Dike hidden danger detection method

Abstract

The invention discloses an embankment hidden danger detection method. The method comprises the following steps that 1, a resistivity thermodynamic diagram of an embankment is obtained; 2, preprocessing and enhancing the resistivity thermodynamic diagram in sequence; 3, constructing an FCN model, wherein the FCN model comprises an input layer, an encoder layer, a decoder layer and an output layer; the output of the FCN model is a resistivity thermodynamic diagram which displays a low-resistance area and is covered with a mask; and 4, training the constructed FCN model, and taking the resistivity thermodynamic diagram processed in the step 2 as the input of the FCN model to obtain a detection result. According to the method, the FCN model is combined to identify the low-resistance region, and compared with manual interpretation, the detection precision and the detection efficiency can be effectively improved.

Description

Technical Field

[0001] The invention relates to the technical field of embankment detection, in particular to a method for detecting hidden dangers in embankments. Background Art

[0002] As vital water conservancy infrastructure, levees shoulder the crucial task of defending against floods and protecting people's lives and property. However, levees face various safety hazards over their long-term operation, among which piping and holes are among the most common and significant. Piping occurs when water flows through gaps or weak channels within a levee, carrying away soil particles and causing water gushing. If not addressed promptly, this can lead to levee failure. Holes, formed internally due to defects in levee construction, changes in geological conditions, or external damage, also weaken the levee's structural stability.

[0003] Traditional methods for detecting levee hazard risks rely primarily on manual experience, examining resistivity thermograms and other detection data to determine the presence of piping and holes. This approach is not only inefficient but also susceptible to human error, leading to frequent missed detections and false alarms. With the continuous advancement of science and technology, deep learning technology has achieved remarkable results in areas such as image recognition and classification, providing new insights and methods for levee hazard detection. Summary of the Invention

[0004] In view of the deficiencies in the prior art, the present invention provides a method for detecting hidden dangers in dikes to solve the technical problems of low detection efficiency and accuracy caused by relying on manual experience in the prior art.

[0005] The present invention provides a method for detecting hidden dangers in dikes, comprising the following steps:

[0006] Step 1: Obtain the resistivity thermodynamic map of the dam;

[0007] Step 2: Preprocess and enhance the resistivity thermodynamic map in sequence;

[0008] Step 3: Construct an FCN model. The FCN model includes an input layer, an encoder layer, a decoder layer, and an output layer. The output of the FCN model is a resistivity heat map showing the low-resistance area covered with a mask.

[0009] Step 4: Train the constructed FCN model and use the resistivity thermogram processed in step 2 as the input of the FCN model to obtain the detection results.

[0010] Furthermore, in step 1, the specific method for obtaining the resistivity thermodynamic map is:

[0011] By installing a detection device equipped with a transient electromagnetic instrument, detection is carried out at a certain speed interval along the side line of the embankment to obtain the secondary eddy current field attenuation curve, and the conductivity distribution of the underground medium is inverted. Based on the inverted conductivity distribution data, the resistivity thermodynamic map of the embankment is generated.

[0012] Furthermore, in step 2, the resistivity thermodynamic map is preprocessed by:

[0013] The resistivity thermodynamic map is scaled to a standard size and normalized.

[0014] Furthermore, in step 2, the method for enhancing the resistivity thermodynamic map is:

[0015] The preprocessed resistivity thermogram is enhanced by random rotation, translation, flipping and Gaussian noise.

[0016] Furthermore, the rotation angle range is: ±20°; the translation range is: ±10°; and the standard deviation of Gaussian noise is: 0.1.

[0017] Furthermore, in step 3, the loss function of the constructed FCN model is: binary cross entropy.

[0018] Furthermore, the specific formula of the loss function is:

[0019]

[0020] Where y i ∈{0,1} is the true label; p i ∈[0,1] is the model prediction probability.

[0021] Furthermore, in step 3, when constructing the FCN model, the classification threshold of the FCN model is optimized by ROC.

[0022] Furthermore, the classification threshold of the optimized FCN model is 0.05-0.15.

[0023] Furthermore, in step 3, the mask is a binary mask.

[0024] Beneficial effects of the present invention:

[0025] This invention uses the FCN model to identify low-resistance areas, effectively improving detection accuracy and efficiency compared to manual interpretation. Preprocessing and enhancing resistivity thermograms effectively eliminates interference signals and improves the accuracy of subsequent detection. Optimizing the constructed FCN model improves its detection efficiency and accuracy. By intelligently adjusting the FCN model's classification threshold using ROC, the model becomes more sensitive and accurate in identifying low-resistance areas, while balancing missed detections and false alarms, making it more suitable for high-risk embankment sections. BRIEF DESCRIPTION OF THE DRAWINGS DETAILED DESCRIPTION

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0027] The present invention will be further described below with reference to specific examples. Those skilled in the art will appreciate that these examples are intended only to illustrate the present invention and are not intended to limit the scope of the present invention, and that modifications to various equivalent forms of the present invention fall within the scope defined by the appended claims.

[0028] The present invention provides a method for detecting hidden dangers in dikes, comprising the following steps:

[0029] Step 1: Using a detection device equipped with a transient electromagnetic instrument, conduct detection at regular speed intervals along the levee lateral to obtain the secondary eddy current field attenuation curve and invert the underground medium conductivity distribution. Based on the inverted conductivity distribution data, a resistivity thermogram of the levee is generated. The resistivity thermogram is a two-dimensional resistivity pseudo-color image.

[0030] Determine the mask, use the mask as a supervisory signal, and define a binary classification task, where 0: background / high resistance, 1: low resistance target;

[0031] Step 2: Preprocess and enhance the resistivity thermodynamic map in sequence;

[0032] The preprocessing method is:

[0033] The resistivity thermogram is scaled to a standard size and normalized to eliminate sensor gain differences, allowing the model to focus on relative resistivity changes rather than absolute values.

[0034] The enhanced methods are:

[0035] By random rotation and translation, the detection angle offset or positioning error of the vehicle-mounted equipment when moving along the embankment side line is simulated;

[0036] By flipping, the model can cope with changes in the dike direction, such as symmetrical structures on the left and right banks, and enhance the model's learning of direction invariance;

[0037] Gaussian noise processing.

[0038] The rotation angle range is preferably: ±20°; the translation range is preferably: ±10°; and the standard deviation of Gaussian noise is preferably: 0.1.

[0039] Step 3: Build the FCN model, which includes: input layer, encoder layer, decoder layer and output layer;

[0040] Among them, the input of the input layer is set to 256*256*3 image input; the encoder layer includes: 64-channel convolution, 2*2 maximum pooling, 128-channel convolution, 2*2 maximum pooling, 256-channel convolution and 2*2 maximum pooling; the decoder layer includes: 128-channel deconvolution, 64-channel deconvolution and 32-channel deconvolution.

[0041] The encoder layer extracts the spatial distribution features of resistivity at different scales through multi-level convolution and pooling. For example, the shallow network representing the edge of the hole may detect local conductivity mutations, while the deep network representing the piping zone may identify large-scale low-resistance patterns.

[0042] The decoder layer uses deconvolution operations to gradually restore the size of the feature map. Combined with jump connections, it can fuse multi-scale information and improve the detection accuracy of small targets, such as holes with a diameter of less than 1m.

[0043] The loss function of the constructed FCN model is: binary cross entropy, which can minimize the difference between the predicted mask and the true label. The mathematical form is:

[0044]

[0045] Where y i ∈{0,1} is the true label; p i ∈[0,1] is the model prediction probability.

[0046] When constructing the FCN model, the optimal classification threshold is also determined through ROC curve analysis. Lowering the threshold can increase sensitivity and is suitable for high-risk embankment sections. The optimal range of the classification threshold is: 0.05-0.15.

[0047] The output of the FCN model is a masked resistivity thermogram showing low-resistivity areas;

[0048] Specific output includes:

[0049] Original image: Resistivity heat map: red indicates high resistance, green indicates low resistance;

[0050] Prediction mask (binarization mask): Grayscale image showing the low-resistance area determined by the model;

[0051] Overlay effect: A semi-transparent color mask is superimposed on the original image to intuitively display the location of hidden dangers;

[0052] Color mapping: Red indicates high-resistance areas, i.e., dense soil layers; green indicates low-resistance areas, i.e., water-rich anomalies. Drilling can then be used to verify whether these areas are piping.

[0053] Step 4: Train the constructed FCN model and use the resistivity thermogram processed in step 2 as the input of the FCN model to obtain the detection results.

[0054] While the embodiments of the present invention have been described, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are within the scope defined by the appended claims.

Claims

1. A method for detecting hidden dangers in dikes, characterized in that: The steps include: Step 1: Obtain the resistivity thermodynamic map of the dam; Step 2: Preprocess and enhance the resistivity thermodynamic map in sequence; Step 3: Construct an FCN model. The FCN model includes an input layer, an encoder layer, a decoder layer, and an output layer. The output of the FCN model is a resistivity heat map showing the low-resistance area covered with a mask. Step 4: Train the constructed FCN model and use the resistivity thermogram processed in step 2 as the input of the FCN model to obtain the detection results.

2. The dike hidden danger detection method according to claim 1, characterized in that: In step 1, the specific method for obtaining the resistivity thermodynamic map is: By installing a detection device equipped with a transient electromagnetic instrument, detection is carried out at a certain speed interval along the side line of the embankment to obtain the secondary eddy current field attenuation curve, and the conductivity distribution of the underground medium is inverted. Based on the inverted conductivity distribution data, the resistivity thermodynamic map of the embankment is generated.

3. The dike hidden danger detection method according to claim 1 or 2, characterized in that: In step 2, the resistivity thermodynamic map is pre-processed as follows: The resistivity thermodynamic map is scaled to a standard size and normalized.

4. The dike hidden danger detection method according to claim 3, characterized in that: In step 2, the method for enhancing the resistivity thermodynamic map is: The preprocessed resistivity thermogram is enhanced by random rotation, translation, flipping and Gaussian noise.

5. The dike hidden danger detection method according to claim 4, characterized in that: The rotation angle range is: ±20°; the translation range is: ±10°; and the standard deviation of Gaussian noise is: 0.

1.

6. The dike hidden danger detection method according to claim 1, characterized in that: In step 3, the loss function of the constructed FCN model is: binary cross entropy.

7. The dike hidden danger detection method according to claim 6, characterized in that: The specific formula of the loss function is: Where y i ∈{0,1} is the true label; p i ∈[0,1] is the model prediction probability.

8. The dike hidden danger detection method according to claim 1, characterized in that: In the step 3, when constructing the FCN model, the classification threshold of the FCN model is optimized by ROC.

9. The dike hidden danger detection method according to claim 8, characterized in that: The classification threshold of the optimized FCN model is 0.05-0.

15.

10. The dike hidden danger detection method according to claim 1, characterized in that: In step 3, the mask is a binary mask.

Images