Portable device and method for rapidly detecting silt content in sand

Abstract

The invention provides a device and a method for rapidly detecting the silt content in sand. The device comprises a mud and sand sampler, a sample bottle, a photographing device, and a white contrast sticker, a transparent test window and a black contrast sticker which are arranged on the sample bottle. The detection method comprises the following steps: mixing a mud sand sample with water in a sample bottle, taking a picture, extracting RGB parameters of a black and white contrast area and a transparent test window in the picture, and calculating the mud content by using a neural network model. The influence of sunny days and cloudy weather on light rays is considered in model training, and the detection accuracy is improved. The device is portable, is simple to operate, is suitable for on-site rapid detection, and is especially suitable for quality monitoring of sandstone materials in constructional engineering.

Description

Technical Field

[0001] The invention belongs to the field of civil engineering application and is a technology capable of quickly identifying mud content of mud and sand. Background Art

[0002] A large amount of excavated mud and sand is often generated in engineering activities, such as the Western Land-Sea New Channel (Pinglu) Canal Project, which is planned to be about 135 kilometers long and involves the excavation and construction of hundreds of millions of earth and stone. The excavated mud and sand need to be tested for mud content on the spot in a timely and rapid manner, and the mud content is used to determine whether it is directly utilized as a resource, utilized as a resource after sand washing, or discarded. However, the current project mainly relies on the drying and weighing method, that is, drying-weighing-screening-drying-weighing-calculation, which is mainly completed by manual operation. The detection steps are complicated, time-consuming, and require a large oven, which is inefficient. It is unable to meet the real-time detection needs on the spot, and can only be transported to the storage yard / sand washing yard for unified sampling and laboratory testing, and then transported again for resource utilization or landfill disposal, resulting in a large waste of manpower and material resources.

[0003] The existing method based on taking photos of mud cannot cope with the impact of changes in weather, light, and equipment on mud photos, and cannot obtain sufficiently accurate data; in addition, it is not integrated into a complete set of equipment, is not portable, and cannot be used for on-site measurement. For the sand excavated in real time, if it cannot be tested in time, it can only be transported to the storage yard or sand washing yard first, and then classified and disposed of after testing, which will waste a lot of manpower and material resources. Summary of the invention

[0004] In view of the above problems existing in the prior art, in order to identify the mud content on-site and quickly, the present invention provides a portable device and method for quickly detecting the mud content in sand.

[0005] The present invention first provides a device for detecting mud content in sand, comprising a container capable of containing water and sand, wherein the outer surface of the container has connected black and white, wherein the standard RGB color code of the black is RGB (0,0,0), and the standard RGB color code of the white is RGB (255,255,255); a transparent test window adjacent to both the black and white is also provided on the outer surface of the container, wherein the transparent test window is surrounded by the black and white, and the transparency of the transparent test window is such that an object in the container can be displayed without color difference; an area on the outer surface of the container corresponding to a certain plane projection of the container simultaneously includes the black, white and transparent test windows, so that the black, white and transparent test windows can appear simultaneously on a photo when taking a photo;

[0006] The steps of using the device to detect the mud content in sand include: adding sampled sand into a container, adding a specific volume of water in advance or temporarily, so that the water level is at least higher than the lowest end of the transparent test window.

[0007] Preferably, the shape of the container includes one or more of a cylinder and a cube; the container can hold liquid, the material of the container includes transparent plastic, and the container can be selected from a plastic bottle of a specific volume.

[0008] The present invention also provides a method for measuring mud content using the above device, comprising the following steps:

[0009] Step 1: Take the mud and sand to be tested with a volume of L1 as the sample, and the weight of the mud and sand with a volume of L1 (mL) is M (g);

[0010] Step 2: Put all the mud and sand of volume L into the device of claim 1, add water of volume L2 (mL), shake well, and let stand until the volume of the aqueous solution is sufficient to cover at least part of the transparent test window;

[0011] Step 3: Take a photo so that the white, black, and transparent test windows appear in the same photo;

[0012] Step 4: Extract the RGB parameters of the black, white, and transparent test windows in the photo;

[0013] Step 5: Correct the RGB parameters R′, G′, and B′ of the transparent test window according to the RGB parameters of black and white. The correction method is based on formulas (1), (2), and (3):

[0014] R′=(R 测试区域 -R 黑色对比区域 ) / (R 白色对比区域 -R 黑色对比区域 ) (1)

[0015] G′=(G 测试区域 -G 黑色对比区域 ) / (G 白色对比区域 -G 黑色对比区域 ) (2)

[0016] B′=(B 测试区域 -B 黑色对比区域 ) / (B 白色对比区域 -B 黑色对比区域 ) (3)

[0017] Step 6: According to the weather when the photo was taken, select the BPCNN model under sunny conditions or the BPCNN model under cloudy conditions, and input , , and into the BPANN model. The model output is the mud concentration C (g / mL) in the water shown in the photo, which is converted into the mud content in the mud sand as C*L2 / M*100%.

[0018] Preferably, the sunny day means that the sun is visible and not blocked by clouds, and the cloudy day means that the sun is blocked by clouds and cannot be seen when taking pictures.

[0019] Preferably, the adopted training samples cover samples with mud concentrations of 1%, 3%, 6%, 10%, 15%, 20%, 30%, 40%, 50%, and 60% in the solution.

[0020] Preferably, the rest time is 15 s to allow the grit to settle.

[0021] Compared with the prior art, the present invention has the following advantages and effects:

[0022] 1. Portable: The detection device of the present invention is small and light, with a volume not exceeding 30cm*30cm*30cm and a weight not exceeding 1kg. For example, a 500mL mineral water bottle can be used as the detection device of the present invention. It is portable and easy to operate, and can realize real-time detection of mud content in sand fields;

[0023] 2. Efficient: One person can complete all operations on the mud content in sand. The detection time of a mud and sand sample is less than or equal to 5 minutes, and the result can be obtained in about 3 minutes.

[0024] 3. Accuracy:

[0025] 1) The present invention calculates the mud content based on the mud image. After a specific settling time, only mud particles with a particle size less than 0.075 mm will remain in the water. At this time, the color depth of the mud image is linearly related to the mud content, and the model inversion accuracy is high;

[0026] 2) The detection device is set with a black and white pure color contrast area, which can correct the error caused by light brightness and color difference caused by different mobile phone photos;

[0027] 3) Two models were fitted based on the measured data under the influence of different degrees of reflection brought by sunny and cloudy days. The above measures greatly improved the prediction accuracy of mud content in sand (the error can reach 0.5%), with strong adaptability and high robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a sampler style diagram, where 1 is the scale line and 2 is the mud and sand sampler.

[0029] Figure 2 This is a diagram of the structure of a pure water sample bottle, where 3 is a white contrast area, 4 is a transparent sampling area, and 5 is a black contrast area.

[0030] Figure 3 A batch of samples collected on site are used for model training.

[0031] Figure 4 This is a photo of a field sample. The mud content is obtained in real time by analyzing the RGB parameters in the photo through the model of the present invention. DETAILED DESCRIPTION

[0032] The specific implementation modes of the present invention are further described in detail below with reference to the accompanying drawings.

[0033] Example 1

[0034] like Figure 2 As shown, the present invention provides a device for quickly detecting mud content, which includes: a 5ml mud and sand sampler (such as Figure 1 ), a sample bottle with a volume of at least 500ml (hereinafter referred to as a 500ml sample bottle or a 500ml pure water sample bottle), a camera device such as a mobile phone, a white contrast sticker, a transparent test window and a black contrast sticker set on the 500ml sample bottle. The standard RGB color code of the black contrast sticker is RGB (0,0,0), and the standard RGB color code of the white contrast sticker is RGB (255,255,255); the transparent test window is adjacent to both the black contrast sticker and the white contrast sticker, the container of the transparent test window is transparent, the transparent test window is surrounded by black and white, and the transparency of the transparent test window is such that the object in the container can be displayed without color difference; a certain plane projection of the sample bottle The area on the sample bottle includes the black contrast sticker, the white contrast sticker and the transparent test window at the same time.

[0035] The shape of the sample bottle is cylindrical, and experimenters in this field can also set it to common shapes such as cube.

[0036] Among them, the sampling volume of the mud and sand sampler is 5ml. Volume sampling instead of weighing sampling is adopted in order to obtain a fixed volume of mud and sand to cope with the errors caused by different water content and weight of mud and sand on site.

[0037] 500 ml of water is stored in a 500 ml sample bottle to retain the mud in the mud and sand in the water to form mud slurry, which is convenient for the next step of testing. White contrast stickers, black contrast stickers and a transparent test window in the middle are prefabricated on the sample bottle. When taking pictures with a mobile phone, the image RGB parameters of the white contrast stickers and black contrast stickers can be introduced and compared with the RGB parameters of the transparent test window to calculate the final mud content, which can effectively reflect the changes in light and dark conditions to reduce the deviation of the data caused by light conditions.

[0038] Example 2

[0039] This embodiment demonstrates a method for establishing a mathematical model for calculating mud content. Pure mud of different masses is taken and mixed with a fixed volume of pure water to prepare mud with known mud content. The image RGB parameters of the black and white contrast areas and the transparent test window are extracted by taking photos under different weather and light intensity conditions, and a neural network model training is carried out.

[0040] The specific implementation is as follows: 3 typical types of mud and sand were collected from the sand mining sites of the 13th and 14th sections of Pinglu Canal and the Pingnan Gudu section of Qinzhou, Guangxi. Pure mud samples with weights of 0.10g, 0.15g, 0.20g, 0.25g, 0.30g, 0.35g, 0.40g, 0.50g, 1.00g, 1.50g, 2.00g, 2.50g, 3.00g, 4.00g, 5.00g and 6.00g were taken from each typical type of mud and sand. A total of 48 groups of pure mud samples with the above weights were obtained from the three typical types of mud and sand. 48 groups of mud samples were prepared according to the following method, and 254 images under different light conditions (sunny and cloudy) were obtained:

[0041] Place each pure mud sample in a sample bottle, add 500 ml of water to each bottle, and obtain the mud concentration in the solution. Then, shake the sample bottle up and down for 10 seconds and let it stand for 15 seconds. The transparent test window is facing away from the sun (to prevent direct sunlight), and the pure water sample bottle is held vertically above the head. The sample bottle is at an elevation angle of 45° to the horizontal plane (to prevent reflections). The transparent test window of the pure water sample bottle is facing the camera phone, making sure there is no reflection in the photo area, and focus the phone on the transparent test window to shoot. Extract the RG B parameters of the black and white contrast areas and the transparent test window in the middle from the photos taken.

[0042] A neural network is used to establish the relationship between the mud concentration in the water and the RGB parameters of these three regions. The neural network used in the present invention is BPANN, which is a relatively common multi-layer feedforward neural network using a back-propagation algorithm, and has a strong nonlinear function fitting ability. BPANN usually includes an input layer, several hidden layers and an output layer. Each layer is composed of a different number of "neurons". Each neuron passes the weighted sum of the input of the previous layer to the next layer until it reaches the output layer. When the output is inconsistent with the expected output, the error will be back-propagated to adjust the weights and biases of the neurons in each layer. This process is repeated until the error between the predicted output and the expected output is acceptable. Ultimately, the neural network model can be used to make predictions using similar input data.

[0043] In order to improve the convergence speed and accuracy of the model, the input and output data were normalized before training. The data values ​​of each column of the data set are mapped to the range [0,1]. The prediction results will also be reverse normalized to the original value interval. After normalization (Formulas 4, 5, and 6), a total of 254 sets of images were obtained. We used 80% of the data as a training set to train the model; 20% of the data was used as a test set to test the model. The following table shows the 254 sets of sample data used to establish and verify the model. The input features of the neural network are R′, G′, and B′, and the output feature is the mud concentration in the water (the mud weight in the table below is divided by the volume of the sampled water 500mL).

[0044] R′=(R 测试区域 -R 黑色对比区域 ) / (R 白色对比区域 -R 黑色对比区域 ) (1)

[0045] G′=(G 测试区域 -G 黑色对比区域 ) / (G 白色对比区域 -G 黑色对比区域 ) (2)

[0046] B′=(B 测试区域 -B 黑色对比区域 ) / (B 白色对比区域 -B 黑色对比区域 ) (3)

[0047] Among them, R 测试区域 is the R value of the transparent test window, R 黑色对比区域 is the R value of the black contrast area, R 白色对比区域 is the R value of the white contrast area; G 测试区域 is the G value of the transparent test window, G 黑色对比区域 is the G value of the black contrast area, G 白色对比区域 is the G value of the white contrast area; B 测试区域 is the B value of the transparent test window, B 黑色对比区域 is the B value of the black contrast area, B 白色对比区域 is the B value of the white contrast area.

[0048] Table 1 254 groups of data

[0049]

[0050]

[0051]

[0052]

[0053]

[0054]

[0055]

[0056] During the training process, the topology and training hyperparameters of the BPANN model were optimized. The topology of the final optimized model consists of 1 input layer, 2 hidden layers, and 1 output layer. The input layer consists of 3 neurons, corresponding to 3 input features. The 2 hidden layers are fully connected, with 64 and 32 neurons in each layer, respectively. The output layer consists of 1 neuron, corresponding to the predicted value. The hidden layer uses the relu function as the activation function, introducing nonlinear factors to make the neural network more capable of mapping nonlinear functions. The hyperparameters used in training include loss function, optimizer, epoch number, and batch size. The specific settings are as follows: Loss function: Mean Squared Error (MSE) Optimizer: Adam, learning rate is 0.001 Training epoch number: 2000 Batch size: 64 In each epoch, the model updates parameters according to the training data, and adjusts the network weights by calculating the value of the loss function to reduce the error. Finally, through the training and optimization of the model, accurate prediction of the input data is achieved. In particular, it was found during training that the error of establishing a model for the concentration of mud in water that is applicable to all weather conditions is too large (there is always reflection on a sunny day, which causes a large difference from no reflection on a cloudy / overcast day). Therefore, when generating data, we separated two different weather conditions, namely sunny and cloudy weather, and modeled them separately. Since the neural network model is inconvenient to express, only the test results are shown here as follows:

[0057] The model accuracy test calculation is performed using the data of the validation set and the test set. The neural network model of the present invention has higher detection accuracy when the mud concentration in the water is low. The calculation is based on the fact that the dry weight of 5mL sampled mud and sand is about 10g. The accuracy of the present invention is converted into the mud content of the mud and sand for expression, that is: in sunny weather, for samples with a mud content of 0.5%-3%, the average measurement error is 0.48%, for samples with a mud content of 3%-10%, the average measurement error is 0.91%, and for samples with a mud content of 10%-60%, the average measurement error is 5.39%; in cloudy weather, for samples with a mud content of 0.5%-3%, the average measurement error is 0.23%, for samples with a mud content of 3%-10%, the average measurement error is 0.23%, and for samples with a mud content of 10%-60%, the average measurement error is 8.28%. It can be seen that when the model accuracy is below 10%, the accuracy of the mud content of mud and sand is relatively high, and the error is within 1%, which is suitable for determining the qualified mud content in excavation sites and sand washing sites and can be transferred to resource utilization scenarios; when the mud content of mud and sand is above 10%, the accuracy is slightly poor, but it is also suitable for on-site and real-time application in excavation sites to determine whether it is worth sand washing (mud and sand with sand washing value usually have a mud content of less than 50%. Taking into account the measurement error, mud and sand with a mud content of less than 40% calculated by the neural network simulation of the present invention usually have sand washing value).

[0058] The present invention also provides a method for establishing a mathematical model for calculating mud content 2. In order to compare with the neural network and to perform detection when it is inconvenient to use a computer, a linear fitting method is provided. The pre-processing steps of the method are the same as above. For 254 sets of data, the black and white contrast area of ​​the image and the RGB parameters of the transparent test window are extracted, and R′G′B′ is calculated using formulas (1)(2)(3). Then, the following formula for calculating mud content is obtained by linear fitting:

[0059] Under sunny conditions: Mud content = -0.5142R′-0.0075G′-0.0348B′+0.4590 (4)

[0060] Under cloudy conditions: Mud content = -1.0819R′+0.1807G′+0.1168B′+0.8945 (5)

[0061] Compared with the mathematical model established by the neural network, the traditional linear fitting function has a lower computational complexity and is suitable for scenarios with high real-time requirements and limited computing resources. However, the linear model can only fit linear relationships and cannot handle complex nonlinear problems, resulting in inferior performance to the neural network in mud content detection. When the linear fitting predicts a mud content of less than 10% under sunny conditions, the average error is 4.47%; when predicting a mud content of more than 10% under sunny conditions, the average error is 16.13%; when predicting a mud content of less than 10% under cloudy conditions, the average error is 4.26%; when predicting a mud content of more than 10% under cloudy conditions, the average error reaches 13.96%; if a separate model is built for the mud content below 10%, the error can be reduced to 1.15% on sunny days and 0.69% on cloudy days. Compared with the model that models the overall mud content data, this model is more accurate in detecting low mud content (below 12%). If a more accurate linear fitting of low mud content (below 12%) is required in engineering, the following formula can be obtained by fitting the data with mud content below 12%. The following formula can be used to predict low mud content (below 12%):

[0062] Under sunny conditions: Mud content = 0.3114R′-0.4360G′+0.12285B′+0.0177 (6)

[0063] Under cloudy conditions: Mud content = 0.2303R′-0.4304G′+0.1436B′+0.0525 (7)

[0064] Compared with traditional linear fitting, the neural network algorithm can fit complex nonlinear relationships through multi-layer feedforward networks and nonlinear activation functions, and is suitable for complex mud content detection tasks. BPANN can reduce errors by continuously adjusting neuron weights and biases. In actual tests, BPANN has high detection accuracy.

[0065] Example 3

[0066] The present invention also provides a method for quickly detecting mud content using the device in Example 1 and the mathematical model in Example 2. The detection method comprises the following steps:

[0067] Step 1: Take the mud and sand to be tested with a volume of L1 as the sample, and the weight of the mud and sand with a volume of L1 (mL) is M (g);

[0068] Step 2: Put all the mud and sand of volume L into the device of claim 1, add water of volume L2 (mL), shake well, and let stand until the volume of the aqueous solution is sufficient to cover at least part of the transparent test window;

[0069] Step 3: Take a photo so that the white, black, and transparent test windows appear in the same photo;

[0070] Step 4: Extract the RGB parameters of the black, white, and transparent test windows in the photo;

[0071] Step 5: Correct the RGB parameters R′, G′, and B′ of the transparent test window according to the RGB parameters of black and white. The correction method is based on formulas (1), (2), and (3):

[0072] R′=(R 测试区域 -R 黑色对比区域 ) / (R 白色对比区域 -R 黑色对比区域 ) (1)

[0073] G′=(G 测试区域 -G 黑色对比区域 ) / (G 白色对比区域 -G 黑色对比区域 ) (2)

[0074] B′=(B 测试区域 -B 黑色对比区域 ) / (B 白色对比区域 -B 黑色对比区域 ) (3)

[0075] Step 6: According to the weather when the photo was taken, select the BPCNN model under sunny conditions or the BPCNN model under cloudy conditions, input R′, G′, and B′ into the BPANN model, and the model output is the mud concentration C (g / mL) in the water shown in the photo. The mud content in the mud sand is converted to C*L2 / M*100%.

[0076] Among them, the above-mentioned sunny day means that the sun is visible and not blocked by clouds, and cloudy weather means that the sun is blocked by clouds and cannot be seen when taking pictures.

[0077] More specifically, the detection steps include the following:

[0078] Step 1: Insert the 5ml sampler horizontally into the mud and sand to be tested until it covers the scale line;

[0079] Step 2: Place the mud and sand collected by the 5ml sampler into a customized pure water sample bottle, add water until the water surface is near the bottle mouth and shake up and down for 10 seconds;

[0080] Step 3: Go to a relatively open place, let the evenly shaken pure water sample bottle stand for 15 seconds (according to the Stokes sedimentation formula, after 15 seconds, all sand particles with a particle size greater than 0.075 mm will settle below the transparent test window, and the only factor that affects the color of the mud in the window is the amount of mud particles). With your back to the sun (to prevent direct sunlight), hold the pure water sample bottle vertically above your head, with the bottle tilted downward at a 45° angle (to prevent glare), and make sure that the transparent test window of the pure water sample bottle faces the camera phone, making sure there is no glare in the photo area, and focus the phone on the transparent test window to take the photo;

[0081] Step 4: Use the image recognition applet with the neural network model described in Example 2 (with a built-in algorithm to extract the black and white contrast area of ​​the image and the RGB parameters of the transparent test window, obtain the mud concentration in the water through the model, and convert the mud content according to the above calculation formula of the present invention) to directly obtain the mud content result. In this example, mud and sand samples were taken from a sand washing plant in Pinglu Canal, two samples were taken from each sampling location, and the above experimental operations were performed in parallel. The trained neural network model in Example 2 was used to predict the mud content, and the test results are as follows:

[0082] Table 2 Test results

[0083]

[0084]

[0085] As can be seen from the results in the above table, the measured error is close to the training result, and the detection error of samples with a mud content of less than 10% is within 0.5%, which meets the needs of engineering applications; and the measurement time is about 50 seconds, which is much lower than laboratory testing (laboratory testing time takes 1-2 days). The present invention provides a portable, efficient and accurate device and method for on-site detection of mud content, which greatly improves the efficiency of post-excavation mud treatment, avoids mud accumulation and waste of resources, and has great on-site application value.

[0086] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some or all of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A device for detecting the mud content in sand, comprising a container capable of containing water and sand, It is characterized in that There are connected black and white on the outer surface of the container, the standard RGB color code of the black is RGB (0,0,0), and the standard RGB color code of the white is RGB (255,255,255); a transparent test window adjacent to both the black and white is also provided on the outer surface of the container, the transparent test window is surrounded by the black and white, and the transparency of the transparent test window is such that the object in the container can be displayed without color difference; The area on the outer surface of the container corresponding to a certain plane projection image of the container includes black, white and transparent test windows at the same time; The steps of using the device to detect the mud content in sand include: adding sampled sand into a container, adding a specific volume of water in advance or temporarily, so that the water level is at least higher than the lowest end of the transparent test window.

2. The device according to claim 1, It is characterized in that The shape of the container includes one or more of a cylinder and a cube; the container can hold liquid, and the material of the container includes transparent plastic.

3. A method for measuring mud content using the device according to claim 1, comprising the following steps: Step 1: Take the mud and sand to be tested with a volume of L1 as the sample, and the weight of the mud and sand with a volume of L1 (mL) is M (g); Step 2: Put all the mud and sand of volume L into the device of claim 1, add water of volume L2 (mL), shake well, and let stand until the volume of the aqueous solution is sufficient to cover at least part of the transparent test window; Step 3: Take a photo so that the white, black, and transparent test windows appear in the same photo; Step 4: Extract the RGB parameters of the black, white, and transparent test windows in the photo; Step 5: Correct the RGB parameters R′, G′, and B′ of the transparent test window according to the RGB parameters of black and white. The correction method is based on formulas (1), (2), and (3): R′=(R 测试区域 -R 黑色对比区域 ) / (R 白色对比区域 -R 黑色对比区域 ) (1) G′=(G 测试区域 -G 黑色对比区域 ) / (G 白色对比区域 -G 黑色对比区域 ) (2) B′=(B 测试区域 -B 黑色对比区域 ) / (B 白色对比区域 -B 黑色对比区域 ) (3) Step 6: According to the weather when the photo was taken, select the BPCNN model under sunny conditions or the BPCNN model under cloudy conditions, input R′, G′, and B′ into the BPANN model, and the model output is the mud concentration C (g / mL) in the water shown in the photo. The mud content in the mud sand is converted to C*L2 / M*100%.

4. The method according to claim 3, It is characterized in that The sunny day means that the sun is visible and not blocked by clouds, and the cloudy day means that the sun is blocked by clouds and cannot be seen when taking pictures.

5. The method according to claim 4, It is characterized in that The training samples used cover samples with mud concentrations of 1%, 3%, 6%, 10%, 15%, 20%, 30%, 40%, 50%, and 60% in the solution.

6. The method according to claim 3, It is characterized in that The standing time was 15 s to allow the grit to settle.

Images