Railway construction concealed engineering acceptance system

Abstract

The present application provides a railway construction concealed engineering acceptance system, which collects the images of railway construction concealed engineering, identifies and analyzes the signboards and steel entities in the images based on computer vision, automatically manages the image data, evaluates and analyzes the quality of concealed engineering, and gives management suggestions. Through automatic processing and analysis of the images of railway project concealed engineering, the huge amount of acceptance image data analysis demand generated in long-term large-scale construction engineering can be met, the slow processing speed and non-standard quality analysis and evaluation caused by manual management and acceptance are avoided, the concealed engineering acceptance management efficiency is improved, and wider and higher-density acceptance is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of engineering management, and particularly relates to a railway construction concealed engineering acceptance system. BACKGROUND

[0002] In a railway engineering construction project, the quality control of concealed engineering has a very high management requirement. At present, for the acceptance management of concealed engineering in railway engineering construction, manual inspection, manual data entry and manual quality judgment are mostly adopted. Due to the long construction period and complex engineering of railway engineering project construction, manual management and acceptance are time-consuming and labor-consuming and are prone to errors.

[0003] In the process of railway engineering construction, each unit will collect a large amount of image data according to requirements. These image data are simply stored on the platform system and are not processed and analyzed. It is urgent to further process the engineering images, identify engineering features, extract key information and carry out deepening application.

[0004] In order to implement the lifelong responsibility system of engineering quality, ensure the quality of railway engineering construction and promote the standardized management of concealed engineering image data of railway engineering construction project, it is urgent to have a railway construction concealed engineering acceptance system to uniformly manage and evaluate the data generated in the concealed engineering acceptance project of railway construction. SUMMARY

[0005] In view of this, the present application provides a railway construction concealed engineering acceptance system to eliminate or improve one or more defects in the prior art, solve the problem that the prior art cannot effectively manage and analyze concealed engineering acceptance image data.

[0006] The present application provides a railway construction concealed engineering acceptance system, which comprises:

[0007] A concealed engineering image acquisition module is configured to acquire a first to-be-detected image of a railway construction concealed engineering according to a set standard based on a specified link or a storage block;

[0008] An image data management module is configured to receive the first to-be-detected image uploaded by the concealed engineering image acquisition module and classify and store the first to-be-detected image according to a set project classification standard;

[0009] An image quality detection module is configured to detect the quality of the first to-be-detected image stored in the image data management module according to the set standard, wherein the first to-be-detected image with unqualified quality is marked and the concealed engineering image acquisition module is instructed to reacquire the image;

[0010] An image preprocessing module is configured to preprocess the first image to be detected that is detected and output by the image quality detection module to obtain a second image to be detected, so as to meet the input requirements of subsequent image recognition and analysis.

[0011] An image content recognition module is configured to recognize and mark a position frame of a signboard in the second image to be detected based on a pre-trained signboard recognition model, and recognize and mark a position frame of a steel bar entity in the second image to be detected based on a pre-trained steel bar entity recognition model.

[0012] An element judgment module is configured to judge whether the second image to be detected contains a specified element according to the recognition results of the signboard and the steel bar entity by the image content recognition module, mark the second image to be detected that does not contain the specified element, and instruct the concealed engineering image acquisition module to re-acquire.

[0013] An image analysis module is configured to crop a signboard sub-image according to the position frame of the signboard marked by the image content recognition module from the second image to be detected processed by the image content recognition module, and perform signboard classification and marking based on a pre-trained signboard classification model; and perform image segmentation on the position frame of the steel bar entity to obtain a steel bar entity sub-image, and perform steel bar welding fullness detection and marking based on a pre-trained steel bar welding fullness recognition model.

[0014] A construction quality evaluation module is configured to perform quality evaluation according to the signboard classification result and the steel bar welding fullness detection result of the second image to be detected detected by the image analysis module, and generate a secondary acceptance instruction for unqualified images.

[0015] The image data management module is further configured to classify and store the second image to be detected that is evaluated as qualified by the construction quality evaluation module according to the set project classification standard.

[0016] In some embodiments, the set standard includes image acquisition of one or more target concealed engineering objects, and a first set number of the first images to be detected are acquired for each target concealed engineering object during the image acquisition process, the first images to be detected are shot according to set shooting angle requirements, shooting distance requirements, definition requirements, brightness requirements and frame size requirements, and time and acquisition personnel information are marked.

[0017] In some embodiments, the image data management module classifies and stores the first images to be detected according to the set project classification standard, including:

[0018] The first images to be detected are classified and stored in levels according to the project stage, the project main object and the sub-projects of the project main object of the railway construction concealed engineering.

[0019] In some embodiments, the image quality detection module performs quality detection on the first image to be detected stored in the image management module according to the set standard, including sharpness detection, noise detection, exposure detection, contrast detection, color saturation detection, compression quality detection, and frame size detection.

[0020] In some embodiments, the pre-training step of the signboard recognition model comprises:

[0021] obtaining a first training sample set containing a plurality of samples, each sample comprising a first hidden engineering sample image and a labeled position box of a signboard as a label;

[0022] obtaining a first initial neural network performing an image detection task, taking the first hidden engineering sample image of each sample as input and outputting the position box of the signboard; the first initial neural network is a convolutional neural network, a residual network, a recurrent neural network, or a U-Net network;

[0023] training the first initial neural network using the first training sample set to obtain the signboard recognition model.

[0024] In some embodiments, the pre-training step of the steel bar entity recognition model comprises:

[0025] obtaining a second training sample set containing a plurality of samples, each sample comprising a second hidden engineering sample image and a labeled position box of a steel bar entity as a label;

[0026] obtaining a second initial neural network performing an image detection task, taking the second hidden engineering sample image of each sample as input and outputting the position box of the steel bar entity; the second initial neural network is a convolutional neural network, a residual network, a recurrent neural network, or a U-Net network;

[0027] training the second initial neural network using the second training sample set to obtain the steel bar entity recognition model.

[0028] In some embodiments, the pre-training step of the signboard classification model comprises:

[0029] obtaining a third training sample set containing a plurality of samples, each sample comprising a sample signboard image and a labeled signboard category as a label; the signboard category is used to represent the project type of the hidden engineering, and each project type matches a set process standard;

[0030] obtain a third initial neural network, the third initial neural network performing a classification task, taking the sample identification plate image of each sample as input and outputting a class of identification plate, the third initial neural network being a convolutional neural network, a residual network, a recurrent neural network or a U-Net network;

[0031] train the third initial neural network using the third training sample set to obtain the identification plate classification model.

[0032] In some embodiments, the pre-training step of the steel bar welding fullness recognition model comprises:

[0033] obtain a fourth training sample set, the fourth training sample set containing a plurality of samples, each sample comprising a steel bar entity image and labeling steel bar saturation as a label;

[0034] obtain a fourth initial neural network, the fourth initial neural network taking the steel bar entity image of each sample as input and outputting steel bar saturation, the fourth initial neural network being a convolutional neural network, a residual network, a recurrent neural network or a U-Net network;

[0035] train the fourth initial neural network using the fourth training sample set to obtain the steel bar welding fullness recognition model.

[0036] In some embodiments, the image preprocessing module performs image scaling, image cropping, color space transformation, contrast enhancement, noise removal and / or normalization on the first to-be-detected image to obtain the second to-be-detected image.

[0037] In some embodiments, the image analysis module is further configured to query the process requirements of the concealed engineering according to the identification plate classification result in the second to-be-detected image, and label the process requirements in the second to-be-detected image.

[0038] The beneficial effects of the present application are at least:

[0039] The railway construction concealed engineering acceptance system of the present application collects images of railway construction concealed engineering, identifies and analyzes identification plates and steel bars in the images based on computer vision, automatically manages image data, evaluates and analyzes the quality of concealed engineering, and gives management suggestions. Through automatic processing and analysis of concealed engineering images of railway projects, the processing demand of huge amount of acceptance image data generated in long-term and large-scale construction engineering can be met, the slow processing speed and non-standard quality analysis and evaluation caused by human management and acceptance can be avoided, the efficiency of concealed engineering acceptance management is improved, and wider and higher density acceptance is realized.

[0040] Additional advantages, objects, and features of the application will be set forth in part by the description that follows, and will in part be apparent to those of ordinary skill in the art upon examination of the following or can be learned from practice of the application. The advantages and features of the application can be realized and obtained by means of the instrumentalities specifically pointed out in the appended claims.

[0041] Those skilled in the art will appreciate that the objects and advantages of the application can be accomplished by the BRIEF DESCRIPTION OF DRAWINGS

[0042] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description, serve to explain the principles of the application. In the drawings:

[0043] Figure 1 The structural schematic diagram of the railway construction concealed engineering acceptance system according to an embodiment of the present application.

[0044] Figure 2 The concealed engineering acceptance management implementation flowchart of the railway construction concealed engineering acceptance system according to an embodiment of the present application. DETAILED DESCRIPTION

[0045] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the embodiments and drawings. Herein, the illustrative embodiments of the present application and the descriptions thereof are used to explain the present application, but are not used to limit the present application.

[0046] Herein, it also needs to be explained that, in order to avoid the present application being obscured by unnecessary details, only the structures and / or processing steps closely related to the scheme according to the present application are shown in the drawings, and other details not closely related to the present application are omitted.

[0047] It should be emphasized that the terms "comprises / comprising" when used in this specification, specify the presence of stated features, elements, steps or components, but do not preclude the presence or addition of one or more other features, elements, steps or components.

[0048] Herein, it also needs to be explained that, if not specially stated, the term "connection" herein can not only mean direct connection, but also means indirect connection with an intermediate.

[0049] In the following, the embodiments of the present application will be described with reference to the drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.

[0050] At present, the acceptance management of concealed engineering in railway construction projects mostly relies on manual image collection, analysis and archiving, which makes the acceptance efficiency of concealed engineering low, the processing capacity small, and the quality analysis and evaluation more subjective. Therefore, the present application provides a railway construction concealed engineering acceptance system,

[0051] A set of computer vision processing-based specifications is established. After the image data is uploaded to the system, it can automatically identify whether the information such as signboards, concealed engineering entities and inspectors in the image is comprehensive, and whether the steel structure meets the requirements. It can quickly analyze and process the data within a specified time to determine whether the image data is qualified, and push the determination result to the data uploader. If the image data is unqualified, the corresponding image data will be prompted to be retaken.

[0052] It should be noted in advance that the first, second, third and the like in the present application are only used to distinguish different subjects, and are not limited to the order.

[0053] Specifically, the present application provides a railway construction concealed engineering acceptance system, which comprises a concealed engineering image collection module, an image data management module, an image quality detection module, an image preprocessing module, an image content recognition module, an element judgment module, an image analysis module and a construction quality evaluation module. The aforementioned modules can be electronic devices capable of loading and executing computer programs such as single-chip microcomputers and central processing units, or can be deployed to run on the cloud.

[0054] The concealed engineering image collection module is used to obtain first detection images of railway construction concealed engineering collected according to set standards based on a specified link or storage block. In some embodiments, the set standards include image collection of one or more target concealed engineering objects, and a first set number of first detection images of each target concealed engineering object are collected during the image collection process. The first detection images are taken according to set shooting angle requirements, shooting distance requirements, clarity requirements, brightness requirements and frame size requirements, and are marked with time and collection personnel information. For example, railway project engineering usually has a large time span and a wide range, and concealed engineering is divided according to different time periods and different locations. Each part needs to be sampled separately. Further, the preliminary collected concealed engineering images in the sampling process are first detection images. The present application can not only set standards for the number of collected objects and collected images, but also set standards for the images themselves to meet the needs of subsequent processing, for example, the sharpness, contrast, signal-to-noise ratio, etc. can be limited for clarity, the image can be required to reach a specified average brightness, and the minimum brightness and maximum brightness can be limited. The frame size can be limited by pixel size, scale size, etc. Further, standards can be proposed for the image collection process, for example, the concealed engineering is required to be taken under specific lighting conditions and within a specified distance range.

[0055] The image data management module is configured to receive the first to-be-detected image uploaded by the concealed engineering image collection module, and classify and store the first to-be-detected image according to the set project classification standard. In this embodiment, the image data management module can classify and store the first to-be-detected image according to the project stage, the project main object, and the sub-projects constituting the project main object of the railway construction concealed engineering.

[0056] The image quality detection module is configured to detect the quality of the first to-be-detected image stored in the image data management module according to the set standard, and mark and instruct the concealed engineering image collection module to re-collect the first to-be-detected image that does not meet the quality standard.

[0057] In some embodiments, the image quality detection module can detect the quality of the first to-be-detected image stored in the image data management module according to the set standard, which can include sharpness detection, noise detection, exposure detection, contrast detection, color saturation detection, compression quality detection, and frame size detection.

[0058] The sharpness detection is used to evaluate the definition and details of the image, and to determine whether the image is blurred or too sharp. The noise detection is used to detect the noise points, particles or other interfering elements in the image, and to determine whether the image contains noise. The exposure detection is used to evaluate the brightness level of the image, and to determine whether the image is overexposed (too bright) or underexposed (too dark). The contrast detection is used to evaluate the difference between colors and brightness in the image, and to determine whether the contrast of the image is appropriate. The color saturation detection is used to detect the saturation of colors in the image, and to determine whether the colors of the image are too bright or too dark. The compression quality detection is used to detect the quality loss of the compressed image, and to determine whether the image has obvious distortion or compression artifacts.

[0059] The limitations of image quality detection in this application are only partially listed, and it should be understood that other detection items can also be set according to the needs of actual application scenarios.

[0060] During the quality detection process, if the quality of the first to-be-detected image meets the requirements, it will enter the next stage of recognition and analysis processing, and if it does not meet the requirements, it will be instructed to re-collect. This screening can effectively avoid subsequent recognition errors and save computing power while ensuring processing and analysis effects.

[0061] The image preprocessing module is configured to preprocess the first to-be-detected image that has passed the quality detection of the image quality detection module to obtain a second to-be-detected image, so as to meet the input requirements of subsequent image recognition and analysis. In some embodiments, the image preprocessing module performs image scaling, image cropping, color space conversion, contrast enhancement, noise removal, and / or normalization on the first to-be-detected image to obtain the second to-be-detected image.

[0062] Image scaling is the adjustment of the size of an image, usually to fit the input requirements of a particular model or algorithm, or to save computational resources. Image cropping is to cut out the region of interest in the image as needed, remove irrelevant background or frame, and focus attention on the key area. Color space transformation is to convert an image from one color space to another, such as RGB to grayscale image, RGB to HSV, RGB to Lab, etc., in order to better handle the color information of the image. Contrast enhancement is to enhance the contrast in the image by histogram equalization, adaptive histogram equalization, contrast stretching, etc., so that the details in the image are clearer. Noise removal is to eliminate noise in the image using filters or noise reduction algorithms to improve image quality. Normalization is to map the pixel values of an image to a specific range, such as [0, 1] or [-1, 1], in order to better adapt to the input requirements of the model.

[0063] The image content recognition module identifies and marks the position frame of the sign in the second to-be-detected image based on a pre-trained sign recognition model, and identifies and marks the position frame of the steel bar entity in the second to-be-detected image based on a pre-trained steel bar entity recognition model.

[0064] The image content recognition module identifies the positions of the sign and the steel bar entity for subsequent analysis and processing.

[0065] In some embodiments, the pre-training steps of the sign recognition model include steps S101-S103:

[0066] Step S101: Obtain a first training sample set, which contains multiple samples, each sample including a first concealed engineering sample image and a labeled position frame of a sign as a label.

[0067] Step S102: Obtain a first initial neural network, which performs an image detection task, takes the first concealed engineering sample image of each sample as input, and outputs the position frame of the sign; the first initial neural network is a convolutional neural network, a residual network, a recurrent neural network, or a U-Net network.

[0068] Step S103: Train the first initial neural network using the first training sample set to obtain a sign recognition model.

[0069] In some embodiments, the pre-training steps of the steel bar entity recognition model include steps S201-S203:

[0070] Step S201: Obtain a second training sample set, which contains multiple samples, each sample including a second concealed engineering sample image and a labeled position frame of a steel bar entity as a label.

[0071] Step S202: Obtain a second initial neural network, the second initial neural network performs an image detection task, takes a second hidden engineering sample image of each sample as input and outputs a position box where a reinforcing bar entity is located; the second initial neural network is a convolutional neural network, a residual network, a recurrent neural network or a U-Net network.

[0072] Step S203: Train the second initial neural network by using a second training sample set to obtain a reinforcing bar entity recognition model.

[0073] The element judgment module is configured to determine whether the specified element is contained in the second detection image according to the identification results of the sign and the reinforcing bar entity by the image content recognition module, mark the second detection image which does not contain the specified element and instruct the hidden engineering image acquisition module to re-acquire. If no sign or reinforcing bar entity is identified in the detection of a hidden engineering image, it means that the necessary element required for acceptance is lacking, and the hidden engineering image needs to be re-acquired.

[0074] The image analysis module is configured to crop a sign sub-image according to the position box of the sign marked by the image content recognition module and perform sign classification and marking based on a pre-trained sign classification model, and perform image segmentation on the reinforcing bar entity according to the position box of the reinforcing bar entity to obtain a reinforcing bar entity sub-image, and perform reinforcing bar welding fullness detection and marking based on a pre-trained reinforcing bar welding fullness recognition model.

[0075] In some embodiments, the pre-training step of the sign classification model includes steps S301-S303:

[0076] Step S301: Obtain a third training sample set, the third training sample set contains a plurality of samples, each sample includes a sample sign image and labels a sign category as a label; the sign category is used to represent the project type of the hidden engineering, and each project type matches a set process standard.

[0077] Step S302: Obtain a third initial neural network, the third initial neural network performs a classification task, takes the sample sign image of each sample as input and outputs the category of the sign; the third initial neural network is a convolutional neural network, a residual network, a recurrent neural network or a U-Net network.

[0078] Step S303: Train the third initial neural network by using the third training sample set to obtain a sign classification model.

[0079] In some embodiments, the pre-training step of the reinforcing bar welding fullness recognition model includes steps S401-S403:

[0080] Step S401: Obtain a fourth training sample set, the fourth training sample set containing a plurality of samples, each sample including a steel bar entity image and labeling the steel bar saturation as a label.

[0081] Step S402: Obtain a fourth initial neural network, the fourth initial neural network taking the steel bar entity image of each sample as input and outputting the steel bar saturation; the fourth initial neural network is a convolutional neural network, a residual network, a recurrent neural network, or a U-Net network.

[0082] Step S403: Train the fourth initial neural network using the fourth training sample set to obtain a steel bar welding saturation recognition model.

[0083] The construction quality evaluation module is configured to perform quality evaluation according to the sign classification result and the steel bar welding saturation detection result in the second to-be-detected image detected by the image analysis module, and generate a secondary acceptance instruction for unqualified images.

[0084] The sign can be used to mark the specific content of the concealed engineering object, and accordingly, the construction process requirements can be retrieved according to the project object, which includes the requirement for the steel bar welding saturation. In the analysis process, based on the object type marked by the sign, the steel bar welding saturation requirement that should be met is queried, and by comparing the actual steel bar welding saturation obtained by identification, the monitoring result of whether it is qualified can be obtained.

[0085] In some embodiments, the image analysis module is further configured to query the process requirements of the concealed engineering according to the sign classification result in the second to-be-detected image, and mark the process requirements in the second to-be-detected image.

[0086] The image data management module is further configured to classify and store the second to-be-detected image that is evaluated as qualified by the construction quality evaluation module according to the set project classification standard.

[0087] In some embodiments, the image analysis module is further configured to query the process requirements of the concealed engineering according to the sign classification result in the second to-be-detected image, and mark the process requirements in the second to-be-detected image.

[0088] Further, the face recognition of the acceptance personnel is needed to judge the specific identity of the acceptance personnel, based on which the acceptance authority is queried, and only when the acceptance authority meets the requirements, the second to-be-detected image is continued to be processed, otherwise the image collection is reinitiated.

[0089] Specifically, to achieve the identity recognition of the acceptance personnel, the following solutions can be implemented: Data collection: Collect a large amount of image data containing the faces of acceptance personnel as a training set. Ensure that the image data comes from a wide range of sources, including acceptance personnel images of different shooting angles, shooting scenes, expressions, and lighting conditions.

[0090] Data preprocessing: Preprocess the collected image data. This includes image scaling, cropping, grayscale or color conversion, etc. to standardize the input data and improve the robustness of the model.

[0091] Data augmentation: Perform augmentation operations on image data to expand the diversity and quantity of the data set. Random rotation, translation, scaling, mirroring, etc. can be performed to increase the model's ability to recognize faces of different angles and scales.

[0092] Labeling data: Label the faces of the people in the image data containing the faces of the acceptance personnel. Key point labeling can be used to mark important feature points of the face, such as eyes, nose, mouth, etc. Face bounding box labeling can also be used to mark the location of the person's face, as well as the subject label of the acceptance personnel.

[0093] Model construction: Select a suitable computer vision model architecture, such as a convolutional neural network (CNN), or a deep learning model such as FaceNet, ArcFace, etc. These models can learn the feature representation of a face and have good recognition performance.

[0094] Model training: Train the model using the labeled data. Input the image into the model to perform face recognition training tasks. Use appropriate optimization algorithms such as stochastic gradient descent (SGD) during training, and set appropriate learning rates.

[0095] Model evaluation: Use the validation set to evaluate the performance of the model. Calculate the accuracy, recall, F1 value, etc. of the model on the face recognition task to evaluate the effectiveness of the model.

[0096] Model optimization: Based on the evaluation results, optimize and improve the model. You can try adjusting the model architecture, increasing the number of network layers, introducing regularization techniques, adjusting the loss function, etc.

[0097] Model deployment: In practical applications, deploy the trained model. Apply the model to new image data for acceptance personnel face recognition. Use GPU or specialized accelerators to improve inference speed.

[0098] Specifically, the railway construction concealed engineering acceptance system described in the application implements the following management steps for concealed engineering acceptance management:

[0099] 1) The concealed engineering image acquisition module acquires the first to-be-detected image according to the standard based on the terminal device deployed on the construction site, which can be automatic acquisition or manual uploading by the construction personnel.

[0100] 2) The influence data management module classifies and stores the first to-be-detected image acquired by the concealed engineering image acquisition module according to the time and source for subsequent processing.

[0101] 3) The image quality detection module detects the quality of the first to-be-detected image classified and stored by the influence data management module, and determines that the first to-be-detected image is unqualified and requires reacquisition for unqualified quality.

[0102] 4) For the first to-be-detected image of qualified quality, the image preprocessing module is handed over to process the second to-be-detected image obtained by image scaling, image cropping, color space transformation, contrast enhancement, noise removal and / or normalization.

[0103] 5) The image content recognition module recognizes the second to-be-detected image, which is mainly used to detect the positions of the signboard and the steel entity in the second to-be-detected image, and in other embodiments, the construction personnel can also be identified.

[0104] 6) On this basis, the element judgment module judges whether the second to-be-detected image contains the signboard and the steel entity that need to be analyzed according to the recognition result of the image content recognition module, and if they exist, further analysis is performed, and if they do not exist, it is indicated that the image of the related railway construction project concealed engineering is reacquired.

[0105] 7) The image analysis module analyzes the second to-be-detected image processed by the image content recognition module, crops the signboard sub-image according to the position frame of the signboard marker, and performs signboard classification and labeling based on the pre-trained signboard classification model; and crops the steel entity sub-image according to the position frame of the steel entity, and performs steel welding fullness detection and labeling based on the pre-trained steel welding fullness recognition model.

[0106] 8) The construction quality evaluation module, based on the analysis completed by the image analysis module, queries the construction standard of the corresponding concealed engineering object according to the identified signboard type in each image, and checks the identified steel welding fullness according to the standard to determine whether it is qualified. In other embodiments, it is also used to check the steel spacing, grid size and other items.

[0107] 9) The image data management module classifies and stores the second to-be-detected image evaluated by the construction quality evaluation module according to the set project classification standard.

[0108] The core of the embodiment is a computer vision-based concealed engineering image processing software, which forms a structured extraction technology of intelligently identifying personnel, signboard information, acceptance quality, acceptance conclusion and other parameters in the concealed engineering process acceptance based on image AI recognition and other machine vision algorithms.

[0109] By using an AI intelligent measurement method and an image feature point extraction algorithm, the steel bar spacing, steel bar grid size, steel bar welding fullness and the like are recognized, intelligent analysis and intelligent calculation are performed on the measurement results, the design parameters and the acceptance standards are automatically compared, and the steel bar process acceptance conclusion is formed.

[0110] This image recognition work actually mainly involves two aspects of method and data, the encoding-decoding and transformer modules which perform better in image classification are used on the method, which can optimize the recognition performance, or the old algorithms such as U-Net, resenet and deeplap can be used. As for the image data used for training, on the one hand, the images uploaded in the existing concealed engineering image management system are used, and on the other hand, the concealed engineering image acceptance data collected according to the specification requirements can be used.

[0111] In summary, the railway construction concealed engineering acceptance system disclosed in the present application can collect images of railway construction concealed engineering, identify and analyze the signboards and steel bars in the images based on computer vision, automatically manage image data, evaluate and analyze the quality of concealed engineering, and give management suggestions. Through automatic processing and analysis of the images of the concealed engineering of the railway project, the huge amount of image data processing demand generated in the long-term and large-scale construction engineering can be met, the slow processing speed and non-standard quality analysis and evaluation caused by manual management and acceptance are avoided, the concealed engineering acceptance management efficiency is improved, and the acceptance range is wider and the density is higher.

[0112] Those skilled in the art should understand that the exemplary components, systems and methods described in connection with the embodiments disclosed herein can be implemented in hardware, software or a combination thereof. Whether the implementation is in hardware or software depends on the specific application and design constraints imposed on the overall system. Skilled persons can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a functional card, etc. When implemented in software, the elements of the present application are program or code segments used to perform the required tasks. The program or code segments can be stored in a machine-readable medium or transmitted through a data signal carried in a carrier wave on a transmission medium or communication link.

[0113] It is to be expressly understood that the invention is not limited to the specific configurations and processes described above and illustrated in the drawings. For the sake of clarity, detailed descriptions of known methods are omitted. In the above embodiments, several specific steps are described and illustrated as examples. However, the method processes of the present invention are not limited to the specific steps described and illustrated, and various changes, modifications and additions can be made thereto by one of ordinary skill in the art without departing from the spirit of the present invention, and the order of the steps can be changed.

[0114] In the present invention, features described and / or illustrated with respect to one embodiment can be used in the same or a similar way in one or more other embodiments, and / or combined with or substituted for features of other embodiments.

[0115] The above description is only preferred embodiments of the present invention, and is not intended to limit the present invention. The embodiments of the present invention can be variously changed and modified by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of the present invention.

Claims

1. A railway construction concealed engineering acceptance system, characterized in that: The system comprises: A hidden engineering image acquisition module is used to acquire a first image to be detected of a hidden engineering project of railway construction according to set standards based on a designated link or storage block; An image data management module, configured to receive the first image to be detected uploaded by the concealed engineering image acquisition module, and classify and store the first image to be detected according to a set project classification standard; an image quality detection module for performing quality detection on the first image to be detected stored in the image data management module according to the set standard, wherein the first image to be detected with unqualified quality is marked and the concealed engineering image acquisition module is instructed to re-acquire the image; An image preprocessing module is used to preprocess the first image to be detected that has passed the detection and output by the image quality detection module to obtain a second image to be detected, so as to meet the input requirements of subsequent image recognition and analysis; An image content recognition module, wherein the image content recognition module recognizes the sign in the second image to be detected and marks a location frame based on a pre-trained sign recognition model, and recognizes the steel bar entity in the second image to be detected and marks a location frame based on a pre-trained steel bar entity recognition model; an element judgment module, configured to judge whether the second image to be detected contains a specified element based on the recognition results of the identification plate and the steel bar entity by the image content recognition module, mark the second detection image that does not contain the specified element, and instruct the hidden engineering image acquisition module to re-acquire the image; An image analysis module is configured to, from the second image to be detected processed by the image content recognition module, crop a sign sub-image based on the position frame of the sign mark, and classify and label the sign based on a pre-trained sign classification model; and, to obtain a steel bar entity sub-image based on the position frame of the steel bar entity, and detect and label the steel bar welding fullness based on a pre-trained steel bar welding fullness recognition model; a construction quality evaluation module, configured to perform quality evaluation based on the identification plate classification results and steel bar welding fullness detection results in the second image to be detected detected by the image analysis module, and generate a secondary acceptance instruction for unqualified ones; The image data management module is further configured to classify and store the second images to be detected that are evaluated as qualified by the construction quality evaluation module according to the set project classification standard.

2. The railway construction concealed engineering acceptance system according to claim 1 is characterized in that: The set standard includes image acquisition of one or more target hidden engineering objects. During the image acquisition process, a first set number of the first images to be detected are acquired for each target hidden engineering object. The first images to be detected are shot according to set shooting angle requirements, shooting distance requirements, clarity requirements, brightness requirements and frame size requirements, and the time and acquisition personnel information are marked.

3. The railway construction concealed engineering acceptance system according to claim 1 is characterized in that: The image data management module classifies and stores the first image to be detected according to a set project classification standard, including: The first to-be-detected images are hierarchically classified and stored according to the project phases, project main objects, and sub-projects constituting the project main objects of the railway construction concealed project.

4. The railway construction concealed engineering acceptance system according to claim 1, characterized in that: The image quality detection module performs quality detection on the first image to be detected stored in the image data management module according to the set standard, including: sharpness detection, noise detection, exposure detection, contrast detection, color saturation detection, compression quality detection and frame size detection.

5. The railway construction concealed engineering acceptance system according to claim 1 is characterized in that: The pre-training steps of the sign recognition model include: Obtain a first training sample set, wherein the first training sample set includes a plurality of samples, each sample including a first concealed engineering sample image and annotated with a signboard location frame as a label; Obtaining a first initial neural network, wherein the first initial neural network performs an image detection task, takes the first covert engineering sample image of each sample as input and outputs a location frame of the sign; the first initial neural network is a convolutional neural network, a residual network, a recurrent neural network, or a U-Net network; The first initial neural network is trained using the first training sample set to obtain the sign recognition model.

6. The railway construction concealed engineering acceptance system according to claim 1, characterized in that: The pre-training step of the steel bar entity recognition model includes: Obtain a second training sample set, where the second training sample set includes a plurality of samples, each sample including a second hidden engineering sample image and annotated with a location box of a steel bar entity as a label; Obtaining a second initial neural network, wherein the second initial neural network performs an image detection task, takes the second hidden engineering sample image of each sample as input and outputs a location box of the steel bar entity; the second initial neural network is a convolutional neural network, a residual network, a recurrent neural network, or a U-Net network; The second initial neural network is trained using the second training sample set to obtain the steel bar entity recognition model.

7. The railway construction concealed engineering acceptance system according to claim 1 is characterized in that: The pre-training steps of the sign classification model include: Obtain a third training sample set, the third training sample set comprising a plurality of samples, each sample comprising a sample signboard image and annotated with a signboard category as a label; the signboard category is used to characterize a project type of concealed engineering, and each project type matches a set process standard; Obtaining a third initial neural network, wherein the third initial neural network performs a classification task, takes the sample sign image of each sample as input and outputs the category of the sign; the third initial neural network is a convolutional neural network, a residual network, a recurrent neural network, or a U-Net network; The third initial neural network is trained using the third training sample set to obtain the sign classification model.

8. The railway construction concealed engineering acceptance system according to claim 1 is characterized in that: The pre-training steps of the steel bar welding fullness recognition model include: Obtaining a fourth training sample set, wherein the fourth training sample set includes a plurality of samples, each sample including an image of a steel bar entity and annotated with steel bar saturation as a label; Obtaining a fourth initial neural network, wherein the fourth initial neural network takes the steel bar entity image of each sample as input and outputs steel bar saturation; the fourth initial neural network is a convolutional neural network, a residual network, a recurrent neural network, or a U-Net network; The fourth initial neural network is trained using the fourth training sample set to obtain the steel bar welding fullness recognition model.

9. The railway construction concealed engineering acceptance system according to claim 1, characterized in that: The image preprocessing module performs image scaling, image cropping, color space transformation, contrast enhancement, noise removal and / or normalization on the first image to be detected to obtain the second image to be detected.

10. The railway construction concealed engineering acceptance system according to claim 1, characterized in that: The image analysis module is further configured to query the process requirements of the concealed project based on the classification result of the signboard in the second image to be detected, and mark the process requirements in the second image to be detected.

Images