps executed by a computer in the method of the first aspect.

[0043] The positive progress effect of the present application is that:

[0044] The nuclear power valve drawing information recognition and collection method can efficiently complete the nuclear power valve drawing information recognition and collection work, and can reduce the cost of nuclear power valve drawing review, system design optimization and the like, and meanwhile, the model self-strengthening and self-upgrading are carried out for the recognition model, so that the model can maintain high recognition ability, and provide support for nuclear power station system equipment design optimization, drawing review, data verification and the like. BRIEF DESCRIPTION OF DRAWINGS

[0045] The above and other features, properties and advantages of the present application will become more apparent through the following description with reference to the accompanying drawings and embodiments, in which:

[0046] Figure 1 Flow chart of an embodiment of the nuclear power valve drawing information recognition and collection method of the present application;

[0047] Figure 2 Flow chart of the nuclear power valve drawing preprocessing of the present application;

[0048] Figure 3 Schematic diagram of drawing pixel inspection and rotation geometry inspection;

[0049] Figure 4 Flow chart of the nuclear power valve drawing information judgment of the present application;

[0050] Figure 5 Flow chart of the nuclear power valve drawing information recognition of the present application;

[0051] Figure 6 Flow chart of the nuclear power valve drawing information collection of the present application;

[0052] Figure 7 Flow chart of the nuclear power valve drawing information evaluation of the present application;

[0053] Figure 8 Structural schematic diagram of the collection system of the present application. DETAILED DESCRIPTION

[0054] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings.

[0055] Reference will now be made in detail to the embodiments of the present application, examples of which are illustrated in the accompanying drawings. Wherever possible, the same reference numbers will be used throughout the drawings to refer to the same or like parts. In addition, although the terms used in the present application are selected from generally known and used terms, some of the terms mentioned in the description of the present application can be selected by the applicant in his or her judgment, and the detailed meanings thereof are described in relevant parts of the description. Furthermore, the present application should not be understood to be limited only to the actual terms used but also to also the meanings of each term underlying the terms.

[0056] The present application proposes a nuclear power valve drawing information recognition and collection method, which can refer to Figure 1 , Figure 1 The flow of the nuclear power valve drawing information recognition and collection is shown, wherein the nuclear power valve drawing is sequentially subjected to pretreatment, judgment, recognition, collection and evaluation after input, and the model training of the valve drawing information recognition model supporting CRNN is carried out, the obtained drawing information is evaluated and corrected, and the classified and integrated output information file can also be used for model training to ensure the adaptability and applicability of model self-learning and self-strengthening.

[0057] Specifically, the collection method comprises the following steps:

[0058] Step 1, obtaining a target valve drawing and pretreating the target valve drawing in terms of geometric requirements and format to obtain a first target valve drawing.

[0059] In step 1, the target valve drawing is first received, and then the target valve drawing is converted into a file meeting the geometric requirements and format requirements.

[0060] Referring to Figure 2 , the pretreatment specifically comprises:

[0061] Step 11, reading the target valve drawing by using imread() of OpenCV;

[0062] Step 12, checking and judging the format of the target valve drawing, and converting the format of the target valve drawing which is not in JPG format into JPG format; for example, converting the format of the valve drawing file which is in “*.pdf” format into JPG format.

[0063] Step 13, completing the drawing angle check based on pixel point coordinates and the drawing pixel height conversion by pixel operation to obtain the first target valve drawing.

[0064] Specifically, in step 13, Numpy is used for paper pixel operation, and the vertical coordinates of the black frame on the left side of the first column of the paper are obtained. As shown in Figure 3 , if all the pixel values of the coordinates obtained are 0 (three-way channel pixel value B = G = R = 0), that is, pure black pixels, it indicates that the paper is currently in a vertical state and does not need to be rotated. Otherwise, the pixel coordinate position of the pixel value of 0 is obtained to the right of the column, and the angle between the front and rear two pixel coordinates is formed at this time, indicating that the current paper is skewed. The angle between the pixel coordinate position direction and the vertical direction is calculated, and the paper is rotated using the getRotationMatrix2D() method of OpenCV with the lower left corner of the paper as the rotation coordinate center point and the obtained deflection angle as the rotation angle.

[0065] Then the size() method of OpenCV is used to obtain the pixel values of the width and height of the paper. Since the CRNN model used in the subsequent paper information recognition module requires the input paper to have a fixed geometric pixel height of 32, the dsize() method of OpenCV is also used to perform geometric transformation on the input file that does not meet the pixel height requirement.

[0066] Step 2, using the valve paper information recognition model of CRNN to pre-identify the first target valve paper information, and sequentially performing paper image information recognition removal, paper pixel channel merging, paper information filtering and threshold processing to obtain the second target valve paper.

[0067] Referring to Figure 4 , step 2 includes:

[0068] Step 21, receiving the first target valve paper, and performing pixel inspection and key valve identification on the first target valve paper information to obtain a third target valve paper without abnormal pixel points.

[0069] In step 21, by obtaining the coordinate pixels of the first target valve paper, the abnormal pixel point region between 0-255 is introduced into the valve paper information recognition model based on CRNN for pre-identification of key information.

[0070] If there is no abnormal pixel point, step 22 is entered; if there is an abnormal pixel point, the paper key information is checked, and the valve paper information recognition model of CRNN is used for key information identification. For the paper whose key information identification fails and belongs to the key valve, error information is fed back to the valve paper information database, and for the paper whose key information identification fails and does not belong to the key valve, the corresponding key information is invalidated (the process in Figure 2 is ended).

[0071] Step 22, image information recognition is performed on the third target valve paper.

[0072] Because the present application is aimed at character parameter information recognition and collection in drawings, the image information of valve assembly drawings, detailed design detail information and the like existing in the drawings does not belong to the information to be recognized and collected, and will cause interference to the recognition and collection of the character parameter information, and even lead to information recognition failure. Therefore, the valve drawing image recognition model of CNN (Convolutional Neural Network) is adopted in the present application, the existing valve organization image data is used to carry out model training, and is used for image recognition in the present step, the position of the valve assembly drawing and the detailed design detail information image is positioned after the recognition, and the image pixel value in the positioning frame is set to 255 by using the pixel operation of Numpy. That is, the position region becomes pure white, so as to remove the corresponding image information, and the recognition and collection of the character parameter information will not cause recognition interference after the processing.

[0073] Step 23, after the image recognition is completed, the third target valve drawing file three-way channel is merged by using the merge() method of OpenCV, and the fourth target valve drawing is obtained.

[0074] That is, after the drawing image information recognition and removal, the drawing file does not need to use the three-way RGB (red, green and blue optical three primary colors) channel, the drawing file three-way channel is merged by using the merge() method of OpenCV in the present step, so as to facilitate the subsequent drawing pixel threshold processing.

[0075] Step 24, the merged fourth target valve drawing is subjected to bilateral filtering processing.

[0076] That is, the drawing file after the drawing pixel channel merging is subjected to bilateral filtering processing, the noise information in the drawing is removed, the detail layer side information is reduced, and the image smoothing processing work is completed under the premise of effectively protecting the image boundary information. Among them, the bilaterFilter() method of OpenCV is adopted, the pixel gravity center is positioned as the drawing gravity center pixel coordinate, and the color calculation range is set to 255.

[0077] Step 25, the drawing after the filtering processing is subjected to drawing pixel binarization threshold processing, so that the pixel information in the drawing is highlighted.

[0078] That is, the drawing file after the filtering processing is subjected to drawing pixel binarization threshold processing, so that the pixel information in the drawing is highlighted, forming a strong pixel contrast and differentiation effect, and the drawing pixel information is highlighted, which is helpful to improve the recognition effect in the subsequent information recognition.

[0079] Optionally, the threshold() method of OpenCV is adopted in the present application, with 200 as the threshold value, 255 as the maximum value, the pixel value of less than or equal to 200 pixel points being set to 0 for pure black display, and the pixel value of greater than 200 pixel points being set to 255 for pure white display, that is:

[0080]

[0081] After the threshold processing of the drawing pixel binarization, the pixel value of the to-be-recognized region of the drawing has obvious distinguishing effect, which will effectively improve the recognition effect of the drawing information.

[0082] Step 3, using the valve drawing information recognition model of CRNN to recognize the information of the to-be-recognized region of the second target valve drawing.

[0083] Referring to Figure 5 , the valve drawing information recognition model of CRNN is divided into CRNN convolution layer, CRNN cycle layer and CRNN transcription layer, which respectively complete the extraction of text features of to-be-recognized information, the extraction of sequence features and the generation of probability feature sequence, and the prediction of probability feature sequence and the conversion into character sequence.

[0084] Specifically, the CRNN convolution layer is used to extract the features of the to-be-recognized information of the input drawing, and the convolution feature map is obtained through convolution, pooling, normalization and other operations, which is used as the input of the CRNN cycle layer. The CRNN convolution layer uses convolutional neural network as the core tool for feature extraction, constructs 6 times of convolution operation, 4 times of maximum pooling operation, and performs 2 times of batch normalization operation, a total of 13 times of convolution layer operation, and the related operation parameters are shown in Table 1, wherein C is the channel number, k is the convolution kernel size, s is the convolution kernel step, p is the convolution padding size, w is the pooling kernel size, and the activation function uses rectified linear function ReLU (Rectified Linear Unit).

[0085] Table 1 CRNN convolution layer operation

[0086] Sequence number Computational type Hyperparameters Activation function 1 Convolution C:64, k:3x3, s:1, p:1 ReLU 2 Max pooling w:2x2, s:2 - 3 Convolution C:128, k:3x3, s:1, p:1 ReLU 4 Max pooling w:2x2, s:2 - 5 Convolution C:256, k:3x3, s:1, p:1 ReLU 6 Convolution C:256, k:3x3, s:1, p:1 ReLU 7 Max pooling w:1x2, s:2 - 8 Convolution C:512, k:3x3, s:1, p:1 ReLU 9 Batch normalization - - 10 Convolution C:512, k:3x3, s:1, p:1 ReLU 11 Batch normalization - - 12 Max pooling w:1x2, s:2 - 13 Convolution C:512, k:2x2, s:1, p:0 ReLU

[0087] It can be known that the operation purpose of convolution is feature detection, the input matrix of convolution operation represents the to-be-recognized image region, and the matrix elements correspond to the image pixels, representing the gray information. The convolution kernel slides on the image according to the step, and the convolution actually performs matrix inner product calculation, and the feature map is obtained through convolution calculation. The feature Figure 1 is generally a matrix or a tensor.

[0088] Pooling, also known as aggregation or downsampling, the purpose of the pooling operation is feature selection, each element of the input matrix of the pooling operation identifies a feature, representing the detection value of the feature. The pooling kernel slides on the image by a step, and the maximum value operation is performed on the matrix parameters covered by the sliding of the pooling kernel. Pooling will output a reduced matrix.

[0089] Batch normalization is a normalization operation on the net input of each layer of the neural network on each batch sample, which completes the feature scale transformation, can accelerate the learning convergence speed of the neural network, and to some extent prevent gradient disappearance and gradient explosion. Gradient disappearance and gradient explosion may exist in neural network gradient descent calculation, which will cause parameter stop updating and parameter overflow, and further cause model learning difficulty and calculation convergence failure; and the input value is normalized to keep it within a certain range, such as between 0 and 1, which can reduce the difference in value range, accelerate learning convergence, and suppress gradient disappearance and gradient explosion.

[0090] The activation function is an important part of the neural network, which is used to establish a non-linear relationship between the input and the output, so as to expand the learning ability and expression ability of the neural network. Actual problems are generally non-linear problems, and the activation function can effectively express the non-linear relationship. Commonly used activation functions include S-shaped function, hyperbolic tangent function, and rectified linear function. In this application, the rectified linear function ReLU is used. The derivative of ReLU is always 1 in the positive interval, which can avoid the problem of gradient explosion. At the same time, compared with other activation functions, the function is simple to calculate, and can maintain good sparse activation performance, improve the generalization ability of the model, and the function relationship is shown in the following formula.

[0091] f(x) = relu(x) = max(0, x);

[0092] After multiple convolution, pooling and normalization operations, the to-be-identified information extracts the character features of the drawing information to obtain a feature map as the input of the CRNN recurrent layer.

[0093] The CRNN recurrent layer adopts a bidirectional LSTM module to form a bidirectional long short-term memory neural network (BiLSTM, Bidirectional Long Short-term Memory Neural Network), which is used to extract the character features input by the CRNN convolution layer into sequence features and generate a probability feature sequence. BiLSTM is based on LSTM, and LSTM is based on traditional RNN model. By introducing a special gate structure, the transmission of input information is controlled, so that the network model can retain or discard information. The input gate structure of LSTM is used to update the information state of the unit; the forget gate structure is used to decide whether the current information is retained or discarded; the output gate structure is used to control the information output by the current unit and pass the information to the next LSTM unit.

[0094] The LSTM adopts a top-to-bottom sequence direction when processing information, and cannot fully utilize the context information in the character recognition work. The BiLSTM breaks the limitation of the one-way LSTM by adopting two opposite direction LSTMs, and the two-way LSTMs work independently without information interaction. Finally, the calculation results are fused and output, so that the context information can be obtained and processed at the same time when processing the drawing information, and the final prediction result of the LSTM is more accurate and reliable.

[0095] The CRNN cycle layer generates a probability feature sequence by adopting the BiLSTM, which is used as the input of the CRNN transcription layer.

[0096] The CRNN transcription layer adopts a connection temporal classification (CTC) loss function to convert the predicted probability feature sequence input by the CRNN cycle layer into a character sequence. The essence is to obtain the character sequence result with the highest probability combination in the probability feature matrix as the final information recognition result.

[0097] In the character recognition work, the prediction result sequence has no fixed length, and the number of characters corresponding to different features is inconsistent. The traditional mapping function method cannot solve the problem of indefinite length text recognition, and the variable sequence length text recognition is also one of the difficulties in the design of the character recognition model. In addition, the introduction of blank positions or repeated character information in the cycle layer processing will lead to distortion of the recognition result and reduce the recognition ability of the model. The CRNN adopts the CTC loss function, which can perform end-to-end training, omit the intermediate process, and only focus on the input and output ends of the data information, so that the recognition efficiency is effectively improved. At the same time, the CTC loss function introduces the blank character “-” mechanism, inserts a blank symbol “-” between the repeated information characters, and separates the redundant characters. Therefore, when the CRNN transcription layer converts the probability feature sequence into a character sequence, the same and continuous characters in the existing character sequence are not merged, but the repeated characters and blank symbols “-” are deleted first, which can effectively solve the problem of character repetition and redundancy in the character information recognition. This process is also called the decoding process.

[0098] The CRNN transcription layer removes the duplicate probability distribution sequence of the cycle layer and takes out the character with the maximum probability to form the final recognition result.

[0099] Step 4, using the valve drawing information recognition model of the CRNN to extract and collect the information of the recognition result in step 3, and classifying and integrating the collected information.

[0100] Reference Figure 6 Step 4 includes:

[0101] Step 41, receiving the identified result information in step 3, using the end-to-end identification mode of the valve drawing information recognition model of the CRNN to extract the drawing information.

[0102] Specifically, the application adopts a key feature matching mechanism, the key words come from the valve parameter information required to be stored in the valve drawing information database, and the units of the parameter information are used as identifiable key features, such as the units of temperature, pressure, and weight. Compared with other text information, the features are more prominent, which helps to accurately obtain correct parameter information. The key feature information is used as a screening key index for text comparison with the currently identified information. The information that matches successfully is used as the extracted information, and the invalid information is discarded. The information extraction is returned until all key features are used.

[0103] Step 42, classifying and integrating the extracted information.

[0104] Specifically, the drawing information obtained by the drawing information classification and integration submodule is classified, such as drawing source information, design parameter information, etc. A temporary evaluation file is generated to store the information integration results of the current submodule. The result file will be used for correctness judgment and evaluation by the information evaluation module.

[0105] Step 5, classifying and integrating the drawing, and then operating the correct and effective information into the valve drawing information database; wherein the valve drawing information database is associated with the valve drawing information recognition model of the CRNN.

[0106] Referring to Figure 7 , step 5 includes:

[0107] Step 51, receiving the classified and integrated drawing information in step 4, comparing the drawing information with the design key information library composed of valve design files, finding out the error information through comparison, and correcting the error information.

[0108] It can be seen that the comparison parameters used in the application come from the design key information library composed of valve design files, which saves the correct data of key information and can be used for parameter correctness comparison. In the parameter information judgment, if the judgment information is incorrect, the error information is fed back to the valve drawing information database for the data item, and the information is rewritten to correct the error problem; if the judgment information is correct, it is transferred to step 52.

[0109] Step 52, classifying the correct information to the valve drawing information database; wherein the valve drawing information database is associated with the valve drawing information recognition model of the CRNN.

[0110] That is, the main information required for information data warehousing includes drawing file attribution information (nuclear power project number, item number, file submission unit, data table number), valve design parameter information (nuclear safety classification, seismic grade, design temperature, design pressure, nominal size, valve pound level, main material, flow resistance parameter), valve envelope requirement information (valve weight, valve center of gravity), valve test parameter information (shell water pressure test, valve seat sealing test, closing test, packing seal test), and other information. For electric, pneumatic or electromagnetic valves, related information of electric devices, pneumatic devices, etc. is also required.

[0111] The warehoused information obtained thereby is not only associated with the data of the current drawing, but also ensures the correctness and effectiveness of the identified data information, so that the valve drawing information database will be used for further training and strengthening of the valve drawing information recognition model of the CRNN. After self-learning and self-strengthening, the valve drawing information recognition model of the CRNN based on the existing drawing information judgment, recognition, collection, evaluation and training will be more in line with the information recognition collection work requirements of the nuclear power project drawing, and the model work will be more targeted and reliable.

[0112] In summary, the present application can meet the requirements of "drawing processing-information recognition-information collection-data sorting-data warehousing-information correction-model strengthening" for nuclear valve drawing information processing, and can efficiently complete the nuclear valve drawing information recognition collection work.

[0113] Furthermore, the present application can be used to guide the self-learning and self-strengthening of the nuclear valve drawing information recognition model, realize the improvement of the model capability, provide technical support for the nuclear power station valve selection, drawing review, equipment design, system optimization, data verification and other work, and reduce the cost of nuclear valve drawing review, system design optimization and other work.

[0114] Reference Figure 8 As shown in the drawings, the present application also provides a collection system 1000, which includes a memory 1001 and a processor 1002. The memory can store instructions executable by the processor 1002. The processor 1002 can execute the instructions to implement the steps performed by the computer in the nuclear valve drawing information recognition collection method introduced in the above embodiments.

[0115] It can be understood that it should be noted that the above-mentioned memory and processor are not limited to a specific memory and processor. Furthermore, in the embodiment adopting a distributed structure, the specific execution terminal of each step can be adjusted according to the actual situation, and the specific scheme of implementing each step in a specific terminal should not limit the protection scope of the present application.

[0116] According to another aspect of the present application, the present application also provides a computer readable medium.

[0117] The computer readable medium provided by the present application has computer instructions thereon. The computer instructions, when executed by a processor, can implement the steps performed by the program in the method as introduced in the above embodiments.

[0118] According to yet another aspect of the present application, the present application also provides a computer program product.

[0119] The computer readable medium provided by the present application has computer instructions thereon. The computer instructions, when executed by a processor, can implement the steps performed by the program in the method as introduced in the above embodiments.

[0120] The various illustrative logical blocks, circuits, and circuitry described in connection with the embodiments disclosed herein can be implemented or performed with a general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. The general purpose processor can be a microprocessor, but in the alternative, the processor can be any conventional processor, controller, microcontroller, or state machine. A processor can also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration.

[0121] The steps of a method or algorithm described in connection with the embodiments disclosed herein can be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module can reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such that the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium can be integral to the processor. The processor and the storage medium can reside in an ASIC. The ASIC can reside in a user terminal. In the alternative, the processor and the storage medium can reside as discrete components in a user terminal.

[0122] In one or more exemplary embodiments, the functions described can be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functions can be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A storage media can be any available media that can be accessed by a computer. By way of example, and not limitation, such computer-readable media can comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. Also, any connection is properly termed a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. Disk and disc, as used herein, includes compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and blu-ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

[0123] The application has been disclosed with reference to preferred embodiments. However, it will be appreciated that those skilled in the art will be able to devise various arrangements which, although not explicitly described or shown herein, embody the principles of the application and are included within its spirit and scope. Therefore, to the extent that the application has been described with reference to specific embodiments, it is understood that it is covered by the claims and their equivalents.

Claims

1. A method for identifying and collecting information on a nuclear power valve drawing, characterized in that, The application relates to a valve drawing information recognition method based on a CRNN. The method comprises the following steps: step 1, obtaining a target valve drawing and pre-processing the target valve drawing in terms of geometric requirements and formats to obtain a first target valve drawing; step 2, pre-recognizing the information of the first target valve drawing by using a valve drawing information recognition model of the CRNN, sequentially removing the image information of the drawing, merging the pixel channels of the drawing, filtering and threshold processing the information of the drawing to obtain a second target valve drawing; step 3, recognizing the information of a to-be-recognized region of the second target valve drawing by using the valve drawing information recognition model of the CRNN; step 4, extracting and collecting the information of the recognition result in step 3 by using the valve drawing information recognition model of the CRNN, and classifying and integrating the collected information; and step 5, evaluating the correctness of the classified and integrated drawing, and then operating the correct and effective information into a valve drawing information database; wherein the valve drawing information database is associated with the valve drawing information recognition model of the CRNN. The step 1 comprises the following steps: step 11, reading the target valve drawing by using imread() of OpenCV; step 12, checking and judging the format of the target valve drawing, and converting the format of the target valve drawing which is not in JPG format into JPG format; and step 13, completing the drawing angle check and drawing pixel height transformation based on the pixel point coordinates by using pixel operation to obtain the first target valve drawing. The step 2 comprises the following steps: step 21, receiving the first target valve drawing, and performing pixel check and key valve recognition on the information of the first target valve drawing to obtain a third target valve drawing without abnormal pixel points; step 22, performing image information recognition on the third target valve drawing; step 23, after the image recognition is completed, merging the three-way channels of the third target valve drawing file by using the merge() method of OpenCV to obtain a fourth target valve drawing; step 24, performing bilateral filtering processing on the merged fourth target valve drawing; and step 25, adopting pixel binarization threshold processing on the drawing after the filtering processing to highlight the pixel information in the drawing. In the step 21, the coordinate pixels of the first target valve drawing are obtained, the abnormal pixel point region between 0 and 255 is introduced into the valve drawing information recognition model based on the CRNN for pre-recognition of key information; if there is no abnormal pixel point, the step 22 is entered; if there is an abnormal pixel point, the valve drawing information recognition model of the CRNN is used for key information recognition, error information is fed back to the valve drawing information database for the drawing which fails in key information recognition and belongs to a key valve, and the corresponding key information is invalidated for the drawing which fails in key information recognition and does not belong to a key valve. ​ 2. The nuclear power valve chart information recognition and collection method according to claim 1, characterized in that, ​ ​ ​ ​ 3. The nuclear power valve chart information recognition and collection method according to claim 1, characterized in that, ​ ​ ​ ​ ​ ​ 4. The nuclear power valve chart information recognition and collection method according to claim 3, characterized in that, ​ ​ 5. The nuclear power valve chart information recognition and collection method according to claim 3, characterized in that, In the step 22, the valve drawing image recognition model of the CNN is used to remove the image information that does not belong to the to-be-recognized collection in the third target valve drawing; wherein, the valve assembly drawing and the detailed design detail image information in the third target valve drawing are the image information that does not belong to the to-be-recognized collection.

6. The nuclear power valve chart information recognition and collection method according to claim 1, characterized in that, In the step 3, the valve drawing information recognition model of the CRNN includes a CRNN convolution layer, a CRNN cycle layer and a CRNN transcription layer; the CRNN convolution layer is used to complete the extraction of the text features of the to-be-recognized information; the CRNN cycle layer is used to extract the sequence features and generate the probability feature sequence; and the CRNN transcription layer is used to predict the probability feature sequence and convert it into a character sequence.

7. The nuclear power valve chart information recognition and collection method according to claim 1, characterized in that, The step 4 includes: In step 41, the identified result information in the step 3 is received, and the end-to-end recognition mode of the valve drawing information recognition model of the CRNN is used to extract the drawing information; In step 42, the extracted information is classified and integrated.

8. The nuclear power valve chart information recognition and collection method according to claim 1, characterized in that, The step 5 includes: In step 51, the classified and integrated drawing information in the step 4 is received, and the drawing information is compared with the design key information database composed of the valve design file, error information is found out through the comparison, and the error information is corrected; In step 52, the correctly checked information is sorted into the valve drawing information database; wherein, the valve drawing information database is associated with the valve drawing information recognition model of the CRNN.

9. A computer program product, characterised in that, A computer program is included, which is executed by a processor to realize the steps executed by a computer in the method of any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, A computer program is included, which is executed by a processor to realize the steps executed by a computer in the method of any one of claims 1-8.

11. A harvesting system characterized by, It includes: A memory capable of storing instructions executable by a processor; A processor capable of executing the instructions to realize the steps executed by a computer in the method of any one of claims 1-8.

Images