Concrete quality detection method based on mixing truck platform
By collecting concrete images and physical features inside the cement tanker and using deep learning algorithms for real-time detection, the problem of difficulty in real-time monitoring of the internal state of concrete in the prior art is solved, and high accuracy and real-time concrete quality detection is achieved.
Patent Information
- Application Number
- CN202510184617.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-19
- Publication Date
- 2025-05-13
AI Technical Summary
The prior art is difficult to monitor the internal state of concrete in real time and accurately during concrete transportation, resulting in inconsistent concrete quality at the construction site and inconsistent with the transportation process, and there are errors and safety risks.
The concrete quality detection method based on the mixing transport truck platform is used to determine whether the concrete quality is abnormal by collecting real-time concrete images inside the cement tanker, combining physical characteristics, and using deep learning algorithms (such as the YOLOv8 model) for real-time detection and classification.
It realizes non-contact real-time detection of concrete status inside the cement tanker, improves the accuracy and real-timeness of concrete quality inspection, and ensures that the quality of concrete is qualified during transportation and use.
Smart Images
Figure CN119974242A_ABST
Abstract
Description
Technical Field
[0001] The disclosed embodiments relate to the field of data processing technology, and in particular to a concrete quality detection method based on a mixer truck platform. Background Art
[0002] At present, in the process of building construction, the quality of concrete directly affects the safety and service life of the project. At present, concrete quality detection mainly relies on manual inspection on the construction site. This method is not only inefficient and has poor real-time performance, but also has a high risk of error and personnel safety risks. Especially during the transportation of concrete, due to the mixing and vibration inside the tank truck, the state of concrete will continue to change, resulting in the actual concrete quality on the construction site being inconsistent with that during transportation.
[0003] At present, similar systems have been installed in foreign countries by installing physical sensors on concrete transport vehicles, which can monitor concrete parameters such as temperature, slump, and mixing speed in real time. These online monitoring systems not only improve the uniformity of concrete, but also reduce on-site rejections due to substandard concrete quality. For example, in the United States and Europe, many concrete companies have applied this technology to their transport vehicles, which has significantly improved the transportation quality and construction efficiency of concrete.
[0004] The quality monitoring technology for concrete transportation in China started late, mainly due to the complex internal environment of the tank truck and the lack of technical means for real-time monitoring. Due to the narrow and dark environment inside the tank truck, traditional physical sensors or contact measurement methods have great limitations in real-time and non-destructive monitoring of concrete status. These technologies are more applied to static scenes in the laboratory and cannot meet the needs of accurate identification of concrete status in actual transportation.
[0005] However, although foreign physical sensor systems can effectively monitor the basic parameters of concrete, these methods also have certain limitations in complex transportation environments. For example, these systems mainly detect the external characteristics of concrete (such as temperature and mixing speed), but it is difficult to accurately capture the subtle changes inside the concrete, especially segregation, over-mixing or changes in physical properties that occur during transportation.
[0006] It can be seen that the concrete quality detection method based on the mixer truck platform can detect the concrete state in real time inside the cement tank truck without contact. Summary of the invention
[0007] In view of this, an embodiment of the present disclosure provides a concrete quality detection method based on a mixer truck platform, which at least partially solves some of the problems existing in the prior art.
[0008] The present disclosure provides a concrete quality detection method based on a mixer truck platform, comprising:
[0009] Step 1, collecting real-time concrete images during transportation by cement tank trucks;
[0010] Step 2, input the real-time concrete image and physical characteristics into the target model to obtain the quality inspection result corresponding to the current concrete;
[0011] Step 3, determine whether the quality inspection result is abnormal, if so, generate a decision plan, if not, return to step 1.
[0012] According to a specific implementation of the embodiment of the present disclosure, the target model is a yo l ov8 model.
[0013] According to a specific implementation of the embodiment of the present disclosure, step 2 specifically includes:
[0014] Step 2.1, input the real-time concrete image into the changeable convolution kernel layer in the backbone network of the yo l ov8 model and combine it with the attention module to obtain the original feature map, and perform global average pooling on the original feature map to obtain the global description vector;
[0015] Step 2.2, performing an activation scaling operation on the global description vector to generate a scaling factor and weighting the feature map accordingly to obtain a weighted feature map;
[0016] Step 2.3, collecting real-time physical data during cement tanker transportation and converting it into physical features;
[0017] Step 2.4, input the weighted feature map and physical features into the Neck layer of the yo l ov8 model for feature fusion to obtain the fused features;
[0018] Step 2.5: predict the quality inspection result corresponding to the current concrete based on the fusion features.
[0019] According to a specific implementation of the embodiment of the present disclosure, step 2.1 specifically includes:
[0020] Step 2.1.1, input the real-time concrete image into the changeable convolution kernel layer combined with the attention module in the backbone network of the yo l ov8 model to obtain image features;
[0021] Step 2.1.2, generate displacement according to image features, change the convolution kernel layer combined with the attention module by learning the offset, and these offsets will be applied to the sampling position of the convolution operation;
[0022] Step 2.1.3, use the changeable convolution kernel layer combined with the attention module to perform convolution operation on the image features. For each offset sampling point, use the weight of the convolution kernel to perform weighted summation to generate a new feature map, where the expression of the new feature map is
[0023]
[0024] Where (x+i+Δi,y+j+Δj) represents each offset sampling point, K(i,j) represents the weight of the convolution kernel, I represents the input image, and I(x+i+Δi,y+j+Δj) represents the pixel value of the input image at the offset position (x+i+Δi,y+j+Δj);
[0025] Step 2.1.4, perform global average pooling on the feature map rows through the self-attention mechanism, calculate the global average value for each channel, and generate a global description vector, where the expression of the global description vector is
[0026]
[0027] Among them, Z c represents the global description vector, c represents the number of channels, H is the height of the feature map, and W is the width.
[0028] According to a specific implementation of the embodiment of the present disclosure, step 2.2 specifically includes:
[0029] Step 2.2.1: Activate and scale the global description vector through two fully connected layers to generate a scale factor for weighting
[0030] s=σ(W2*δ(W1*z))
[0031] Among them, s is the scale factor used for weighting, W1 and W2 are the weight matrices of the fully connected layer, δ is the ReLU activation function, and σ is the sigmode function;
[0032] Step 2.2.2: Weight the original feature map by the proportional factor and adjust the response intensity of each channel to obtain the weighted feature map.
[0033] O1(c,i,j)=s c *O(c,i,j)
[0034] Among them, s c is the weight on channel c, and O1 is the final weighted feature map.
[0035] According to a specific implementation of the embodiment of the present disclosure, step 2.3 specifically includes:
[0036] Step 2.3.1, collecting pulse signals corresponding to multiple different types of sensors used to monitor data during cement tanker transportation, and amplifying and shaping the pulse signals through a signal processing circuit and then inputting them into an analog-to-digital converter to obtain a digital signal;
[0037] Step 2.3.2, analyze the digital signal and obtain the physical characteristics.
[0038] According to a specific implementation of the embodiment of the present disclosure, the expression of the fusion feature is:
[0039] F fused =αF img +βF phy
[0040] Among them, F fused represents the feature map after multimodal fusion, α and β are the weight parameters that need to be learned, F img represents the weighted feature map, F phy Represents physical characteristics.
[0041] The concrete quality detection scheme based on the mixer truck platform in the disclosed embodiment includes: step 1, collecting real-time concrete images during cement tanker transportation; step 2, inputting the real-time concrete images and physical characteristics into the target model to obtain the quality detection results corresponding to the current concrete; step 3, judging whether the quality detection results are abnormal, if so, generating a decision plan, if not, returning to step 1.
[0042] The beneficial effects of the embodiments of the present disclosure are as follows: through the scheme of the present disclosure, by collecting concrete state images inside cement tankers and combining physical features, real-time detection and classification are performed through deep learning algorithms. By deploying the optimized YOLOv8 model inside cement tankers to perform real-time state classification of concrete, concrete feature information can be effectively identified and monitored to ensure that the concrete is of qualified quality when in use. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0044] Figure 1 A schematic flow chart of a concrete quality detection method based on a mixer truck platform provided in an embodiment of the present disclosure;
[0045] Figure 2A schematic diagram of a specific application process of a concrete quality detection method based on a mixer truck platform provided in an embodiment of the present disclosure;
[0046] Figure 3 A schematic diagram of a specific hardware deployment structure provided for an embodiment of the present disclosure. DETAILED DESCRIPTION
[0047] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.
[0048] The following describes the embodiments of the present disclosure through specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.
[0049] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.
[0050] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The drawings only show components related to the present disclosure rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.
[0051] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.
[0052] The embodiment of the present disclosure provides a concrete quality detection method based on a mixer truck platform, which can be applied to the summary of the concrete quality detection process in a construction scene.
[0053] See also Figure 1 , is a flow chart of a concrete quality detection method based on a mixer truck platform provided by an embodiment of the present disclosure. Figure 1 and Figure 2 As shown, the method mainly comprises the following steps:
[0054] Step 1, collecting real-time concrete images during transportation by cement tank trucks;
[0055] In specific implementation, the disclosed embodiment adopts a high-performance edge computing platform based on RK3588 to support real-time data processing and analysis. Unlike the traditional method that relies on the server for reasoning and analysis, RK3588 provides efficient local computing capabilities, which can perform real-time concrete image acquisition and analysis during cement tanker transportation, reduce data transmission delays, and significantly improve the response speed of the system. Figure 3 :Device 1 is the power supply, device 2 is rk3588, and device 3 is the camera. When in use, the camera can be used to collect real-time concrete images during the transportation of cement tank trucks for subsequent operation processes.
[0056] Step 2, input the real-time concrete image and physical characteristics into the target model to obtain the quality inspection result corresponding to the current concrete;
[0057] Optionally, the target model is a yo l ov8 model.
[0058] Furthermore, the step 2 specifically includes:
[0059] Step 2.1, input the real-time concrete image into the changeable convolution kernel layer in the backbone network of the yo l ov8 model and combine it with the attention module to obtain the original feature map, and perform global average pooling on the original feature map to obtain the global description vector;
[0060] Step 2.2, performing an activation scaling operation on the global description vector to generate a scaling factor and weighting the feature map accordingly to obtain a weighted feature map;
[0061] Step 2.3, collecting real-time physical data during cement tanker transportation and converting it into physical features;
[0062] Step 2.4, input the weighted feature map and physical features into the Neck layer of the yo l ov8 model for feature fusion to obtain the fused features;
[0063] Step 2.5: predict the quality inspection result corresponding to the current concrete based on the fusion features.
[0064] Furthermore, the step 2.1 specifically includes:
[0065] Step 2.1.1, input the real-time concrete image into the changeable convolution kernel layer combined with the attention module in the backbone network of the yo l ov8 model to obtain image features;
[0066] Step 2.1.2, generate displacement according to image features, change the convolution kernel layer combined with the attention module by learning the offset, and these offsets will be applied to the sampling position of the convolution operation;
[0067] Step 2.1.3, use the changeable convolution kernel layer combined with the attention module to perform convolution operation on the image features. For each offset sampling point, use the weight of the convolution kernel to perform weighted summation to generate a new feature map, where the expression of the new feature map is
[0068]
[0069] Where (x+i+Δi,y+j+Δj) represents each offset sampling point, K(i,j) represents the weight of the convolution kernel, I represents the input image, and I(x+i+Δi,y+j+Δj) represents the pixel value of the input image at the offset position (x+i+Δi,y+j+Δj);
[0070] Step 2.1.4, perform global average pooling on the feature map rows through the self-attention mechanism, calculate the global average value for each channel, and generate a global description vector, where the expression of the global description vector is
[0071]
[0072] Among them, Z c represents the global description vector, c represents the number of channels, H is the height of the feature map, and W is the width.
[0073] Furthermore, the step 2.2 specifically includes:
[0074] Step 2.2.1: Activate and scale the global description vector through two fully connected layers to generate a scale factor for weighting
[0075] s=σ(W2*δ(W1*z))
[0076] Among them, s is the scale factor used for weighting, W1 and W2 are the weight matrices of the fully connected layer, δ is the ReLU activation function, and σ is the sigmode function;
[0077] Step 2.2.2: Weight the original feature map by the proportional factor and adjust the response intensity of each channel to obtain the weighted feature map.
[0078] O1(c,i,j)=s c *O(c,i,j)
[0079] Among them, s c is the weight on channel c, and O1 is the final weighted feature map.
[0080] Furthermore, the step 2.3 specifically includes:
[0081] Step 2.3.1, collecting pulse signals corresponding to multiple different types of sensors used to monitor data during cement tanker transportation, and amplifying and shaping the pulse signals through a signal processing circuit and then inputting them into an analog-to-digital converter to obtain a digital signal;
[0082] Step 2.3.2, analyze the digital signal and obtain the physical characteristics.
[0083] Furthermore, the expression of the fusion feature is:
[0084] F fused =αF img +βF phy
[0085] Among them, F fused represents the feature map after multimodal fusion, α and β are the weight parameters that need to be learned, F img represents the weighted feature map, F phy Represents physical characteristics.
[0086] In the specific implementation, yolov8 is used as the detection and classification framework. In Backbone, the following ordinary convolution layers are changed to changeable convolution kernel layers combined with attention modules to improve the extraction of higher-level features. At the same time, physical features and visual features are fused in the Neck layer. In order to improve the accuracy of concrete detection, the following improvements are made:
[0087] Data preprocessing uses dark channel prior algorithm to improve clarity in poor lighting conditions;
[0088] The changeable convolution kernel is designed in combination with the self-attention mechanism. In view of the fact that concrete is in a state of continuous flow, the present invention uses a changeable convolution kernel for feature extraction and learns variable convolution kernel displacement to capture irregular surface features. At the same time, the self-attention mechanism is combined to enhance the feature response on the channel and dynamically adapt to irregular surface features. The steps are as follows:
[0089] Input feature map: provided by the replaced convolutional layer in the backbone, assuming that the input feature map is F∈R C*H*W, where C is the number of channels, H is the height of the feature map, and W is the width.
[0090] Generate offsets Δi, Δj: The convolution kernel can be changed by learning offsets Δi, Δj, and these offsets will be applied to the sampling positions of the convolution operation.
[0091] Convolution kernel weighting: For each offset sampling point (x+i+Δi, y+j+Δj), the convolution kernel weight K(i,j) is used for weighted summation to generate a new feature map. The specific convolution operation is as follows, where k is the size of the convolution kernel and O(x,y) is the output feature map.
[0092]
[0093] Global average pooling: Perform global average pooling on O(x,y) through the self-attention mechanism, calculate the global average for each channel c, and generate a global description vector z∈RC:
[0094]
[0095] Activation and scaling: The global description vector z is activated and scaled through two fully connected layers to generate a scale factor s∈RC for weighting:
[0096] s=σ(W2*δ(W1*z))
[0097] Among them, W1 and W2 are the weight matrices of the fully connected layer, δ is the ReLU activation function, and σ is the sigmode function.
[0098] Feature map weighting: The original feature map O is weighted by the generated weight s to adjust the response strength of each channel:
[0099] O1(c,i,j)=s c *O(c,i,j)
[0100] Among them, s c is the weight on channel c, and O1 is the final weighted feature map.
[0101] The weighted feature map is used in the attention mechanism. By calculating the weights of specific regions or channels, the network can focus on the important parts of the image or feature map and ignore the irrelevant parts. The weighting process allows the network to dynamically adjust the attention to different feature map regions, thereby improving the performance of the model.
[0102] Multimodal feature fusion This invention innovatively fuses visual information with the speed, current, and voltage of the motor in the Neck part of yolov8, enhances the model's sensitivity to state changes, and forms a detection framework for multimodal data fusion. The pulse signals generated by various sensors are usually very weak and need to be amplified and shaped by a signal processing circuit to generate a clear square wave pulse signal. The amplified and shaped pulse signal is sent to the analog-to-digital converter (ADC), and the ADC converts the pulse signal into a digital signal (the above process is completed inside the sensor). The converted digital signal is output to the RK3588 industrial computer through the RS485 communication interface. The industrial computer can obtain physical sensor features such as the mixing drum speed, current, and voltage by analyzing these digital signals. These physical features F phy The fusion is weighted into the yolov8 fully connected layer. The fusion method adopts weighted fusion, which introduces weights between image features and physical sensor features, so that different features are weighted and summed according to their importance:
[0103] F fused =αF img +βF phy
[0104] α and β are weight parameters that need to be learned, which represent the relative importance of image features and physical features. The detection head generates bounding boxes, assigns confidence scores, and classifies the boxes according to their categories. The probability of each category is calculated through the sigmo id function, and then the global category loss is calculated. The raw output of the detection head needs to go through non-maximum suppression (NMS) to remove redundant detection boxes and select the best detection result.
[0105] The application of multimodal feature mixing is to improve the accuracy of concrete classification by combining visual image features and physical sensor features. The state of concrete can also be judged by using physical sensors (speed sensors, current transformers, voltage sensors, etc.). Concrete requires different speeds, currents, and voltages in different states. The weighted feature map above is completely based on image features. The state of concrete can be detected based on image features. In the Neck part of YOLOv8, the weighted image features and physical sensor features are weighted and summed, and different features are weighted according to their importance.
[0106] Advantages of multimodal feature fusion of visual information and physical sensor features: 1. Improve detection accuracy: Combining visual and physical information can provide a more comprehensive understanding of the concrete state and reduce misjudgment. 2. Enhance sensitivity to state changes: Physical sensors can capture subtle changes that are difficult to reflect in images, such as segregation, over-mixing, etc. 3. Improve the robustness of the model: Even if the image quality is poor or interfered by the environment, the physical sensor information can still provide a reliable reference.
[0107] Step 3, determine whether the quality inspection result is abnormal, if so, generate a decision plan, if not, return to step 1.
[0108] In specific implementation, when the system detects abnormal concrete conditions inside the cement tanker (such as segregation, thinness, dryness, etc.), the system will automatically issue an alarm and suggest that the operator make corresponding adjustments based on the classification results of the concrete conditions, such as adjusting the mixing speed or adding an appropriate amount of water. At the same time, the detection data will be uploaded to the remote server through the Internet of Things module for remote monitoring and decision-making by managers.
[0109] Based on the Internet of Things technology, the system supports uploading concrete status classification results and video information to the cloud server in real time. Through big data analysis, the generated intelligent report not only contains real-time classification results, but also includes historical data, trend analysis and abnormal event records of concrete quality. The construction party can make data-driven decisions through these reports to ensure that the quality of concrete meets the standards during transportation and use.
[0110] The concrete quality detection method based on the mixer truck platform provided in this embodiment uses the concrete state image collected inside the cement tanker and combines it with physical features to perform real-time detection and classification through a deep learning algorithm. By deploying the optimized YOLOv8 model inside the cement tanker to perform real-time state classification of concrete, concrete feature information can be effectively identified and monitored to ensure that the concrete is of qualified quality when in use.
[0111] It should be understood that various parts of the present disclosure may be implemented in hardware, software, firmware or a combination thereof.
[0112] The above is only a specific implementation of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present disclosure should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.
Claims
1. A concrete quality detection method based on a mixer truck platform, characterized in that: include: Step 1, collecting real-time concrete images during transportation by cement tank trucks; Step 2, input the real-time concrete image and physical characteristics into the target model to obtain the quality inspection result corresponding to the current concrete; Step 3, determine whether the quality inspection result is abnormal, if so, generate a decision plan, if not, return to step 1.
2. The method according to claim 1, characterized in that The target model is the yolov8 model.
3. The method according to claim 2, characterized in that The step 2 specifically includes: Step 2.1, input the real-time concrete image into the changeable convolution kernel layer in the backbone network of the yolov8 model and combine it with the attention module to obtain the original feature map, and perform global average pooling on the original feature map to obtain the global description vector; Step 2.2, performing an activation scaling operation on the global description vector to generate a scaling factor and weighting the feature map accordingly to obtain a weighted feature map; Step 2.3, collecting real-time physical data during cement tanker transportation and converting it into physical features; Step 2.4, input the weighted feature map and physical features into the Neck layer of the yolov8 model for feature fusion to obtain the fused features; Step 2.5: predict the quality inspection result corresponding to the current concrete based on the fusion features.
4. The method according to claim 3, characterized in that The step 2.1 specifically includes: Step 2.1.1, input the real-time concrete image into the changeable convolution kernel layer in the backbone network of the yolov8 model combined with the attention module to obtain the image features; Step 2.1.2, generate displacement according to image features, change the convolution kernel layer combined with the attention module by learning the offset, and these offsets will be applied to the sampling position of the convolution operation; Step 2.1.3, use the changeable convolution kernel layer combined with the attention module to perform convolution operation on the image features. For each offset sampling point, use the weight of the convolution kernel to perform weighted summation to generate a new feature map, where the expression of the new feature map is Where (x+i+Δi,y+j+Δj) represents each offset sampling point, K(i,j) represents the weight of the convolution kernel, I represents the input image, and I(x+i+Δi,y+j+Δj) represents the pixel value of the input image at the offset position (x+i+Δi,y+j+Δj); Step 2.1.4, perform global average pooling on the feature map rows through the self-attention mechanism, calculate the global average value for each channel, and generate a global description vector, where the expression of the global description vector is Among them, Z c represents the global description vector, c represents the number of channels, H is the height of the feature map, and W is the width.
5. The method according to claim 4, characterized in that The step 2.2 specifically includes: Step 2.2.1: Activate and scale the global description vector through two fully connected layers to generate a scale factor for weighting s=σ(W2*δ(W1*z)) Among them, s is the scale factor used for weighting, W1 and W2 are the weight matrices of the fully connected layer, δ is the ReLU activation function, and σ is the sigmode function; Step 2.2.2: Weight the original feature map by the proportional factor and adjust the response intensity of each channel to obtain the weighted feature map. O1(c,i,j)=s c *O(c,i,j) Among them, s c is the weight on channel c, and O1 is the final weighted feature map.
6. The method according to claim 5, characterized in that The step 2.3 specifically includes: Step 2.3.1, collecting pulse signals corresponding to multiple different types of sensors used to monitor data during cement tanker transportation, and amplifying and shaping the pulse signals through a signal processing circuit and then inputting them into an analog-to-digital converter to obtain a digital signal; Step 2.3.2, analyze the digital signal and obtain the physical characteristics.
7. The method according to claim 6, characterized in that The expression of the fusion feature is: F fused =αF img +βF phy Among them, F fused represents the feature map after multimodal fusion, α and β are the weight parameters that need to be learned, F img represents the weighted feature map, F phy Represents physical characteristics.