A cement-modified expansive soil ecological base material and its preparation method

Abstract

The present application discloses a cement-modified expansive soil ecological base material and a preparation method thereof. The cement-modified expansive soil ecological base material includes expansive soil, cement, ferrous sulfate, water-retaining agent, organic fertilizer, peat and polyacrylamide. Among them, the cement accounts for 3% - 10% of the dry soil mass of the expansive soil, the ferrous sulfate accounts for 1% - 2% of the dry soil mass of the expansive soil, the water-retaining agent accounts for 0.05 - 0.2% of the dry soil mass of the expansive soil, the organic fertilizer accounts for 0.5% - 2% of the dry soil mass of the expansive soil, the peat accounts for 3% - 10% of the dry soil mass of the expansive soil, and the polyacrylamide accounts for 0.05% - 0.2% of the dry soil mass of the expansive soil. This material can inhibit the expansion of the expansive soil itself, enhance the overall stability of the replaced expansive soil slope, and at the same time is suitable for plant growth to achieve the purpose of ecological slope protection. The present invention also provides a preparation method for the cement-modified expansive soil ecological base material, which is simple and easy to operate.

Description

Technical Field

[0001] This application relates to the field of intelligent preparation, and more specifically, to a cement-modified expansive soil ecological base material and a preparation method thereof. Background Art

[0002] The cement-modified expansive soil ecological base material is a material for soil improvement. It improves the physical and chemical properties of expansive soil by adding cement and other additives. This modified soil is usually used for vegetation restoration and ecological slope protection, which can improve the soil stability, reduce the expansibility of the soil, and increase its bearing capacity at the same time.

[0003] During the preparation process of the cement-modified expansive soil ecological base material, it is necessary to evenly mix the sieved expansive soil and cement to ensure that the cement is uniformly distributed in the entire expansive soil base material, so as to improve the overall performance and consistency of the base material. However, the control of the traditional mixing time usually depends on human experience or standard operating procedures. This method is not intelligent enough and is prone to uneven mixing because the operator may not be able to precisely control the mixing time of each batch, thus affecting the stability of the mixing process.

[0004] Therefore, an optimized preparation method of the cement-modified expansive soil ecological base material is desired. Summary of the Invention

[0005] To solve the above technical problems, this application is proposed. The embodiments of this application provide a cement-modified expansive soil ecological base material and a preparation method thereof. It collects the mixing state monitoring video in real time by a camera, and uses video processing and analysis technology based on a deep learning neural network to perform key frame sampling and mixing texture feature analysis on the mixing state monitoring video, so as to automatically judge whether to stop mixing according to the texture time series aggregation representation features of each mixing state key frame. In this way, the need for manual monitoring can be reduced, the labor cost can be lowered, and the automation level of the mixing process can be improved, so as to realize a more intelligent preparation process of the cement-modified expansive soil ecological base material.

[0006] According to one aspect of this application, a preparation method of a cement-modified expansive soil ecological base material is provided, which includes:

[0007] Drying the expansive soil raw material and then performing rolling and sieving treatments in sequence to obtain the sieved expansive soil;

[0008] Evenly mixing the sieved expansive soil with cement accounting for 3% - 10% of the dry mass of the expansive soil to obtain a cement-expansive soil mixture;

[0009] Measure the pH value in the cement-expansive soil mixture, add ferrous sulfate accounting for 1% - 2% of the dry mass of the expansive soil to the cement-expansive soil mixture, and simultaneously adjust the pH value to 5.5 - 7.0 to obtain the adjusted cement-expansive soil;

[0010] Add a water-retaining agent accounting for 0.05% - 0.2% of the dry mass of the expansive soil, organic fertilizer accounting for 0.5% - 2% of the dry mass of the expansive soil, peat accounting for 3% - 10% of the dry mass of the expansive soil, and polyacrylamide accounting for 0.05% - 0.2% of the dry mass of the expansive soil to the adjusted cement-expansive soil, and mix evenly to obtain a mixture;

[0011] Conduct a light compaction test on the mixture to obtain the maximum dry density and the optimum moisture content of the mixture;

[0012] Based on the optimum moisture content, add water to the mixture and mix it to prepare a soil sample with the optimum moisture content to obtain the cement-modified expansive soil ecological base material;

[0013] Among them, uniformly mixing the sieved expansive soil with cement accounting for 3% - 10% of the dry mass of the expansive soil to obtain a cement-expansive soil mixture includes:

[0014] Obtain a monitoring video of the stirring state collected by a camera;

[0015] Sample key frames from the monitoring video of the stirring state to obtain a time series of stirring state key frames;

[0016] Extract texture features from each stirring state key frame in the time series of stirring state key frames to obtain a time series of stirring texture feature maps;

[0017] Pass each stirring texture feature map in the time series of stirring texture feature maps through a multi-scale semantic feature extraction module to obtain a time series of stirring texture multi-scale semantic feature vectors;

[0018] Input the time series of stirring texture multi-scale semantic feature vectors into a node message propagation fusion network that fuses node importance to obtain a stirring texture time series propagation aggregation representation vector as a stirring texture time series propagation aggregation representation feature;

[0019] Based on the stirring texture time series propagation aggregation representation feature, obtain a control instruction, and the control instruction is used to indicate whether to stop stirring.

[0020] According to another aspect of the present application, there is provided a cement-modified expansive soil ecological base material, and the cement-modified expansive soil ecological base material is prepared by the preparation method of the aforementioned cement-modified expansive soil ecological base material.

[0021] Compared with the prior art, a cement-modified expansive soil ecological base material and its preparation method provided by the present application collect a mixing state monitoring video in real time by a camera, and use video processing and analysis techniques based on a deep learning neural network to perform key frame sampling and mixing texture feature analysis on the mixing state monitoring video, so as to automatically judge whether to stop mixing according to the texture time series aggregation representation features of each mixing state key frame. In this way, the need for manual monitoring can be reduced, the labor cost can be lowered, the automation level of the mixing process can be improved, and thus a more intelligent preparation process of the cement-modified expansive soil ecological base material can be realized. Description of the Drawings

[0022] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above and other objects, features and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0023] Figure 1 It is a flowchart of the preparation method of the cement-modified expansive soil ecological base material according to the embodiment of the present application.

[0024] Figure 2 It is a flowchart of uniformly mixing the sieved expansive soil with cement accounting for 3% - 10% of the dry soil mass of the expansive soil to obtain a cement-expansive soil mixture in the preparation method of the cement-modified expansive soil ecological base material according to the embodiment of the present application.

[0025] Figure 3 It is a schematic structural diagram of uniformly mixing the sieved expansive soil with cement accounting for 3% - 10% of the dry soil mass of the expansive soil to obtain a cement-expansive soil mixture in the preparation method of the cement-modified expansive soil ecological base material according to the embodiment of the present application.

[0026] Figure 4 It is a flowchart of respectively passing each mixing texture feature map in the time series of the mixing texture feature map through a multi-scale semantic feature extraction module to obtain a time series of mixing texture multi-scale semantic feature vectors in the preparation method of the cement-modified expansive soil ecological base material according to the embodiment of the present application.

[0027] Figure 5 It is a flowchart of training the texture feature extractor based on the dilated convolutional layer, the multi-scale semantic feature extraction module, the node message propagation fusion network integrating node importance, and the mixing controller based on the classifier in the preparation method of the cement-modified expansive soil ecological base material according to the embodiment of the present application. Detailed Description of the Embodiments

[0028] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Instead, these embodiments are provided to more thoroughly and completely understand the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not used to limit the protection scope of the present disclosure.

[0029] It should be understood that the various steps recited in the method embodiments of the present disclosure can be executed in a different order and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this regard.

[0030] In the description of the embodiments of the present disclosure, the term "including" and its like terms should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc. may refer to different or the same objects. There may also be other explicit and implicit definitions hereinafter.

[0031] It should be noted that the modification of "one" and "a plurality" mentioned in the present disclosure is illustrative rather than restrictive. Those skilled in the art should understand that unless otherwise clearly specified in the context, it should be understood as "one or more".

[0032] Cement-modified expansive soil ecological base material is a material for soil improvement. It improves the physical and chemical properties of expansive soil by adding cement and other additives. This modified soil is usually used for vegetation restoration and ecological slope protection, which can improve the soil stability, reduce the expansibility of the soil, and increase its bearing capacity at the same time.

[0033] In the preparation process of the cement-modified expansive soil ecological base material, it is necessary to uniformly mix the sieved expansive soil and cement to ensure that the cement is evenly distributed throughout the expansive soil base material, so as to improve the overall performance and consistency of the base material. However, the control of the traditional mixing time usually depends on human experience or standard operating procedures. This method is not intelligent enough and is prone to uneven mixing because the operator may not be able to accurately control the mixing time of each batch, thus affecting the stability of the mixing process.

[0034] Based on this, the present application proposes a preparation method of a cement-modified expansive soil ecological base material, which includes: drying the expansive soil raw material and then successively performing rolling and sieving treatments to obtain the sieved expansive soil; uniformly mixing the sieved expansive soil with cement accounting for 3% to 10% of the dry weight of the expansive soil to obtain a cement-expansive soil mixture; measuring the pH value in the cement-expansive soil mixture, and adding ferrous sulfate accounting for 1% to 2% of the dry weight of the expansive soil to the cement-expansive soil mixture, and at the same time adjusting the pH value to 5.5 - 7.0 to obtain the adjusted cement-expansive soil; adding a water-retaining agent accounting for 0.05% to 0.2% of the dry weight of the expansive soil, organic fertilizer accounting for 0.5% to 2% of the dry weight of the expansive soil, peat accounting for 3% to 10% of the dry weight of the expansive soil, and polyacrylamide accounting for 0.05% to 0.2% of the dry weight of the expansive soil to the adjusted cement-expansive soil, and uniformly mixing to obtain a mixture; performing a light compaction test on the mixture to obtain the maximum dry density and the optimum moisture content of the mixture; and, based on the optimum moisture content, adding water and mixing to the mixture to prepare a soil sample with the optimum moisture content to obtain the cement-modified expansive soil ecological base material.

[0035] In the process of uniformly mixing the sieved expansive soil with cement accounting for 3% to 10% of the dry weight of the expansive soil to obtain a cement-expansive soil mixture, it collects a stirring state monitoring video in real time by a camera, and uses video processing and analysis technology based on a deep learning neural network to perform key frame sampling and stirring texture feature analysis on the stirring state monitoring video, so as to automatically judge whether to stop stirring according to the texture time-series aggregation representation features of each stirring state key frame. In this way, the need for manual monitoring can be reduced, the labor cost can be lowered, and the automation level of the stirring process can be improved, thereby realizing a more intelligent preparation process of the cement-modified expansive soil ecological base material.

[0036] Figure 1 It is a flowchart of the preparation method of the cement-modified expansive soil ecological base material according to an embodiment of the present application. As Figure 1As shown, the preparation method of the cement-modified expansive soil ecological base material according to the embodiment of the present application includes: S110, drying the expansive soil raw material and then successively performing rolling and sieving treatments to obtain the sieved expansive soil; S120, uniformly mixing the sieved expansive soil with cement accounting for 3% - 10% of the dry mass of the expansive soil to obtain a cement-expansive soil mixture; S130, measuring the pH value in the cement-expansive soil mixture, adding ferrous sulfate accounting for 1% - 2% of the dry mass of the expansive soil to the cement-expansive soil mixture, and simultaneously adjusting the pH value to 5.5 - 7.0 to obtain the adjusted cement-expansive soil; S140, adding a water-retaining agent accounting for 0.05% - 0.2% of the dry mass of the expansive soil, organic fertilizer accounting for 0.5% - 2% of the dry mass of the expansive soil, peat accounting for 3% - 10% of the dry mass of the expansive soil, and polyacrylamide accounting for 0.05% - 0.2% of the dry mass of the expansive soil to the adjusted cement-expansive soil, and uniformly mixing to obtain a mixture; S150, performing a light compaction test on the mixture to obtain the maximum dry density and the optimum moisture content of the mixture; and S160, based on the optimum moisture content, adding water and mixing to the mixture to prepare a soil sample with the optimum moisture content to obtain the cement-modified expansive soil ecological base material.

[0037] Figure 2 It is a flowchart of uniformly mixing the sieved expansive soil with cement accounting for 3% - 10% of the dry mass of the expansive soil to obtain a cement-expansive soil mixture in the preparation method of the cement-modified expansive soil ecological base material according to the embodiment of the present application. Figure 3 It is a schematic structural diagram of uniformly mixing the sieved expansive soil with cement accounting for 3% - 10% of the dry mass of the expansive soil to obtain a cement-expansive soil mixture in the preparation method of the cement-modified expansive soil ecological base material according to the embodiment of the present application. Specifically, in the embodiment of the present application, as Figure 2 and Figure 3As shown, the sieved expansive soil is evenly mixed with cement accounting for 3% - 10% of the dry mass of the expansive soil to obtain a cement-expansive soil mixture, including: S210, obtaining a mixing state monitoring video collected by a camera; S220, performing key frame sampling on the mixing state monitoring video to obtain a time series of mixing state key frames; S230, respectively extracting texture features for each mixing state key frame in the time series of mixing state key frames to obtain a time series of mixing texture feature maps; S240, respectively passing each mixing texture feature map in the time series of mixing texture feature maps through a multi-scale semantic feature extraction module to obtain a time series of mixing texture multi-scale semantic feature vectors; S250, inputting the time series of mixing texture multi-scale semantic feature vectors into a node message propagation fusion network that fuses node importance to obtain a mixing texture time series propagation aggregation representation vector as a mixing texture time series propagation aggregation representation feature; and, S260, based on the mixing texture time series propagation aggregation representation feature, obtaining a control instruction, where the control instruction is used to indicate whether to stop mixing.

[0038] In step S210, a mixing state monitoring video collected by a camera is obtained. It should be understood that considering that the mixing state monitoring video is recorded in real time by a camera and shows video data of the state changes of the expansive soil and cement mixture during the mixing process. Based on this, in the technical solution of the present application, obtaining a mixing state monitoring video collected by a camera can check the uniformity of mixing and timely detect possible uneven areas or agglomeration phenomena during the mixing process, providing a basis and support for the precise control of the mixing time.

[0039] In step S220, key frame sampling is performed on the mixing state monitoring video to obtain a time series of mixing state key frames. Correspondingly, considering that the mixing state monitoring video contains a large number of frame images, and not every frame is equally important for the monitoring and analysis of the mixing state, and it also contains some background information unrelated to the mixing process. Based on this, in the technical solution of the present application, key frame sampling is performed on the mixing state monitoring video to obtain a time series of mixing state key frames. It should be understood that through key frame sampling, representative mixing state frame images can be selected as key frames, and these mixing state key frames can retain important mixing state information and changes over time, so as to more accurately reflect the temporal state changes of the entire mixing process.

[0040] In step S230, texture feature extraction is performed on each of the key frames of the stirring state in the time series of the key frames of the stirring state to obtain a time series of stirring texture feature maps. Specifically, in an embodiment of the present application, performing texture feature extraction on each of the key frames of the stirring state in the time series of the key frames of the stirring state to obtain a time series of stirring texture feature maps includes: passing each of the key frames of the stirring state in the time series of the key frames of the stirring state through a texture feature extractor based on an atrous convolutional layer to obtain the time series of the stirring texture feature maps. It should be understood that considering that each of the key frames of the stirring state in the time series of the key frames of the stirring state contains information about the stirring texture features during the stirring process, that is, the size and distribution of the expansive soil particles, and the crushing condition of the particles during the stirring process. And the atrous convolutional layer can help the network better learn the fine texture features in the image, thereby improving the representation ability and discrimination of the texture features, which is very important for distinguishing the texture features in different stirring time states. Therefore, in the technical solution of the present application, each of the key frames of the stirring state in the time series of the key frames of the stirring state is respectively passed through a texture feature extractor based on an atrous convolutional layer to respectively capture and extract the local texture features in each key frame, so as to obtain a time series of stirring texture feature maps.

[0041] In step S240, each of the stirring texture feature maps in the time series of the stirring texture feature maps is respectively passed through a multi-scale semantic feature extraction module to obtain a time series of stirring texture multi-scale semantic feature vectors. Correspondingly, considering that each of the stirring texture feature maps in the time series of the stirring texture feature maps expresses global and local semantic feature information about the stirring texture, such as the overall distribution of cement and expansive soil in the entire stirring container and the crushing and refinement conditions of the particles. Based on this, in order to more comprehensively understand and master the scene understanding of the stirring texture at each time point, in the technical solution of the present application, each of the stirring texture feature maps in the time series of the stirring texture feature maps is respectively passed through a multi-scale semantic feature extraction module to obtain a time series of stirring texture multi-scale semantic feature vectors. Specifically, the multi-scale semantic feature extraction module captures the background features of the overall stirring time series state and the temporal changes of the local texture features of the stirring state by respectively extracting the global stirring texture semantic feature information and the local stirring texture semantic feature information in each of the stirring texture feature maps, and then fuses the extracted global and local semantic features of the stirring texture to comprehensively consider the overall stirring state and the local texture features in the stirring texture feature map at each time point, so as to obtain a time series of stirring texture multi-scale semantic feature vectors.

[0042] Figure 4It is a flowchart of obtaining a time series of stirring texture multi-scale semantic feature vectors by respectively passing each stirring texture feature map in the time series of the stirring texture feature map through a multi-scale semantic feature extraction module in the preparation method of the cement-modified expansive soil ecological base material according to an embodiment of the present application. Specifically, in the embodiment of the present application, as Figure 4 shown, obtaining a time series of stirring texture multi-scale semantic feature vectors by respectively passing each stirring texture feature map in the time series of the stirring texture feature map through a multi-scale semantic feature extraction module includes: S310, extracting a stirring texture global semantic feature vector from the stirring texture feature map; S320, extracting a stirring texture local semantic feature vector from the stirring texture feature map; and S330, fusing the stirring texture global semantic feature vector and the stirring texture local semantic feature vector to obtain the stirring texture multi-scale semantic feature vector.

[0043] More specifically, in the embodiment of the present application, extracting a stirring texture global semantic feature vector from the stirring texture feature map includes: performing upsampling on the stirring texture feature map to obtain an upsampled stirring texture feature map; performing global average pooling processing on each feature matrix along the channel dimension in the upsampled stirring texture feature map to obtain an upsampled stirring texture semantic channel feature vector; and performing point convolution encoding on the upsampled stirring texture semantic channel feature vector to obtain the stirring texture global semantic feature vector.

[0044] More specifically, in the embodiment of the present application, extracting a stirring texture local semantic feature vector from the stirring texture feature map includes: performing upsampling on the stirring texture feature map to obtain an upsampled stirring texture feature map; performing point convolution encoding on the upsampled stirring texture feature map to obtain a channel-modulated upsampled stirring texture feature map; and performing two-dimensional convolution encoding on the channel-modulated upsampled stirring texture feature map to obtain the stirring texture local semantic feature vector.

[0045] In step S250, the time series of the stirring texture multi-scale semantic feature vectors is input into a node message propagation fusion network that fuses node importance to obtain a stirring texture time-series propagation aggregation representation vector as the stirring texture time-series propagation aggregation representation feature. It should be understood that considering that stirring is a dynamic process, the time series of the stirring texture multi-scale semantic feature vectors reflects the semantic features of the stirring texture at each time point, and there are also mutual associations and dependencies between the various stirring texture multi-scale semantic feature vectors. Therefore, in order to more clearly understand the evolution of the stirring texture features over time and thus better capture the temporal dynamic characteristics of the stirring process, in the technical solution of this application, the time series of the stirring texture multi-scale semantic feature vectors is input into a node message propagation fusion network that fuses node importance to obtain a stirring texture time-series propagation aggregation representation vector. It is worth mentioning that the node message propagation is a commonly used technique in graph neural networks. By transmitting information on the graph structure, each node can gather the information of surrounding nodes, thereby better representing the features of the entire graph. That is to say, in this solution, the time series of the stirring texture feature vectors can be regarded as a graph structure, and the stirring texture multi-scale semantic feature vectors at each time step can be regarded as a node. By propagating through the node message propagation fusion network that fuses node importance, the network can more flexibly process the stirring texture multi-scale semantic features of different nodes by considering the importance and contribution degree of each node, thereby better representing and understanding the features and overall temporal changes of the stirring texture state.

[0046] Specifically, in the embodiments of the present application, inputting the time series of the stirring texture multi-scale semantic feature vectors into a node message propagation fusion network that fuses node importance to obtain a stirring texture time series propagation aggregation representation vector as a stirring texture time series propagation aggregation representation feature, includes: using a scoring function to measure the significant factors of the first n-1 stirring texture multi-scale semantic feature vectors in the time series of the stirring texture multi-scale semantic feature vectors to obtain n-1 stirring texture multi-scale semantic significant factors; using the n-1 stirring texture multi-scale semantic significant factors as weights, calculating the position-wise weighted sum of the first n-1 stirring texture multi-scale semantic feature vectors in the time series of the stirring texture multi-scale semantic feature vectors to aggregate the first n-1 stirring texture multi-scale semantic feature vectors in the time series of the stirring texture multi-scale semantic feature vectors to obtain a rear node set stirring texture multi-scale semantic aggregation representation vector; using a preset hyperparameter to linearly modulate the nth stirring texture multi-scale semantic feature vector in the time series of the stirring texture multi-scale semantic feature vectors to obtain a modulated nth stirring texture multi-scale semantic feature vector; calculating the position-wise sum of the modulated nth stirring texture multi-scale semantic feature vector and the rear node set stirring texture multi-scale semantic aggregation representation vector to obtain a stirring texture multi-scale semantic full node aggregation representation vector; and inputting the stirring texture multi-scale semantic full node aggregation representation vector into a multi-layer perceptron model to obtain the stirring texture time series propagation aggregation representation vector.

[0047] More specifically, in the embodiments of the present application, using a scoring function to measure the significant factors of the first n-1 stirring texture multi-scale semantic feature vectors in the time series of the stirring texture multi-scale semantic feature vectors to obtain n-1 stirring texture multi-scale semantic significant factors, includes: respectively calculating the matrix products between each weight coefficient matrix and the first n-1 stirring texture multi-scale semantic feature vectors to obtain the first n-1 stirring texture multi-scale semantic weighted feature vectors; adding the corresponding bias vector to each stirring texture multi-scale semantic weighted feature vector in the first n-1 stirring texture multi-scale semantic weighted feature vectors to obtain the first n-1 stirring texture multi-scale semantic biased feature vectors; inputting the first n-1 stirring texture multi-scale semantic biased feature vectors into a sigmoid function for activation to obtain the first n-1 stirring texture multi-scale semantic activation feature vectors; calculating the products between each stirring texture multi-scale semantic activation feature vector in the first n-1 stirring texture multi-scale semantic activation feature vectors and the weight coefficient vector to obtain the first n-1 stirring texture multi-scale semantic correlation degrees; and inputting the first n-1 stirring texture multi-scale semantic correlation degrees into a softmax function to obtain the n-1 stirring texture multi-scale semantic significant factors.

[0048] In the embodiments of the present application, specifically, inputting the time series of the stirring texture multi-scale semantic feature vectors into a node message propagation fusion network that fuses node importance to obtain a stirring texture time series propagation aggregation representation vector, including: using the node message propagation fusion network that fuses node importance to process the time series of the stirring texture multi-scale semantic feature vectors according to the following propagation fusion formula to obtain the stirring texture time series propagation aggregation representation vector; wherein, the propagation fusion formula is:

[0049]

[0050]

[0051] wherein, v i represents the i-th stirring texture multi-scale semantic feature vector in the time series of the stirring texture multi-scale semantic feature vectors, W i represents the weight coefficient matrix, B i is the bias vector, represents the weight coefficient vector, sigmoid(·) represents the sigmoid function, softmax(·) represents the softmax function, e(v i ) represents the stirring texture multi-scale semantic significance factors at each position in the sequence of the stirring texture multi-scale semantic significance factors, n is the number of feature vectors in the time series of the stirring texture multi-scale semantic feature vectors, v n represents the n-th stirring texture multi-scale semantic feature vector in the time series of the stirring texture multi-scale semantic feature vectors, ∈ is the preset hyperparameter, MLP(·) represents the multi-layer perceptron, and v c is the stirring texture time series propagation aggregation representation vector.

[0052] In step S260, based on the stirring texture time series propagation aggregation representation feature, a control instruction is obtained, and the control instruction is used to indicate whether to stop stirring. Specifically, in the embodiments of the present application, obtaining a control instruction based on the stirring texture time series propagation aggregation representation feature includes: inputting the stirring texture time series propagation aggregation representation vector into a stirring controller based on a classifier to obtain the control instruction. That is, classification processing is performed using the stirring texture time series propagation aggregation representation feature obtained by fusing node messages with the time series of the stirring texture multi-scale semantic feature vectors, so as to automatically determine whether to stop stirring. In this way, the need for manual monitoring can be reduced, the labor cost can be lowered, the automation level of the stirring process can be improved, and thus a more intelligent preparation process of cement-modified expansive soil ecological base materials can be realized.

[0053] It is worth mentioning that those of ordinary skill in the art should be aware that before applying a deep neural network model for inference, the deep neural network model needs to be trained first so that the deep neural network can implement a specific function.

[0054] Specifically, in the embodiments of the present application, a training step is further included: for training the texture feature extractor based on the dilated convolutional layer, the multi-scale semantic feature extraction module, the node message propagation fusion network that fuses node importance, and the classifier-based stirring controller.

[0055] Figure 5 It is a flowchart for training the texture feature extractor based on the dilated convolutional layer, the multi-scale semantic feature extraction module, the node message propagation fusion network that fuses node importance, and the classifier-based stirring controller in the method for preparing a cement-modified expansive soil ecological base material according to the embodiments of the present application. As Figure 5 shown, the training step includes: S410, obtaining training data, where the training data includes a stirring state monitoring video collected by a camera, and a true control instruction, and the true control instruction is the true value of whether to stop stirring; S420, performing key frame sampling on the training stirring state monitoring video to obtain a time series of training stirring state key frames; S430, respectively passing each training stirring state key frame in the time series of training stirring state key frames through the texture feature extractor based on the dilated convolutional layer to obtain a time series of training stirring texture feature maps; S440, respectively passing each training stirring texture feature map in the time series of training stirring texture feature maps through the multi-scale semantic feature extraction module to obtain a time series of training stirring texture multi-scale semantic feature vectors; S450, inputting the time series of training stirring texture multi-scale semantic feature vectors into the node message propagation fusion network that fuses node importance to obtain a training stirring texture time series propagation aggregation representation vector; S460, passing the training stirring texture time series propagation aggregation representation vector through the classifier-based stirring controller to obtain a training control instruction; S470, calculating the cross-entropy loss function value between the training control instruction and the true control instruction to obtain a classification loss function value; and S480, based on the classification loss function value and through backpropagation of gradient descent, training the texture feature extractor based on the dilated convolutional layer, the multi-scale semantic feature extraction module, the node message propagation fusion network that fuses node importance, and the classifier-based stirring controller.

[0056] It should be understood that here, considering the multi-scale image semantic feature non-correspondence of the source image semantic non-correspondence of each training stirring texture multi-scale semantic feature vector in the time series of the training stirring texture multi-scale semantic feature vectors, when fusing the node importance in the node message propagation fusion, the training stirring texture time series propagation aggregation representation vector will also have a knowledge shift of the image semantic distribution caused by the difference in the propagation of the node importance of the image semantic feature distribution, thereby causing the uncertainty of its image semantic feature distribution with respect to the class probability understanding of the classifier-based stirring controller, reducing the speed of the classification training and the accuracy of the classification result.

[0057] Based on this, in a preferred embodiment, inputting the training stirring texture time series propagation aggregation representation vector into the classifier-based stirring controller to obtain a training control instruction includes: determining a stop class probability value corresponding to the representation of stopping stirring obtained by the training stirring texture time series propagation aggregation representation vector through the classifier-based stirring controller, and subtracting the stop class probability value from one to obtain a non-stop class probability value; respectively calculating the power functions of each eigenvalue of the training stirring texture time series propagation aggregation representation vector with the stop class probability value as the exponent and the power function with the non-stop class probability value as the exponent to obtain a training stirring texture time series propagation aggregation stop class vector and a training stirring texture time series propagation aggregation non-stop class vector; performing a dot product of the training stop class probability value and the training stirring texture time series propagation aggregation non-stop class vector to obtain a first training stirring texture time series propagation aggregation cross-class vector, and performing a dot product of the non-stop class probability value and the training stirring texture time series propagation aggregation stop class vector to obtain a second training stirring texture time series propagation aggregation cross-class vector; after performing a dot product of the first training stirring texture time series propagation aggregation cross-class vector and the second training stirring texture time series propagation aggregation cross-class vector, further performing a dot addition with the dot product result of the stop class probability value and the training stirring texture time series propagation aggregation stop class vector to obtain the optimized training stirring texture time series propagation aggregation representation vector; and inputting the optimized training stirring texture time series propagation aggregation representation vector into the classifier-based stirring controller to obtain the training control instruction.

[0058] That is, when classifying the training stirring texture time-series propagation aggregation representation vector through the classifier-based stirring controller, in order to achieve unsupervised domain adaptation from the feature distribution domain of the training stirring texture time-series propagation aggregation representation vector to the probability distribution domain of classification probability, the stop class probability value and non-stop class probability value obtained by passing the training stirring texture time-series propagation aggregation representation vector through the classifier-based stirring controller are used as domain proxies, and the moving average of the probability distributions of the training stirring texture time-series propagation aggregation representation vector based on the training stirring texture time-series propagation stop class vector and the training stirring texture time-series propagation non-stop class vector is performed through power function class distribution interaction as the exponent, and the knowledge transfer from the unlabeled classification feature to the labeled probability distribution is achieved by superimposing the feature domain probability distribution knowledge of the training stirring texture time-series propagation aggregation representation vector itself, thereby promoting the classification operation of the training stirring texture time-series propagation aggregation representation vector through the classifier-based stirring controller, that is, improving the speed of classification training and the accuracy of control instructions of the classifier-based stirring controller. In this way, the need for manual monitoring can be reduced, the labor cost can be lowered, the automation level of the stirring process can be improved, and thus a more intelligent preparation process of cement-modified expansive soil ecological base material can be realized.

[0059] In the embodiment of the present application, specifically, when inputting the training stirring texture time-series propagation aggregation representation vector into the classifier-based stirring controller to obtain a training control instruction, optimizing the training stirring texture time-series propagation aggregation representation vector to obtain an optimized training stirring texture time-series propagation aggregation representation vector includes: optimizing the training stirring texture time-series propagation aggregation representation vector with the following optimization formula to obtain the optimized training stirring texture time-series propagation aggregation representation vector; where the optimization formula is:

[0060] V′ = (p⊙V 1-p )⊙[(1 - )⊙V p ⊕(p⊙V p )

[0061] where V is the training stirring texture time-series propagation aggregation representation vector, p is the stop class probability value corresponding to indicating stop stirring obtained by passing the training stirring texture time-series propagation aggregation representation vector through the classifier-based stirring controller, 1 - is the non-stop class probability value, V 1-p is the training stirring texture time-series propagation non-stop class vector, V p is the training stirring texture time-series propagation stop class vector, ⊙ is element-wise multiplication, ⊕ is element-wise addition, and V′ is the optimized training stirring texture time-series propagation aggregation representation vector.

[0062] In summary, the preparation method of the cement-modified expansive soil ecological base material based on the embodiments of the present application is elucidated. It collects the mixing state monitoring video in real time by a camera, and uses video processing and analysis techniques based on deep learning neural networks to perform key frame sampling and mixing texture feature analysis on the mixing state monitoring video, so as to automatically judge whether to stop mixing according to the texture time series aggregation representation features of each mixing state key frame. In this way, the need for manual monitoring can be reduced, the labor cost can be lowered, the automation level of the mixing process can be improved, and thus a more intelligent preparation process of the cement-modified expansive soil ecological base material can be achieved.

[0063] Specifically, a cement-modified expansive soil ecological base material is also provided, and the cement-modified expansive soil ecological base material is prepared by the preparation method of the cement-modified expansive soil ecological base material described above.

[0064] The above are only examples of the principles of the present disclosure, and those skilled in the art can make various modifications without departing from the scope of the present disclosure. The above embodiments are presented for illustration rather than limitation. The present disclosure can also take many forms other than those explicitly described herein. Therefore, it should be emphasized that the present disclosure is not limited to the explicitly disclosed methods and devices, but is intended to include variations and modifications within the spirit scope of the appended claims.

Claims

1. A preparation method of a cement-modified expansive soil ecological base material, characterized in that, Including: Drying the expansive soil raw material and then successively performing rolling and sieving treatments to obtain the sieved expansive soil; Uniformly mixing the sieved expansive soil with cement accounting for 3% - 10% of the dry mass of the expansive soil to obtain a cement-expansive soil mixture; Measuring the pH value in the cement-expansive soil mixture, adding ferrous sulfate accounting for 1% - 2% of the dry mass of the expansive soil to the cement-expansive soil mixture, and simultaneously adjusting the pH value to 5.5 - 7.0 to obtain the adjusted cement-expansive soil; Adding a water retaining agent accounting for 0.05% - 0.2% of the dry mass of the expansive soil, organic fertilizer accounting for 0.5% - 2% of the dry mass of the expansive soil, peat accounting for 3% - 10% of the dry mass of the expansive soil, and polyacrylamide accounting for 0.05% - 0.2% of the dry mass of the expansive soil to the adjusted cement-expansive soil, and uniformly mixing to obtain a mixture; Performing a light compaction test on the mixture to obtain the maximum dry density and the optimum moisture content of the mixture; Based on the optimum moisture content, adding water to the mixture for mixing to prepare a soil sample with the optimum moisture content to obtain the cement-modified expansive soil ecological base material; Among them, uniformly mixing the sieved expansive soil with cement accounting for 3% - 10% of the dry mass of the expansive soil to obtain a cement-expansive soil mixture, including: Obtaining a stirring state monitoring video collected by a camera; Performing key frame sampling on the stirring state monitoring video to obtain a time series of stirring state key frames; Respectively extracting texture features from each stirring state key frame in the time series of stirring state key frames to obtain a time series of stirring texture feature maps; Respectively passing each stirring texture feature map in the time series of stirring texture feature maps through a multi-scale semantic feature extraction module to obtain a time series of stirring texture multi-scale semantic feature vectors; Inputting the time series of stirring texture multi-scale semantic feature vectors into a node message propagation fusion network that fuses node importance to obtain a stirring texture time series propagation aggregation representation vector as a stirring texture time series propagation aggregation representation feature; Based on the stirring texture time series propagation aggregation representation feature, obtaining a control instruction, and the control instruction is used to indicate whether to stop stirring; The method further includes a training step: for training a texture feature extractor based on a dilated convolutional layer, a multi-scale semantic feature extraction module, a node message propagation fusion network that fuses node importance, and a stirring controller based on a classifier; Among them, the training step includes: Obtaining training data, where the training data includes a stirring state monitoring video collected by a camera, and a true control instruction, and the true control instruction is the true value of whether to stop stirring; Performing key frame sampling on the stirring state monitoring video to obtain a time series of training stirring state key frames; Respectively passing each training stirring state key frame in the time series of training stirring state key frames through the texture feature extractor based on the dilated convolutional layer to obtain a time series of training stirring texture feature maps; Respectively pass each training stirring texture feature map in the time series of the training stirring texture feature maps through the multi-scale semantic feature extraction module to obtain a time series of training stirring texture multi-scale semantic feature vectors; Input the time series of the training stirring texture multi-scale semantic feature vectors into the node message propagation fusion network that fuses node importance to obtain a training stirring texture time-series propagation aggregation representation vector; Pass the training stirring texture time-series propagation aggregation representation vector through the stirring controller based on the classifier to obtain a training control instruction; Calculate the cross-entropy loss function value between the training control instruction and the true control instruction to obtain a classification loss function value; Based on the classification loss function value and through backpropagation of gradient descent, train the texture feature extractor based on the dilated convolutional layer, the multi-scale semantic feature extraction module, the node message propagation fusion network that fuses node importance, and the stirring controller based on the classifier.

2. The preparation method of the cement-modified expansive soil ecological base material according to claim 1, characterized in that, Respectively perform texture feature extraction on each stirring state key frame in the time series of the stirring state key frames to obtain a time series of stirring texture feature maps, including: respectively pass each stirring state key frame in the time series of the stirring state key frames through the texture feature extractor based on the dilated convolutional layer to obtain the time series of the stirring texture feature maps.

3. The preparation method of the cement-modified expansive soil ecological base material according to claim 2, characterized in that, Respectively pass each stirring texture feature map in the time series of the stirring texture feature maps through the multi-scale semantic feature extraction module to obtain a time series of stirring texture multi-scale semantic feature vectors, including: Extract the stirring texture global semantic feature vector from the stirring texture feature map; Extract the stirring texture local semantic feature vector from the stirring texture feature map; Fuse the stirring texture global semantic feature vector and the stirring texture local semantic feature vector to obtain the stirring texture multi-scale semantic feature vector.

4. The preparation method of the cement-modified expansive soil ecological base material according to claim 3, characterized in that, Extract the stirring texture global semantic feature vector from the stirring texture feature map, including: Perform upsampling on the stirring texture feature map to obtain an upsampled stirring texture feature map; Perform global average pooling processing on each feature matrix along the channel dimension in the upsampled stirring texture feature map to obtain an upsampled stirring texture semantic channel feature vector; Perform point convolution encoding on the upsampled stirring texture semantic channel feature vector to obtain the stirring texture global semantic feature vector.

5. The preparation method of the cement-modified expansive soil ecological base material according to claim 4, characterized in that, Extract the stirring texture local semantic feature vector from the stirring texture feature map, including: Perform upsampling on the stirring texture feature map to obtain an upsampled stirring texture feature map; Perform point convolution encoding on the upsampled stirring texture feature map to obtain a channel-modulated upsampled stirring texture feature map; Perform two-dimensional convolution encoding on the channel-modulated upsampled stirring texture feature map to obtain the stirring texture local semantic feature vector.

6. The preparation method of the cement-modified expansive soil ecological base material according to claim 5, characterized in that, Input the time series of the stirring texture multi-scale semantic feature vectors into the node message propagation fusion network that fuses node importance to obtain a stirring texture time-series propagation aggregation representation vector as the stirring texture time-series propagation aggregation representation feature, including: Use a scoring function to measure the significant factors of the first n - 1 stirring texture multi-scale semantic feature vectors in the time series of the stirring texture multi-scale semantic feature vectors to obtain n - 1 stirring texture multi-scale semantic significant factors; Using the n - 1 stirring texture multi-scale semantic significant factors as weights, calculate the position-weighted sum of the first n - 1 stirring texture multi-scale semantic feature vectors in the time series of the stirring texture multi-scale semantic feature vectors to aggregate the first n - 1 stirring texture multi-scale semantic feature vectors in the time series of the stirring texture multi-scale semantic feature vectors to obtain a posterior node set stirring texture multi-scale semantic aggregation representation vector; Use preset hyperparameters to linearly modulate the nth stirring texture multi-scale semantic feature vector in the time series of the stirring texture multi-scale semantic feature vectors to obtain the modulated nth stirring texture multi-scale semantic feature vector; Calculate the position-wise sum of the modulated nth stirring texture multi-scale semantic feature vector and the posterior node set stirring texture multi-scale semantic aggregation representation vector to obtain a stirring texture multi-scale semantic full-node aggregation representation vector; Input the stirring texture multi-scale semantic full-node aggregation representation vector into a multi-layer perceptron model to obtain the stirring texture time series propagation aggregation representation vector.

7. The preparation method of the cement-modified expansive soil ecological base material according to claim 6, characterized in that, Use a scoring function to measure the significant factors of the first n - 1 stirring texture multi-scale semantic feature vectors in the time series of the stirring texture multi-scale semantic feature vectors to obtain n - 1 stirring texture multi-scale semantic significant factors, including: Calculate the matrix product between each weight coefficient matrix and the first n - 1 stirring texture multi-scale semantic feature vectors respectively to obtain the first n - 1 stirring texture multi-scale semantic weighted feature vectors; Add the corresponding bias vector to each stirring texture multi-scale semantic weighted feature vector in the first n - 1 stirring texture multi-scale semantic weighted feature vectors to obtain the first n - 1 stirring texture multi-scale semantic bias feature vectors; Input the first n - 1 stirring texture multi - scale semantic bias feature vectors into the function for activation to obtain the first n - 1 stirring texture multi - scale semantic activation feature vectors; Calculate the product between each stirring texture multi-scale semantic activation feature vector and the weight coefficient vector in the first n - 1 stirring texture multi-scale semantic activation feature vectors to obtain the first n - 1 stirring texture multi-scale semantic correlation degrees; Input the first n - 1 stirring texture multi-scale semantic correlation degrees into a function to obtain the n - 1 stirring texture multi-scale semantic significant factors.

8. The preparation method of the cement-modified expansive soil ecological base material according to claim 7, characterized in that, Based on the stirring texture time series propagation aggregation representation feature, obtain a control instruction, including: inputting the stirring texture time series propagation aggregation representation vector into a stirring controller based on a classifier to obtain the control instruction.

Images