Abnormal witness early warning method and system based on concrete test block condensation state recognition

Abstract

The invention discloses an abnormal witness early warning method and system based on concrete test block condensation state recognition, and the method comprises the steps: carrying out the rough recognition of an extracted test block entity, and classifying whether the surface of a concrete test block is dry or wet, fine identification is carried out on the test block entity with the classification result of surface wettability so as to classify whether the concrete test block is condensed or not, and a condensation state prediction result is obtained; wherein the test block entity is extracted from a shot concrete test block surface image; and according to the normal condensation state corresponding to the stage when the surface image of the concrete test block is shot, comparing the normal condensation state with the condensation state prediction result, and if the comparison result is inconsistent, sending out an abnormal witness early warning signal. According to the method, the condensation state of the concrete test block surface image shot in the sampling stage is quickly judged, whether the concrete test block is abnormal in the sampling stage or not is accurately detected according to the judgment result, and the detection efficiency is improved.

Description

Technical Field

[0001] The present application belongs to the technical field of sampling and identifying concrete test blocks at construction sites, and specifically relates to an abnormal witness warning method and system based on identification of the setting state of concrete test blocks. Background Art

[0002] With the continuous development of the modern construction industry, the quality of building materials plays a vital role in the safety and durability of buildings, and the quality of concrete test blocks is directly related to the safety of the main structure of the construction project. Concrete test blocks are made on-site by staff, and witnesses make witness records to record the information of concrete test blocks in the processes of sampling, sample preparation, labeling, sealing, inspection and on-site testing. In order to prevent the concrete test blocks from being replaced during the sampling stage, the test blocks prepared in advance after being sprinkled with water or slurry are used as witness sampling test blocks to impersonate the test blocks that should be made on-site. On-site personnel are required to judge whether the test blocks are in an unsolidified state to determine whether there is a possibility of abnormal production of the test blocks.

[0003] The existing abnormality identification method mainly takes a photo of the surface of the concrete sample during the sampling stage and enters the image into a QR code. The sampling personnel use the QR code to retrieve the sample information entered in advance to determine whether the test block is in an unset state. This sample state detection method can determine the concrete setting state without testing the inside of the sample, and the cost of taking photos is low, but it is too dependent on manual operation, inefficient, and requires high professional ability and professional quality of witnesses. It is highly subjective, has no unified standards, and is uncontrollable. Summary of the invention

[0004] The present application proposes an abnormal witness warning method and system based on the recognition of the setting state of concrete test blocks. The setting state is quickly judged on the surface image of the concrete test block taken during the sampling stage. The judgment result is used to accurately detect whether there is any witness abnormality in the concrete test block during the sampling stage, thereby improving the detection efficiency.

[0005] The first aspect of the present application provides an abnormal witness warning method based on concrete test block setting state recognition, the method comprising: The extracted test block entity is roughly identified to classify whether the surface of the concrete test block is dry or wet, and then the test block entity with a classification result of wet surface is precisely identified to classify whether the concrete test block is coagulated or uncoagulated, so as to obtain a coagulation state prediction result; wherein the test block entity is extracted from the photographed surface image of the concrete test block; The normal coagulation state corresponding to the stage when the surface image of the concrete test block is taken is compared with the coagulation state prediction result. If the comparison result is inconsistent, a witness abnormality warning signal is issued.

[0006] After the concrete test block is made, take a photo of the surface of the test block to obtain the surface image of the concrete test block; then, after a series of preprocessing on the surface image of the concrete test block, extract the complete and easily recognizable test block entity; perform rough recognition and fine recognition on the test block entity respectively. In the rough recognition, first judge whether the surface of the test block is dry or wet through the image. If the test block with a wet surface is further required to be judged whether it is coagulated or not coagulated inside through more delicate fine recognition, so as to realize more accurate and efficient detection of the quality of the concrete test block, without inserting the instrument into the test block for detection and without relying too much on manual judgment, improving the engineering quality detection efficiency and reducing the detection cost at the same time. Then, compare the obtained coagulation state prediction result with the normal coagulation state corresponding to the stage when the surface image of the concrete test block is taken. If the coagulation state of the concrete test block does not match the normal state that should be presented currently, it is considered that the test block may be replaced, and a witness anomaly warning signal will be sent to provide support for ensuring the engineering quality.

[0007] In a possible implementation method of the first aspect, perform rough recognition on the extracted test block entity to classify whether the surface of the concrete test block is dry or wet, and then perform fine recognition on the test block entity with the classification result of a wet surface to classify whether the concrete test block is coagulated or not coagulated, so as to obtain the coagulation state prediction result. Specifically: Extract features of the test block entity through the backbone network of the preset coagulation state recognition model to obtain the test block feature map; Perform rough recognition on the test block feature map through the rough recognition classifier of the coagulation state recognition model, and classify the test block entity as a dry surface test block or a wet surface test block; Perform fine recognition on the test block feature map classified as a wet surface test block through the feature extraction network based on the transformer structure and the fine recognition classifier of the coagulation state recognition model, and classify the test block entity classified as a wet surface test block as a coagulated water-sprinkled test block, a coagulated mortar-plastered test block or an uncoagulated test block; According to the classification result of the rough recognition and the classification result of the fine recognition, obtain the coagulation state prediction result of the surface image of the concrete test block.

[0008] In a possible implementation method of the first aspect, extract features of the test block entity through the backbone network of the preset coagulation state recognition model to obtain the test block feature map. Specifically: Perform downsampling on the test block entity through the convolution module and the inverted residual module of the backbone network, and then extract features from the downsampling result by stacking several hybrid spatial interaction modules of the backbone network to obtain the test block feature map; Among them, the hybrid spatial interaction module can be designed according to the actual feature extraction task.

[0009] The above solution can extract features from the downsampling results by stacking multiple hybrid spatial interaction modules designed based on actual feature extraction tasks, which can better capture the complex relationships between features, enhance the ability to extract the test block entity features in the picture, and obtain a test block feature map with richer representation.

[0010] In a possible implementation method of the first aspect, the coarse recognition classifier of the condensation state recognition model is used to coarsely recognize the test block feature map, and the test block entity is classified as a surface dry test block or a surface wet test block. Specifically: The test block entity corresponding to the test block feature map is classified as a surface dry test block or a surface wet test block, and the test block feature map corresponding to the surface wet test block is input into the feature extraction network and the fine recognition classifier for fine recognition.

[0011] The above solution initially classifies the test block entity into a surface wet test block and a surface dry test block through the coarse recognition classifier. Since the dryness and wetness of the test block surface are relatively easy to distinguish, through coarse recognition, the test block entities corresponding to the surface dry test blocks and the surface wet test blocks with obvious and easily distinguishable features are classified from the test block feature map first, quickly determining which concrete test blocks are dry and may have abnormal risks.

[0012] In a possible implementation method of the first aspect, the feature extraction network based on the transformer structure and the fine recognition classifier of the condensation state recognition model are used to finely recognize the test block feature map classified as a surface wet test block, and the test block entity classified as a surface wet test block is classified as a coagulated water-sprinkled test block, a coagulated grouted test block, or an uncoagulated test block. Specifically: The test block feature map classified as a surface wet test block is input into the convolutional module to adjust the number of image channels, and then input into the feature extraction network based on the transformer structure to extract local features and global features, obtaining a second feature map; The test block feature map, the feature map output by the convolutional module, and the second feature map are spliced to obtain a feature fusion map corresponding to the surface wet test block; The fine recognition classifier is used to finely recognize the feature fusion map corresponding to the surface wet test block, and the test block entity classified as a surface wet test block is classified as a coagulated water-sprinkled test block, a coagulated grouted test block, or an uncoagulated test block.

[0013] In the above solution, compared with rough recognition, fine recognition requires higher requirements for feature capture. Therefore, the features of the test block entity are more finely extracted first through the feature extraction network based on the transformer structure and the fine recognition classifier, and then the test block feature map, the feature map output by the convolutional module, and the second feature map are spliced together, and whether the test block has solidified is judged with high precision through the interaction and fusion of feature information. Because the solidified state and the non-solidified state have overlapping feature information and it is difficult to judge through surface detection, the fusion of multiple feature maps is adopted here to integrate all comprehensive information for judgment, and which concrete test blocks are non-solidified test blocks are more accurately identified through the combination of local features and all features.

[0014] In a possible implementation method of the first aspect, according to the classification results of rough recognition and fine recognition, the solidification state prediction result of the surface image of the concrete test block is obtained, specifically: According to the classification result of rough recognition, the concrete test block corresponding to the surface image of the concrete test block is classified as a surface dry test block or a surface wet test block; According to the classification result of fine recognition, the concrete test blocks belonging to the surface wet test blocks are classified as solidified water-sprinkled test blocks, solidified plastered test blocks, and non-solidified test blocks.

[0015] In the above solution, the test blocks are roughly divided into dry and wet ones according to the surface state of the test blocks through the rough recognition result first; then the surface wet test blocks are further subdivided into internally solidified and non-solidified ones through the fine recognition result, and the truly non-solidified test blocks are accurately identified.

[0016] In a possible implementation method of the first aspect, the correct solidification state corresponding to the stage when the surface image of the concrete test block is taken is compared with the solidification state prediction result. If the comparison result is inconsistent, a witness anomaly warning signal is issued, specifically: If the stage is the sampling stage, the normal solidification state is set as non-solidified test blocks; If the solidification state prediction result is non-solidified test blocks, the comparison result is consistent, and it is determined that there is no witness anomaly in the concrete test block corresponding to the solidification state prediction result during the sampling stage; If the solidification state prediction result is solidified test blocks, it means that there is a witness anomaly in the concrete test block corresponding to the solidification state prediction result during the sampling stage, and a witness anomaly warning signal is issued.

[0017] The above scheme first determines the characteristics of the concrete test block corresponding to the current detection stage, and then compares the predicted result of the setting state with the normal setting state corresponding to the stage when the surface image of the concrete test block is taken. If the comparison result is consistent, it means that the test block has not been replaced and there is no risk of abnormal production. If the comparison result is inconsistent, it means that the test block prepared in advance may be used as a witness sampling test block without treatment, watering or slurry treatment to impersonate the test block that should be made on site, and a test block abnormality warning signal is issued to warn.

[0018] In a possible implementation method of the first aspect, surface dry test blocks, condensed water test blocks and condensed mortar test blocks are all classified as condensed test blocks.

[0019] In a possible implementation method of the first aspect, the test block entity is extracted from a photographed surface image of the concrete test block, specifically: Taking photos of the surface of the current concrete test block to obtain a plurality of images of the surface of the concrete test block; Preprocessing the concrete test block surface image, and then inputting the preprocessed concrete test block surface image into an instance segmentation model for image segmentation to obtain a test block entity mask of each concrete test block surface image; The test block entity is extracted from the concrete test block surface image according to the position of the test block entity mask in the concrete test block surface image.

[0020] The second aspect of the present application provides an abnormal witness warning system based on concrete test block setting state recognition, the system comprising: a test block state prediction module and a state comparison module; The test block state prediction module is used to roughly identify the extracted test block entity to classify whether the surface of the concrete test block is dry or wet, and then accurately identify the test block entity with a classification result of surface wetness to classify the concrete test block as coagulated or uncoagulated, so as to obtain a coagulation state prediction result; wherein the test block entity is extracted from the photographed surface image of the concrete test block; The state comparison module is used to compare the correct coagulation state corresponding to the stage when the surface image of the concrete test block is taken with the coagulation state prediction result, and if the comparison result is inconsistent, a witness abnormality warning signal is issued. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the implementation manner will be briefly introduced below. Obviously, the drawings described below are only some implementation manners of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 It is a schematic diagram of the specific process of an abnormal witness warning method based on the identification of the setting state of concrete test blocks provided by an embodiment of the present application; Figure 2 It is an image cropping comparison diagram of an abnormal witness warning method based on the identification of the setting state of concrete test blocks provided by an embodiment of the present application; Figure 3 It is a surface diagram of a dry test block of an abnormal witness warning method based on the identification of the setting state of concrete test blocks provided by an embodiment of the present application; Figure 4 It is a diagram of a set - un - set test block of an abnormal witness warning method based on the identification of the setting state of concrete test blocks provided by an embodiment of the present application; Figure 5 It is a diagram of a concrete test block state recognition model of an abnormal witness warning method based on the identification of the setting state of concrete test blocks provided by an embodiment of the present application; Figure 6 It is a structure diagram of the basic module HSI Block of an abnormal witness warning method based on the identification of the setting state of concrete test blocks provided by an embodiment of the present application; Figure 7 It is a structure diagram of the output head module Head Block of an abnormal witness warning method based on the identification of the setting state of concrete test blocks provided by an embodiment of the present application; Figure 8 It is a structure diagram of an abnormal witness warning system based on the identification of the setting state of concrete test blocks provided by an embodiment of the present application. Detailed implementation manners

[0023] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0024] It should be understood that the step numbers used in the text are only for convenience of description and are not used to limit the execution order of the steps.

[0025] First embodiment In the field of engineering quality inspection, external markings, texture anti-counterfeiting, etc. are often used to compare the test block information during witness sampling and sample verification to determine whether the test block has been replaced. During the sampling stage, there may be test blocks made in advance that are not processed, sprinkled with water or slurried as test blocks for witness sampling, pretending to be test blocks that should be made on site. This behavior of passing inferior products off as good ones will affect the authenticity of the test results and further affect the quality of the construction project. Therefore, in order to improve the detection efficiency and accuracy of the witness sampling stage, the embodiment of the present application uses image recognition to quickly and accurately determine whether the concrete test block is in an unset state during the witness sampling stage, identify the test blocks in a set state, and issue an alarm signal in time.

[0026] like Figure 1 As shown, Figure 1 A specific flow chart of an abnormal witness warning method based on identification of the setting state of a concrete test block is provided for a certain embodiment of the present application. The abnormal witness warning method based on identification of the setting state of a concrete test block of the present embodiment includes steps S1 to S3, which are described in detail as follows: Step S1, roughly identifying the extracted test block entity to classify the surface of the concrete test block as dry or wet, and then accurately identifying the test block entity with a classification result of surface wetness to classify the concrete test block as solidified or unsolidified, to obtain a solidification state prediction result; wherein the test block entity is extracted from a photographed surface image of the concrete test block.

[0027] According to relevant regulations, after the concrete test blocks are made, witnesses are required to witness and record them. By recording sampling, sample preparation, labeling, sealing, inspection and on-site testing, it is ensured that the concrete test blocks will not be replaced or replaced with inferior ones, which will affect the quality of the building. During the sampling and inspection stages, witnesses will take photos of the concrete test blocks for record, and use the photos to determine whether the state of the test blocks at the current stage is normal. Because taking photos can be achieved only with a mobile phone, it is possible to verify abnormalities of the test blocks at a low cost, which can promote the development of the construction industry. Among them, during the sampling stage, the concrete test blocks should be in an unset state; during the inspection stage, the concrete test blocks should be in a set state.

[0028] After the concrete is vibrated and compacted, it is an unhardened, water-containing substance with a certain fluidity. Its surface state presents a wet gel state. During the curing process, the cement undergoes a hydration reaction, and the gel-like cement binds the bulk materials such as sand and gravel together, hardens during the curing process, and finally forms a dry and solid concrete block. In the embodiment of the present application, by learning the setting state characteristics of the concrete test block in the deep learning model and summarizing the laws of different setting states, the setting state of the test block can be identified conveniently, simply, directly, and efficiently. Moreover, compared with the prior art method of using a tool to penetrate the concrete test block for identification, taking photos can ensure the integrity of the test block and will not damage the test block.

[0029] If it is detected that the state of the test block in the image does not match the setting state corresponding to the stage, then there is obviously a possibility that the test block has abnormal production or postponed calculation age, and a warning signal will be sent to prompt the personnel to conduct further investigation.

[0030] The embodiment of the present application detects the concrete test block in the sampling stage. In the sampling stage, the concrete test block should be in an unhardened state. The surface of the produced concrete test block is photographed in the sampling stage to obtain multiple surface images of the concrete test block. First, the surface image of the concrete test block is preprocessed, mainly to correct the size of the image to make it conform to the input format required by the instance segmentation model.

[0031] Optionally, in other embodiments, abnormal condition witnessing detection can be performed on concrete test blocks in other stages, such as the curing stage, the storage-out stage, the submission-for-inspection stage, the sample-inspection stage, the compressive-strength stage, etc.

[0032] Specifically, the preprocessing is mainly to scale the surface image of the concrete test block proportionally to the size specified by the instance segmentation model, then convert it to the RGB format and perform standardization processing on the scaled image according to certain mean and variance values, and then convert the standardized image to the Tensor format.

[0033] Optionally, in the embodiment of the present application, the mean values set for image processing are 0.485, 0.456, 0.406; the variance values set for image processing are 0.229, 0.224, 0.225.

[0034] Input the surface image of the pre-processed concrete test block into the instance segmentation model for image segmentation to obtain the test block entity mask corresponding to each surface image of the concrete test block. Among them, if the test block entity mask does not border on the actual edge of the surface image of the concrete test block, it indicates that the test block is completely photographed and meets the requirements for subsequent anomaly detection; if the test block entity mask borders on one actual edge of the surface image of the concrete test block, or borders on two opposite edges of the surface image of the concrete test block, it indicates that it is a triple mold test block, which meets the requirements for subsequent anomaly detection.

[0035] Optionally, the instance segmentation model includes but is not limited to mask r-cnn, solov1, solov2, yolo series, yolact, etc.

[0036] Finally, perform a cropping operation on the test block entity mask. According to the position of the test block entity mask in the surface image of the concrete test block, remove the filling area that appears due to the pre-processing operation and the background captured during shooting in the surface image of the concrete test block, and extract the test block entity that completely shows the surface of the concrete test block.

[0037] To better demonstrate how the embodiment of the present application extracts the test block entity, Figure 2 An image cropping comparison diagram is provided. As shown in the figure, the left side is the surface image of the photographed concrete test block, and the right side is the test block entity extracted based on the surface image of the concrete test block. Compared with the left figure, the extra background around the right figure is removed, which can more completely show the surface of the concrete test block and will not cause interference to the subsequent setting state detection by the extra background area. In addition, Figure 2 The mosaic part in the middle of the concrete test block is the QR code identification of the concrete test block.

[0038] In the embodiment of the present application, by collecting and studying the image data of the entire life cycle of the concrete test block from just being made, cured to being sent for inspection and sampling, these data are input into the deep learning model to observe the different states and characteristics of the test block from not setting to initial setting and then to being set, so that the model can accurately distinguish the dry and wet characteristics of the surface of the concrete test block.

[0039] Because the dry and wet characteristics of the surface are relatively easy to distinguish, and it is more difficult to distinguish whether the inside of the test block has set, so in the embodiment of the present application, the concrete test block is first divided into a surface dry test block and a surface wet test block based on the test block entity. The surface dry test block is the set state test block, and there are various situations in the surface wet test block, which may be an unset test block, a set and watered test block, a set and plastered test block, so it is necessary to further distinguish to identify the unset test block.

[0040] Based on the setting state, the concrete test blocks are divided into surface dry test blocks, set water test blocks, set mortar test blocks and unset test blocks. Among them, surface dry test blocks, set water test blocks and set mortar test blocks are all classified as set test blocks.

[0041] The main characteristics of surface dry test blocks are that the surface of the test blocks generally has cracks and voids, the edges are generally broken, there are cracks between the edges and the mold, there are generally sand or dry crushed concrete particles on the surface, it is generally grayish white, and there may be writing or engraving. Figure 3 The surface image of the dry test block is provided. In the image, the concrete test block has no reflection and has writing on it. There are sand and granular objects on the surface. There are cracks and holes on the edge of the test block, indicating that it is a surface dry test block. In addition, Figure 3 The mosaic part in the middle of the concrete test block is the QR code logo of the concrete test block.

[0042] The main characteristics of the condensed water test block are that the surface of the test block is generally dark gray and there may be reflections on the surface. The other characteristics are similar to those of the surface dry test block.

[0043] The main characteristics of the solidified slurry test block are that there may be a layer of turbid gelatinous cement slurry on the surface of the test block, which is in the shape of a flat spread, but it can be seen that there is a grayish-white solidified test block under the cement slurry; the original holes and cracks on the surface of the test block are filled with mud, and the area may appear darker in color; the cement slurry is generally reflective, and the edges are generally broken or there are cracks with the mold.

[0044] Both the condensed water-sprayed test blocks and the condensed slurry test blocks belong to the condensed test blocks. The characteristics of the two are overlapping. For example, they are dry in themselves but look wet on the surface; the water / wet areas on the surface of the test blocks will reflect light. The difference between the two is: (1) The water on the water-sprayed test block is clear and transparent, and does not affect the observation of the test block surface. Whether it is holes, cracks or ups and downs of wrinkles, they can be clearly seen; but the cement slurry on the slurry test block is turbid, covering most of the test block surface, and almost no holes and cracks can be seen on the surface; (2) The slurry of the test block is usually uneven, with more mud in some areas and less or even no mud in other areas. The color of the mud area and the surface area of the test block are generally quite different, and the color distribution is uneven; (3) The slurry test block will generally have traces of the smoothing action, leaving traces or flat areas around the chip on the surface of the test block; (4) The test block for slurry is prepared in advance, and the surface of the test block is flush with the mold. Therefore, the amount of cement slurry used for slurry is small, very thin, and generally does not contain high sand (otherwise it will protrude and not meet the sampling requirements). The appearance of the slurry is very different from that of the normal uncured test block surface; (5)The cracks between the test block and the mold can be clearly seen in the water-sprayed test block, while the plastered test block is a gully or height difference filled with slurry. (6)Generally, the mold of the plastered test block is plastered with slurry, and the overflow spreads to the mold.

[0045] The above differences can be used as the distinguishing features between the solidified water-sprayed test block and the solidified plastered test block for the solidification state recognition model to learn, so as to accurately distinguish the non-solidified test block, the solidified water-sprayed test block, and the solidified plastered test block.

[0046] The main characteristics of the non-solidified test block are as follows: If there is more water on the surface of the test block, generally the test block has just been made, the water is turbid and colloidal, generally there are no holes, no cracks but there are bubbles, and it is easy to produce reflection, generally dark gray or bright yellow; if there is less water, generally the test block is in the initial setting state, generally there is less reflection, there are large holes but few in number, there are cracks but few in number, and there may be dry crushed concrete particles, etc., generally dark gray or grayish white or bright yellow.

[0047] Figure 4 The figure of the solidified-non-solidified test block is provided. The first row in the figure is the surface figure of the solidified water-sprayed test block. It can be seen that the surface has a certain reflection and the overall color is relatively dark, and the surface is relatively flat and clear; the second row is the surface figure of the solidified plastered test block. It can be seen that the slurry is unevenly distributed and part of the slurry overflows to the mold; the third row is the surface figure of the solidified test block. It can be seen that there is more water on the surface, which is turbid and colloidal, and the slurry is also evenly distributed and easy to produce reflection. In addition, Figure 4 The mosaic part in the middle of the concrete test block is the QR code identification of the concrete test block.

[0048] According to the above characteristic descriptions, the surface dry test block has obvious characteristics, is relatively easy to identify and has a large quantity, while the surface wet test block has a wide range of covering situations and is more difficult to identify. Therefore, in view of this situation, the embodiment of this application designs a two-stage concrete test block solidification state recognition strategy of rough recognition + fine recognition. The test block entity is input into the solidification state recognition model. First, through rough recognition, the surface dry test block and the surface wet test block are distinguished, and then through fine recognition, the non-solidified test block is distinguished from the surface wet test block.

[0049] The specific structure diagram of the solidification state recognition model provided by the embodiment of this application is as Figure 5 shown. Backone is the backbone network. The input test block entity extracts the test block feature map through the backbone network. The test block feature map is then initially classified into the surface dry test block and the surface wet test block through rough recognition by FirstStep. The test block feature map is input into Second Step for fine recognition to obtain the solidification state prediction result of each test block entity.

[0050] Among them, the main function of the backbone network is to extract the features of the test block, which is mainly composed of a convolutional module Conv Block, an inverted residual module Inverted Residual Block, a hybrid spatial interaction module (HybridSpatial Interaction Block, abbreviated as HSI Block) independently designed according to the task characteristics, and an output head module Head Block. The backbone network extracts features from the test block entity, and the process of feature extraction is as follows: 1. Perform downsampling operations through the convolutional module and the inverted residual module; 2. Sufficiently extract features by stacking multiple hybrid spatial interaction modules HSI Block. Among them, the HSI Block uses 5*5 and 7*7 depthwise separable convolutions to implement medium- and long-distance spatial modeling respectively, and then uses dot product operations to achieve high-order spatial interaction, so as to better capture the complex relationships between features, enhance the ability to extract the test block entity features in the picture, and then perform residual splicing to stack new features on the original features to obtain a richer representation.

[0051] Among them, the structure of the HSI Block is as Figure 6 shown. After passing the output feature map of the previous module through a 1*1 convolution, the number of channels is expanded to 3 times, and then it is split into three parts. The first part is processed by sigmoid after passing through a 5*5 depthwise separable convolution, and then directly performs a dot product operation with the second part; after passing the result of the dot product through a 1*1 convolution, it also performs a dot product operation with the third part that has passed through a 7*7 depthwise separable convolution and sigmoid processing; the result of the dot product is concatenated with the output of the previous module that has passed through a 1*1 convolution to obtain a feature map with 2 times the number of channels; finally, after passing through a 1*1 convolution and an SE attention module, the final output is obtained. In addition, Figure 6 CBA is the structure of convolutional layer-batch normalization layer-activation function, full name Convolution-BatchNorm-Activation, which is jointly composed of a convolutional layer, a batch normalization layer and an activation function; DBA is the structure of depthwise separable convolutional layer-batch normalization layer-activation function, full name Deconvolution-BatchNorm-Activation, which is jointly composed of a depthwise separable convolutional layer, a batch normalization layer and an activation function.

[0052] 3. Finally, after passing the output result of the previous part through a convolution operation of the output of the second inverted residual structure, it is concatenated with the output after passing through 4 HSI Blocks, so as to fuse the low-dimensional basic and general feature information with the high-dimensional abstract and complex representation, improve the model performance, and output the final test block feature map.

[0053] Next, the obtained test block feature map will be roughly recognized in the first stage through the Head Block, and the surface dry test blocks with obvious features and easy to distinguish will be classified first.

[0054] Specifically, the rough recognition is realized through a rough recognition classifier, which mainly includes the HeadBlock and can simply classify the test block entity through the test block feature map, and recognize the test block feature map corresponding to the surface dry test block and the test block feature map corresponding to the surface wet test block. Among them, the structure of the Head Block of the rough recognition classifier is as Figure 7 shown.

[0055] Based on the previous description of the surface dry test block, the rough recognition classifier can complete the classification by recognizing whether the surface of the concrete test block is reflective, whether the test block is demolded, and whether there are cracks between the test block and the mold.

[0056] According to the output result of the Head Block of the rough recognition classifier, the test block feature maps representing the surface wet test blocks are distinguished, and these feature maps are input into the second stage for fine recognition.

[0057] First, the test block feature map classified as the surface wet test block is input into the convolution module to adjust the number of image channels, and then the feature map output by the convolution module is input into the feature extraction network based on the transformer structure to extract local features and global features, obtaining the second feature map. Among them, the second stage mainly consists of a convolution module, a feature extraction network based on the transformer structure, and an output head (i.e., a fine recognition classifier). The convolution module can adjust the feature map channels, while the transformer module has the ability of long-distance spatial modeling and the characteristics of high-order spatial interaction, and can well capture the local features and global features of the test block image, realizing the accurate classification of the difficult-to-distinguish categories of uncoagulated and coagulated water-sprayed and plastered test blocks.

[0058] Specifically, first, the test block feature map output by the backbone network is first passed through a convolution module to change the number of channels so that the feature map meets the channel number requirements of the transformer module. Then, the transformer module first performs an unfolding operation on the feature map output by the convolution module, regarding the pixel points in the k*k region as a token, changing the shape of the test block feature map from (B, C, H, W) to (B, C*k*k, H / k, W / k), then using a 1*1 convolution to modify the channels to C, and then transforming and merging the dimensions. At this time, the shape of the test block feature map is (B, H / k*W / k, C); it is input into the self-attention transformer structure for calculation to perform global information interaction and integration; then, a folding operation is performed on the test block feature map. After dimension transformation and 1*1 convolution opposite to the unfolding operation, the shape of the test block feature map is changed from (B, H / k*W / k, C) to the original shape (B, C, H, W), and finally, the constructed second feature map is output.

[0059] Then, the test block feature map output by the backbone network, the feature map output by the convolution module, and the second feature map output by the feature extraction network based on the transformer structure are spliced and then input into the convolution module for feature information interaction and fusion, which not only includes the original information extracted by the backbone network but also encompasses the long-distance spatial interaction information and local feature information of the test block features, fully integrating the overall information for classification operations to obtain the feature fusion map corresponding to the surface wet test block.

[0060] Finally, the feature fusion map is precisely recognized by the precise recognition classifier to recognize the non-set test block, the set water-sprinkled test block, and the set plastered test block. Among them, precise recognition is performed on the color of the concrete test block surface, the shape of the slurry on the test block surface, the number and size of the holes on the test block surface, and the number of bubbles on the test block surface, etc.

[0061] Therefore, precise recognition mainly uses the transformer structure to extract and integrate global feature information on the basis of the rough recognition CNN convolutional neural network.

[0062] Step S2, compare the correct setting state corresponding to the stage when the surface image of the concrete test block is taken with the setting state prediction result. If the comparison result is inconsistent, a witness anomaly warning signal is issued.

[0063] In the embodiment of the present application, according to the setting state prediction result obtained from the setting state recognition model, the concrete test blocks are classified into surface dry test blocks, set water-sprinkled test blocks, set plastered test blocks, and set test blocks, where the surface dry test blocks, set water-sprinkled test blocks, and set plastered test blocks are combined into set test blocks.

[0064] For the sampling stage, the corresponding normal test block state should be the non-set state. Therefore, by determining whether the predicted result of the setting state is the non-set state, it is determined whether the setting state of the concrete test block in the current concrete test block surface image is normal. Specifically, there are the following two cases: 1. When the concrete test block in the concrete test block surface image taken at the sampling stage is in the set state, it indicates that during witnessing, pre-made test blocks may be used without treatment, treated with water spraying or plastering, or used as witnessed sampling test blocks after delaying at least one day after on-site sampling. An abnormal warning signal for the test block should be issued; 2. When the concrete test block in the concrete test block surface image taken at the sampling stage is in the non-set state, it indicates that it conforms to the normal state of the sampling stage, and the predicted result of the setting state is returned.

[0065] Implementing the embodiments of the present application has the following beneficial effects: In the embodiments of the present application, after the concrete test block is made, the surface of the test block is photographed to obtain the concrete test block surface image; then the concrete test block surface image is first segmented to delete the background, filling area, etc. that are irrelevant to the prediction or interfere with the prediction in the image, and the complete and easily recognizable test block entity is extracted; the test block entity is respectively subjected to rough recognition and fine recognition. In the rough recognition, first, it is judged whether the surface of the test block is dry or wet through the image. If the surface of the test block is wet, it is also necessary to judge whether the inside of the test block is set or not through more refined fine recognition, so as to realize more accurate and efficient detection of the quality of the concrete test block. There is no need to insert the instrument into the test block for detection and it does not rely too much on manual judgment, which improves the engineering quality detection efficiency and reduces the detection cost at the same time. Then, the predicted result of the setting state obtained is compared with the normal setting state corresponding to the stage when the concrete test block surface image is taken. If the setting state of the concrete test block does not conform to the normal state that should be presented currently, it is considered that the test block may be replaced, and a witness abnormality warning signal will be issued to provide support for ensuring the engineering quality.

[0066] Second Embodiment Furthermore, in order to implement the abnormal witness warning system based on the recognition of the setting state of the concrete test block corresponding to the above method embodiments to achieve the corresponding functions and technical effects, Figure 8 A structural diagram of an abnormal witness warning system based on the recognition of the setting state of the concrete test block is provided. For the convenience of description, only the parts related to this embodiment are shown. The abnormal witness warning system based on the recognition of the setting state of the concrete test block provided by the embodiments of the present application includes: The test block status prediction module 201 is used to roughly identify the extracted test block entities to classify whether the surface of the concrete test block is dry or wet, and then finely identify the test block entities with a classified result of a wet surface to classify whether the concrete test block has set or not, so as to obtain a setting status prediction result; wherein, the test block entities are extracted from the captured surface image of the concrete test block.

[0067] The status comparison module 202 is used to compare the normal setting status corresponding to the stage when the surface image of the concrete test block is captured with the setting status prediction result. If the comparison result is inconsistent, a witness anomaly warning signal is issued.

[0068] In some embodiments, the test block status prediction module 201 is specifically: According to relevant regulations, after the concrete test block is made, it is required that the witness personnel record. By recording sampling, sample preparation, identification, sealing, sending for inspection, and on-site inspection, etc., it is ensured that operations such as replacement or substitution of the concrete test block with inferior products that affect the building quality will not occur. During the sampling stage and the sample inspection stage, the witness personnel will take pictures of the concrete test block for record, and judge whether the status of the test block is normal at the current stage through the photos. Since taking pictures can be achieved only by a mobile phone, the abnormal verification of the test block can be carried out at low cost, which can promote the development of the construction engineering industry. Among them, during the sampling stage, the concrete test block should be in an unset state; during the sample inspection stage, the concrete test block should be in a set state.

[0069] After the concrete is vibrated and compacted, it is an unhardened, water-containing substance with a certain fluidity, and its surface state presents a wet gel state. During the curing process, the cement undergoes a hydration reaction, and the gel-state cement cements the bulk materials such as sand and gravel together, hardens during the curing process, and finally forms a dry and firm concrete block on the surface. In the embodiment of the present application, by learning the setting state characteristics of the concrete test block in the deep learning model and summarizing the laws of different setting states, the setting state of the test block can be conveniently, simply, directly, and efficiently identified. Moreover, compared with the prior art method of using a tool to penetrate into the concrete test block for identification, taking pictures can ensure the integrity of the test block and will not damage the test block.

[0070] If it is detected that the status of the test block in the image does not match the setting status corresponding to the stage, then there is obviously a possibility that the test block has abnormal production or postponed calculation of the age, and a warning signal will be issued to prompt the personnel to conduct further investigation.

[0071] The embodiments of this application detect concrete test blocks in the sampling stage. During the sampling stage, the concrete test blocks should be in an unhardened state. During the sampling stage, the surfaces of the fabricated concrete test blocks are photographed to obtain multiple surface images of the concrete test blocks. First, preprocess the surface images of the concrete test blocks, mainly by correcting the image size to conform to the input format required by the instance segmentation model.

[0072] Optionally, in other embodiments, abnormal witness detection can be performed on concrete test blocks in other stages, such as the curing stage, the storage-out stage, the submission-for-inspection stage, the sample-verification stage, the compressive-strength stage, etc.

[0073] Specifically, the preprocessing mainly scales the surface images of the concrete test blocks proportionally to the size specified by the instance segmentation model, then converts them to the RGB format and normalizes the scaled images according to a certain mean and variance, and finally converts the normalized images to the Tensor format.

[0074] Optionally, in the embodiments of this application, the means set for image processing are 0.485, 0.456, 0.406; the variances set for image processing are 0.229, 0.224, 0.225.

[0075] Input the preprocessed surface images of the concrete test blocks into the instance segmentation model for image segmentation to obtain the test-block entity masks corresponding to each surface image of the concrete test blocks. Among them, if the test-block entity mask does not border on the actual edge of the surface image of the concrete test block, it means that the test block is completely photographed and meets the requirements for subsequent abnormal detection; if the test-block entity mask borders on one actual edge of the surface image of the concrete test block, or borders on the two opposite edges of the surface image of the concrete test block, it indicates that it is a triple-mold test block and meets the requirements for subsequent abnormal detection.

[0076] Optionally, the instance segmentation model includes but is not limited to mask r-cnn, solov1, solov2, the yolo series, yolact, etc.

[0077] Finally, perform a cropping operation on the test-block entity mask. According to the position of the test-block entity mask in the surface image of the concrete test block, remove the padding areas that appear due to the preprocessing operation and the background captured during shooting in the surface image of the concrete test block, and extract the test-block entity that completely shows the surface of the concrete test block.

[0078] By collecting and studying the image data of the entire life cycle of concrete test blocks from just being fabricated, cured to being submitted for inspection and sample verification, and inputting this data into a deep learning model to observe the different states and characteristics of the test blocks from unhardened to initial setting and then to hardened, so that the model can accurately distinguish the dry-wet characteristics of the surface of the concrete test blocks.

[0079] Because the dry and wet characteristics of the surface are relatively easy to distinguish, and whether the inside of the test block is condensed is more difficult to distinguish, the embodiment of the present application first divides the concrete test blocks into surface dry test blocks and surface wet test blocks based on the test block entity. The surface dry test blocks are test blocks in a condensed state, while there are many situations in the surface wet test blocks, which may be uncondensed test blocks, condensed water test blocks, and condensed slurry test blocks, so it is necessary to further distinguish and identify the uncondensed test blocks.

[0080] Based on the setting state, the concrete test blocks are divided into surface dry test blocks, set water test blocks, set mortar test blocks and unset test blocks. Among them, surface dry test blocks, set water test blocks and set mortar test blocks are all classified as set test blocks.

[0081] The main characteristics of surface dry test blocks are that the surface of the test blocks generally has cracks and voids, the edges are generally broken, there are cracks between the edges and the mold, there are generally sand or dry crushed concrete particles on the surface, it is generally grayish white, and there may be writing or engraving.

[0082] The main characteristics of the condensed water test block are that the surface of the test block is generally dark gray and there may be reflections on the surface. The other characteristics are similar to those of the surface dry test block.

[0083] The main characteristics of the solidified slurry test block are that there may be a layer of turbid gelatinous cement slurry on the surface of the test block, which is in the shape of a flat spread, but it can be seen that there is a grayish-white solidified test block under the cement slurry; the original holes and cracks on the surface of the test block are filled with mud, and the area may appear darker in color; the cement slurry is generally reflective, and the edges are generally broken or there are cracks with the mold.

[0084] Both the condensed water-sprayed test blocks and the condensed slurry test blocks belong to the condensed test blocks. The characteristics of the two are overlapping. For example, they are dry in themselves but look wet on the surface; the water / wet areas on the surface of the test blocks will reflect light. The difference between the two is: (1) The water on the water-sprayed test block is clear and transparent, and does not affect the observation of the test block surface. Whether it is holes, cracks or ups and downs of wrinkles, they can be clearly seen; but the cement slurry on the slurry test block is turbid, covering most of the test block surface, and almost no holes and cracks can be seen on the surface; (2) The slurry of the test block is usually uneven, with more mud in some areas and less or even no mud in other areas. The color of the mud area and the surface area of the test block are generally quite different, and the color distribution is uneven; (3) The slurry test block will generally have traces of the smoothing action, leaving traces or flat areas around the chip on the surface of the test block; (4)The plastering test block was made in advance, and the surface of the test block is already flush with the mold. Therefore, the amount of cement mortar used for plastering is small, very thin, and generally has a low sand content (otherwise it will protrude and does not meet the sampling requirements), and there is a large difference in the appearance of the mortar on the surface of the normal non-set test block; (5)For the water-sprayed test block, the crack between the test block and the mold can be clearly seen. For the plastering test block, there is a gully or height difference filled with mud; (6)Generally, the mold of the plastering test block will be plastered with mortar, and the mortar overflows onto the mold.

[0085] The above-mentioned distinguishing points can be used as the distinguishing features between the set water-sprayed test block and the set plastering test block for the set state recognition model to learn, so as to accurately distinguish the non-set test block, the set water-sprayed test block, and the set plastering test block.

[0086] The main characteristics of the non-set test block are as follows: if there is more water on the surface of the test block, generally the test block has just been made, the water is turbid and in a colloidal state, generally there are no holes, no cracks but there are bubbles, and it is easy to produce reflection, generally in dark gray or bright yellow; if there is less water, generally the test block is in the initial setting state, generally there is less reflection, there are large holes but the number is small, there are cracks but the number is small, and there may be dry crushed concrete particles, etc., generally in dark gray or grayish white or bright yellow.

[0087] According to the above characteristic descriptions, the surface dry test block has obvious characteristics, is relatively easy to identify, and has a large quantity, while the surface wet test block has a wide range of situations and is more difficult to identify. Therefore, in view of this situation, the embodiment of the present application designs a two-stage concrete test block setting state recognition strategy of rough recognition + fine recognition. The test block entity is input into the setting state recognition model, and first, the surface dry test block and the surface wet test block are distinguished through rough recognition, and then the non-set test block is distinguished from the surface wet test block through fine recognition.

[0088] The setting state recognition model provided by the embodiment of the present application includes three main parts: Backone, First Step, and Second Step. Backone is the backbone network. The input test block entity extracts the test block feature map through the backbone network, and the test block feature map is then preliminarily classified into the surface dry test block and the surface wet test block through rough recognition by First Step. The test block feature map is input into Second Step for fine recognition to obtain the setting state prediction result of each test block entity.

[0089] The main function of the backbone network is to extract the features of the test block, which is mainly composed of the convolution module Conv Block, the inverted residual module Inverted Residual Block, the hybrid spatial interaction module (HybridSpatial Interaction Block, HSI Block for short) designed independently according to the characteristics of the task, and the output head module Head Block. The feature extraction process is: 1. Downsampling operation is performed through convolution module and inverted residual module; 2. By superimposing multiple hybrid spatial interaction modules HSI Block, features are fully extracted. HSI Block uses 5*5 and 7*7 depthwise separable convolutions to achieve medium and long-distance spatial modeling respectively, and then uses dot multiplication operations to achieve high-order spatial interaction, so as to better capture the complex relationship between features and enhance the ability to extract the entity features of the test blocks in the image. Then, residual splicing is performed to superimpose new features on the original features to obtain richer representation.

[0090] 3. Finally, the output result of the previous part is convolved with the output of the second inverted residual structure and then concatenated with the output after four HSI blocks, so as to integrate the low-dimensional basic and general feature information with the high-dimensional abstract and complex representation, improve the model performance, and output the final test block feature map.

[0091] Next, the acquired test block feature map is used for the first stage of rough recognition through the Head Block to first classify the surface dry test blocks with obvious features and easy to distinguish.

[0092] Specifically, the coarse identification is achieved through a coarse identification classifier, which mainly includes HeadBlock, which can simply classify the test block entity through the test block feature map, and identify the test block feature map corresponding to the surface dry test block and the test block feature map corresponding to the surface wet test block.

[0093] Based on the previous description of the surface dry test block, the rough identification classifier can complete the classification by identifying whether the surface of the concrete test block is reflective, whether the test block is demolded, whether there is a crack between the test block and the mold, etc.

[0094] According to the output results of the rough recognition classifier Head Block, the test block feature maps representing the surface wetness test blocks are distinguished, and these feature maps are input into the second stage for fine recognition.

[0095] First, input the test block feature map classified as a surface wetness test block into the convolutional module to adjust the number of image channels. Then, input the feature map output by the convolutional module into the feature recognition network based on the transformer structure to extract local and global features, obtaining a second feature map. Among them, the second stage mainly consists of a convolutional module, a feature extraction network based on the transformer structure, and an output head (i.e., a fine recognition classifier). The convolutional module can adjust the channels of the feature map, while the transformer module has the ability of long-distance spatial modeling and high-order spatial interaction characteristics, which can well capture the local and global features of the test block image and achieve accurate classification of the difficult-to-distinguish categories of uncoagulated and coagulated water-sprayed and plastered test blocks.

[0096] Specifically, first change the number of channels of the test block feature map output by the backbone network through the convolutional module so that the feature map meets the channel number requirements of the transformer module. Then, the transformer module first performs an unfolding operation on the feature map output by the convolutional module, regarding the pixel points in the k*k area as a token, changing the shape of the test block feature map from (B, C, H, W) to (B, C*k*k, H / k, W / k). Then, use a 1*1 convolution to modify the channels to C, and then transform and merge the dimensions. At this time, the shape of the test block feature map is (B, H / k*W / k, C); input it into the self-attention transformer structure for calculation to perform global information interaction and integration; then perform a folding operation on the test block feature map, through dimension transformation and 1*1 convolution opposite to the unfolding operation, changing the shape of the test block feature map from (B, H / k*W / k, C) to the original shape (B, C, H, W), and finally output the constructed second feature map.

[0097] Then, splice the test block feature map output by the backbone network, the feature map output by the convolutional module, and the second feature map output by the feature extraction network based on the transformer structure, and then input it into the convolutional module for feature information interaction and fusion, which not only includes the original information extracted by the backbone network, but also includes the long-distance spatial interaction information and local feature information of the test block features, fully integrating the overall information for classification operation to obtain the feature fusion map corresponding to the surface wetness test block.

[0098] Finally, perform fine recognition on the feature fusion map through the fine recognition classifier to identify uncoagulated test blocks, coagulated water-sprayed test blocks, and coagulated plastered test blocks. Among them, perform fine recognition on the color of the concrete test block surface, the shape of the slurry on the test block surface, the number and size of the holes on the test block surface, and the number of bubbles on the test block surface, etc.

[0099] In addition, fine recognition mainly uses a transformer structure on the basis of the CNN convolutional neural network of rough recognition to extract and integrate global feature information.

[0100] In some embodiments, the state comparison module 202 is specifically: According to the predicted condensation state obtained from the condensation state recognition model, the concrete test blocks are classified into surface dry test blocks, water-sprinkled test blocks after condensation, plastered test blocks after condensation, and condensed test blocks, where the surface dry test blocks, water-sprinkled test blocks after condensation, and plastered test blocks after condensation are combined into condensed test blocks.

[0101] For the sampling stage, the corresponding normal state of the test block should be the non-condensed state. Therefore, by judging whether the predicted condensation state result is the non-condensed state, it is determined whether the condensation state of the concrete test block in the current surface image of the concrete test block is normal. There are specifically the following two situations: 1. When the concrete test block in the surface image of the concrete test block taken at the sampling stage is in the condensed state, it indicates that during witnessing, pre-made test blocks may be used without treatment, or after water sprinkling or plastering treatment as the test blocks for witnessed sampling, or after on-site sampling, witnessed sampling is postponed for at least one day. An abnormal warning signal for the test block should be issued. 2. When the concrete test block in the surface image of the concrete test block taken at the sampling stage is in the non-condensed state, it indicates that it conforms to the normal state of the sampling stage, and the predicted condensation state result is returned.

[0102] Implementing the embodiments of the present application has the following beneficial effects: In the embodiments of the present application, after the concrete test block is made, the surface of the test block is photographed to obtain the surface image of the concrete test block; then the surface image of the concrete test block is first segmented, and the background, filling area, etc. that are irrelevant to the prediction or interfere with the prediction in the image are deleted, and the complete and easily recognizable test block entity is extracted; the test block entity is respectively subjected to rough recognition and fine recognition. In rough recognition, first, it is judged whether the surface of the test block is dry or wet through the image. If the test block is a surface wet test block, it is also necessary to judge whether the inside of the test block is condensed or non-condensed through more refined fine recognition, so as to realize more accurate and efficient detection of the quality of the concrete test block. There is no need to insert the instrument into the test block for detection and it is not overly dependent on manual judgment. While improving the efficiency of engineering quality detection, the detection cost is also reduced. Then, the obtained predicted condensation state result is compared with the normal condensation state corresponding to the stage when the surface image of the concrete test block is taken. If the condensation state of the concrete test block does not match the normal state that should be presented currently, it is considered that the test block may be replaced, and a witnessed abnormal warning signal will be issued to provide support for ensuring the engineering quality.

[0103] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present application. It should be understood that the above are only specific embodiments of the present application and are not used to limit the protection scope of the present application. It is particularly noted that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. An abnormal witness warning method based on the recognition of the setting state of concrete test blocks, characterized in that Including: Roughly identify the extracted test block entity to classify whether the surface of the concrete test block is dry or wet, and then finely identify the test block entity with the classification result of wet surface to classify whether the concrete test block has set or not, so as to obtain the prediction result of the setting state; wherein, the test block entity is extracted from the surface image of the photographed concrete test block. Compare the normal setting state corresponding to the stage when the surface image of the concrete test block is photographed with the prediction result of the setting state. If the comparison result is inconsistent, send out a witness anomaly warning signal.

2. The abnormal witness warning method based on the recognition of the setting state of concrete test blocks according to claim 1, characterized in that, The process of roughly identifying the extracted test block entity to classify whether the surface of the concrete test block is dry or wet, and then finely identifying the test block entity with the classification result of wet surface to classify whether the concrete test block has set or not, so as to obtain the prediction result of the setting state is specifically as follows: Extract features from the test block entity through the backbone network of the preset setting state recognition model to obtain the test block feature map. Roughly identify the test block feature map through the rough recognition classifier of the setting state recognition model, and classify the test block entity as a surface dry test block or a surface wet test block. Finely identify the test block feature map classified as a surface wet test block through the feature extraction network and the fine recognition classifier based on the transformer structure of the setting state recognition model, and classify the test block entity classified as a surface wet test block as a set and watered test block, a set and plastered test block or an unset test block. Obtain the prediction result of the setting state of the surface image of the concrete test block according to the classification result of the rough recognition and the classification result of the fine recognition.

3. The abnormal witness warning method based on the setting state recognition of concrete test blocks according to claim 2, characterized in that, The process of extracting features from the test block entity through the backbone network of the preset setting state recognition model to obtain the test block feature map is specifically as follows: Perform downsampling on the test block entity through the convolutional module and the inverted residual module of the backbone network, and then extract features from the downsampling result by stacking several hybrid spatial interaction modules of the backbone network to obtain the test block feature map. Among them, the hybrid spatial interaction module can be designed according to the actual feature extraction task.

4. The abnormal witness warning method based on the identification of the setting state of concrete test blocks according to claim 2, characterized in that, The process of roughly identifying the test block feature map through the rough recognition classifier of the setting state recognition model and classifying the test block entity as a surface dry test block or a surface wet test block is specifically as follows: Classify the test block entity corresponding to the test block feature map as a surface dry test block or a surface wet test block, and input the test block feature map corresponding to the surface wet test block into the feature extraction network and the fine recognition classifier for fine recognition.

5. The abnormal witness warning method based on the identification of the setting state of concrete test blocks according to claim 2, wherein, The process of finely identifying the test block feature map classified as a surface wet test block through the feature extraction network and the fine recognition classifier based on the transformer structure of the setting state recognition model, and classifying the test block entity classified as a surface wet test block as a set and watered test block, a set and plastered test block or an unset test block is specifically as follows: Input the feature map of the test block classified as a surface wetness test block into the convolutional module to adjust the number of image channels, and then input it into the feature extraction network based on the Transformer structure to extract local and global features, obtaining a second feature map; Concatenate the test block feature map, the feature map output by the convolutional module, and the second feature map to obtain a feature fusion map corresponding to the surface wetness test block; Perform fine recognition on the feature fusion map corresponding to the surface wetness test block through the fine recognition classifier, and classify the test block entity classified as a surface wetness test block into a coagulated water-sprayed test block, a coagulated plastered test block, or an uncoagulated test block.

6. The abnormal witness warning method based on the recognition of the setting state of concrete test blocks according to claim 2, wherein, Based on the classification results of the coarse recognition and the fine recognition, obtain the coagulation state prediction result of the surface image of the concrete test block, specifically: According to the classification result of the coarse recognition, classify the concrete test block corresponding to the surface image of the concrete test block into a surface dry test block or a surface wet test block; According to the classification result of the fine recognition, classify the concrete test block belonging to the surface wet test block into a coagulated water-sprayed test block, a coagulated plastered test block, and an uncoagulated test block.

7. The abnormal witness warning method based on the recognition of the setting state of concrete test blocks according to claim 1, wherein Compare the normal coagulation state corresponding to the stage when the surface image of the concrete test block is taken with the coagulation state prediction result. If the comparison result is inconsistent, issue a witness anomaly warning signal, specifically: If the stage is the sampling stage, set the normal coagulation state as an uncoagulated test block; If the coagulation state prediction result is an uncoagulated test block, the comparison result is consistent, and it is determined that there is no witness anomaly for the concrete test block corresponding to the coagulation state prediction result during the sampling stage; If the coagulation state prediction result is a coagulated test block, it indicates that there is a witness anomaly for the concrete test block corresponding to the coagulation state prediction result during the sampling stage, and a witness anomaly warning signal is issued.

8. The abnormal witness warning method based on the identification of the setting state of concrete test blocks according to any one of claims 1 to 7, characterized in that Surface dry test blocks, coagulated water-sprayed test blocks, and coagulated plastered test blocks are all classified as coagulated test blocks.

9. The abnormal witness warning method based on the recognition of the setting state of concrete test blocks according to claim 1, characterized in that The test block entity is extracted from the captured surface image of the concrete test block, specifically: Take pictures of the surface of the current concrete test block to obtain several surface images of the concrete test block; Preprocess the surface image of the concrete test block, and then input the preprocessed surface image of the concrete test block into an instance segmentation model for image segmentation to obtain a test block entity mask for each surface image of the concrete test block; Extract the test block entity from the surface image of the concrete test block according to the position of the test block entity mask in the surface image of the concrete test block.

10. An abnormal witness warning system based on the recognition of the setting state of concrete test blocks, characterized in that Including: A test block state prediction module and a state comparison module; Among them, the test block state prediction module is used to perform coarse recognition on the extracted test block entity to classify whether the surface of the concrete test block is dry or wet, and then perform fine recognition on the test block entity with the classification result of surface wetness to classify whether the concrete test block is coagulated or uncoagulated, obtaining a coagulation state prediction result; among them, the test block entity is extracted from the captured surface image of the concrete test block; The state comparison module is used to compare the normal setting state corresponding to the stage when the surface image of the concrete specimen is taken with the predicted setting state result. If the comparison result is inconsistent, a witness anomaly warning signal is issued.

Images