A detection method for mountain displacement monitoring using unmanned aerial vehicles

Abstract

The present invention discloses a detection method for monitoring mountain displacement using an unmanned aerial vehicle (UAV), belonging to the field of image processing technology. The method first divides the mountain monitoring area into multiple sub-areas and uses an UAV to collect mountain surface images in each sub-area. Next, the soil color coefficients of the image pixels are extracted to obtain a soil color distribution map, which is divided into a normal set and an abnormal set. An overall importance distribution map is obtained based on the difference between the abnormal set soil color coefficient and the normal set's entire soil color mean, and a neighborhood importance distribution map is obtained based on the difference between the abnormal set soil color coefficient and the neighborhood soil color mean. Finally, a monitoring neural network model is used to process the abnormal set soil color distribution map, the overall importance distribution map, and the neighborhood importance distribution map, and output a landslide displacement risk score, thereby accurately assessing landslide displacement risk.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a detection method for mountain displacement monitoring using an unmanned aerial vehicle (UAV). Background Art

[0002] In the field of geological disaster monitoring, mountain displacement monitoring is crucial for protecting people's lives and property and ensuring the stability of infrastructure. Traditional mountain displacement monitoring methods, such as ground-based manual inspections, are not only labor-intensive and financially intensive, but also inefficient and significantly constrained by factors such as terrain and weather, making it difficult to achieve real-time, comprehensive monitoring of large mountain areas. With technological advancements, drone technology is gradually being applied to mountain monitoring. With its advantages of high mobility and wide monitoring range, drones can quickly acquire image data from mountain surfaces and are widely used in mountain displacement monitoring.

[0003] In the field of image detection applications, the rapid development of artificial intelligence technology has led to the development of a variety of advanced methods. Deep learning algorithms enable the automatic identification of geological hazards such as landslides and rockfalls. When a mountain shifts, the underlying soil is brought to the surface, resulting in significant differences between the shifted area and the surrounding area. Currently, image-based mountain displacement monitoring technology primarily collects images of the monitored area and assesses the risk of mountain displacement based on the image's color features. However, this assessment method, which relies solely on image color features, suffers from insufficient accuracy in assessing landslide and displacement risks. Summary of the Invention

[0004] In response to the above-mentioned deficiencies in the prior art, the present invention provides a detection method for mountain displacement monitoring using a drone, which solves the problem of insufficient accuracy in landslide displacement risk assessment in the prior art.

[0005] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: a detection method for mountain displacement monitoring using a drone, comprising the following steps:

[0006] S1. Divide the mountain monitoring area into multiple sub-areas and use a drone to collect a mountain surface image for each sub-area;

[0007] S2. extracting the soil color coefficient of each pixel point on each mountain surface image to obtain a soil color distribution map;

[0008] S3, divide all soil color distribution maps into normal set and abnormal set;

[0009] S4. Obtain an overall importance distribution map based on the difference between the soil color coefficient of the soil color distribution map in the abnormal concentration and the average soil color value of the entire area in the normal concentration;

[0010] S5. Obtain a neighborhood importance distribution map based on the difference between the soil color coefficient of the abnormal concentration soil color distribution map and the neighborhood soil color mean of the neighborhood soil color distribution map;

[0011] S6. Use the monitoring neural network model to process the abnormal concentrated soil color distribution map, the overall importance distribution map and the neighborhood importance distribution map to obtain the landslide displacement risk score.

[0012] Furthermore, S2 includes the following sub-steps:

[0013] S21. In each mountain surface image, subtract the G channel value from the R channel value of each pixel to obtain the RG channel difference;

[0014] S22. In each mountain surface image, add the R channel value and the G channel value of each pixel to obtain the RG channel sum;

[0015] S23. For the same pixel point, take the ratio of the RG channel difference to the RG channel sum as the earth color coefficient;

[0016] S24. Arrange the soil color coefficient of each pixel point according to the corresponding pixel point position to obtain a soil color distribution map.

[0017] Furthermore, S3 includes the following sub-steps:

[0018] S31. Average the soil color coefficients in all soil color distribution maps of the mountain monitoring area to obtain the average soil color value of the entire area;

[0019] S32, averaging the soil color coefficients in each soil color distribution map to obtain a sub-area soil color mean;

[0020] S33, calculating the difference between the soil color mean of each sub-area and the soil color mean of the entire area;

[0021] S34. When the gap is greater than the gap threshold, the corresponding soil color distribution map is classified into an abnormal set;

[0022] S35. Classify other soil color distribution maps into the normal set.

[0023] Furthermore, S4 includes the following sub-steps:

[0024] S41. averaging the soil color coefficients of all soil color distribution maps in the normal set to obtain a normal soil color mean;

[0025] S42. Calculate the overall importance factor based on the difference between the soil color coefficient of each pixel point in the abnormal concentrated soil color distribution map and the normal soil color mean value, and arrange the overall importance factor of each pixel point according to the corresponding pixel point position to obtain the overall importance distribution map.

[0026] Furthermore, S5 includes the following sub-steps:

[0027] S51, extracting a normal concentrated soil color distribution map that is spatially in contact with the abnormal concentrated soil color distribution map to obtain a neighborhood soil color distribution map;

[0028] S52, averaging the soil color coefficients of all neighborhood soil color distribution maps to obtain a neighborhood soil color mean;

[0029] S53. Calculate the neighborhood importance factor based on the difference between the soil color coefficient of each pixel point in the abnormal concentration soil color distribution map and the corresponding neighborhood soil color mean value, and arrange the neighborhood importance factor of each pixel point according to the corresponding pixel point position to obtain a neighborhood importance distribution map.

[0030] Furthermore, the specific process of calculating the overall importance factor in S42 includes: subtracting the normal soil color mean from the soil color coefficient of each pixel to obtain an overall difference, extracting the maximum soil color coefficient and the minimum soil color coefficient of all soil color distribution maps in the mountain monitoring area, subtracting the minimum soil color coefficient from the maximum soil color coefficient to obtain the maximum difference, and using the ratio of the overall difference to the maximum difference as the overall importance factor;

[0031] The specific process of calculating the neighborhood importance factor in S53 includes: using the soil color coefficient of each pixel point to subtract the neighborhood soil color mean to obtain the neighborhood difference, and taking the ratio of the neighborhood difference to the maximum difference as the neighborhood importance factor.

[0032] Furthermore, the monitoring neural network model in S6 includes: an importance image fusion enhancement unit, a first convolutional layer, a second convolutional layer, an attention feature fusion unit AFF, a deep feature acquisition unit and a fully connected layer;

[0033] The first input of the importance image fusion enhancement unit is used to input the overall importance distribution map, the second input is used to input the abnormal concentration soil color distribution map, the third input is used to input the neighborhood importance distribution map, and the output is connected to the input of the first convolution layer and the input of the second convolution layer respectively.

[0034] The input end of the attention feature fusion unit AFF is connected to the output end of the first convolutional layer and the output end of the second convolutional layer respectively, and its output end is connected to the input end of the deep feature acquisition unit;

[0035] The input end of the fully connected layer is connected to the output end of the deep feature acquisition unit, and its output end serves as the output end of the monitoring neural network model.

[0036] Furthermore, the importance image fusion enhancement unit includes: a first shallow feature acquisition module, a second shallow feature acquisition module, a third shallow feature acquisition module and an image feature enhancement module;

[0037] The input end of the first shallow feature acquisition module serves as the first input end of the importance image fusion enhancement unit; the input end of the second shallow feature acquisition module serves as the second input end of the importance image fusion enhancement unit; the input end of the third shallow feature acquisition module serves as the third input end of the importance image fusion enhancement unit;

[0038] The input end of the image feature enhancement module is connected to the output end of the first shallow feature acquisition module, the output end of the second shallow feature acquisition module and the output end of the third shallow feature acquisition module respectively, and its output end serves as the output end of the importance image fusion enhancement unit.

[0039] Furthermore, the expression of the image feature enhancement module is: , where Z is the image fusion enhancement feature, C1 is the output of the first shallow feature acquisition module, C2 is the output of the second shallow feature acquisition module, and C3 is the output of the third shallow feature acquisition module. For element-wise multiplication, is element-wise addition.

[0040] Furthermore, the deep feature acquisition unit includes: a maximum pooling layer, an average pooling layer, a Concat layer and a third convolutional layer;

[0041] The input of the maximum pooling layer is connected to the input of the average pooling layer and serves as the input of the deep feature acquisition unit; the input of the Concat layer is connected to the output of the maximum pooling layer and the output of the average pooling layer respectively, and its output is connected to the input of the third convolutional layer; the output of the third convolutional layer serves as the output of the deep feature acquisition unit.

[0042] The beneficial effects of the present invention are:

[0043] The method first divides the mountain monitoring area into multiple sub-areas. Using drones, the method collects mountain surface images from each sub-area and extracts the soil color coefficient for each pixel to highlight soil characteristics. Subsequently, all soil color distribution maps are divided into a normal set and an abnormal set. Maps in the abnormal set indicate areas where landslide displacement is likely to occur.

[0044] On this basis, the analysis was further deepened through two dimensions: first, the difference between the soil color distribution map in the abnormal concentration and the average soil color of the normal set in the entire area was calculated to generate an overall importance distribution map, which can intuitively reflect the degree of deviation of the soil color coefficient of the abnormal area relative to the overall average state of the mountain; second, the difference between the soil color coefficient of the soil color distribution map in the abnormal concentration and the neighborhood soil color average of the neighborhood soil color distribution map was calculated to obtain a neighborhood importance distribution map, which is used to reflect the difference in the soil color coefficient of the abnormal area relative to the average state of the surrounding normal neighborhood.

[0045] Finally, a monitoring neural network model was used to process the soil color distribution map, the overall importance distribution map, and the neighborhood importance distribution map. The features of the soil color distribution map were enhanced using the corresponding features of the overall importance distribution map and the neighborhood importance distribution map. This multi-dimensional analysis method breaks through the limitations of existing technologies that rely solely on image color feature assessment. It can utilize the differences between displaced areas and normal areas and neighborhoods, effectively addressing the inaccuracy of traditional assessment methods and improving the accuracy of landslide displacement risk assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flow chart of a detection method for mountain displacement monitoring using UAVs;

[0047] Figure 2 This is a schematic diagram of the division of mountain monitoring areas;

[0048] Figure 3 It is a schematic diagram of the positional relationship of the soil color distribution maps that are in contact in space;

[0049] Figure 4 It is a structural diagram of the monitoring neural network model;

[0050] Figure 5 Schematic diagram of the structure of the importance image fusion enhancement unit;

[0051] Figure 6 Schematic diagram of the structure of the deep feature acquisition unit;

[0052] Figure 7 This is a structural diagram of the first shallow feature acquisition module, the second shallow feature acquisition module, and the third shallow feature acquisition module. DETAILED DESCRIPTION

[0053] The specific embodiments of the present invention are described below to facilitate understanding of the present invention by those skilled in the art. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the appended claims, these changes are obvious, and all inventions and creations utilizing the concepts of the present invention are protected.

[0054] like Figure 1 As shown, a detection method for mountain displacement monitoring using a drone includes the following steps:

[0055] S1. Divide the mountain monitoring area into multiple sub-areas and use a drone to collect a mountain surface image for each sub-area;

[0056] S2. extracting the soil color coefficient of each pixel point on each mountain surface image to obtain a soil color distribution map;

[0057] S3, divide all soil color distribution maps into normal set and abnormal set;

[0058] S4. Obtain an overall importance distribution map based on the difference between the soil color coefficient of the soil color distribution map in the abnormal concentration and the average soil color value of the entire area in the normal concentration;

[0059] S5. Obtain a neighborhood importance distribution map based on the difference between the soil color coefficient of the abnormal concentration soil color distribution map and the neighborhood soil color mean of the neighborhood soil color distribution map;

[0060] S6. Use the monitoring neural network model to process the abnormal concentrated soil color distribution map, the overall importance distribution map and the neighborhood importance distribution map to obtain the landslide displacement risk score.

[0061] In the present invention, Figure 2 As shown in Figure 1, S1 specifically involves dividing the mountain monitoring area into M×N sub-areas, where M is the number of rows and N is the number of columns. For each sub-area, the drone collects an image of the mountain surface from directly above.

[0062] In this embodiment, S2 includes the following sub-steps:

[0063] S21. In each mountain surface image, subtract the G channel value from the R channel value of each pixel to obtain the RG channel difference;

[0064] S22. In each mountain surface image, add the R channel value and the G channel value of each pixel to obtain the RG channel sum;

[0065] S23. For the same pixel point, take the ratio of the RG channel difference to the RG channel sum as the earth color coefficient;

[0066] S24. Arrange the soil color coefficient of each pixel point according to the corresponding pixel point position to obtain a soil color distribution map.

[0067] In this embodiment, the formula of the earth color coefficient is: θ i =(R i -G i ) / (R i +G i ), where θ i is the earth color coefficient of the i-th pixel, R i is the R channel value of the i-th pixel, G i is the G channel value of the i-th pixel, where i is a positive integer.

[0068] The present invention converts three-dimensional RGB data into one-dimensional θ iThe value compresses the data volume while retaining key features, highlighting the differences before and after the displacement. Before the landslide occurred, the mountain was covered with vegetation, and the G channel value was greater than the R channel value. After the landslide occurred, when the inner soil was turned to the surface, the R channel value was greater than the G channel value; θ i When θ > 0, the pixel point is biased towards the soil feature (θ i The larger the value, the more obvious the soil color). i When ≤0, the pixel point is biased towards vegetation features.

[0069] The technical solution of the present invention is applicable to areas where mountains are covered with vegetation.

[0070] In this embodiment, S3 includes the following sub-steps:

[0071] S31. Average the soil color coefficients in all soil color distribution maps of the mountain monitoring area to obtain the average soil color value of the entire area;

[0072] S32, averaging the soil color coefficients in each soil color distribution map to obtain a sub-area soil color mean;

[0073] S33, calculating the difference between the soil color mean of each sub-area and the soil color mean of the entire area;

[0074] S34. When the gap is greater than the gap threshold, the corresponding soil color distribution map is classified into an abnormal set;

[0075] S35. Classify other soil color distribution maps into the normal set.

[0076] The present invention establishes a global benchmark for soil characteristics for the entire mountain monitoring area by calculating the average soil color of the entire area. Then, the average soil color of the sub-area of ​​each soil color distribution map is obtained. Then, by calculating the difference between the average soil color of the sub-area and the average of the entire area, the degree of deviation of the soil characteristics of the sub-area relative to the overall mountain is quantified. When the difference exceeds the preset difference threshold, the corresponding soil color distribution map is classified as an abnormal set, otherwise it is classified as a normal set.

[0077] In this embodiment, the specific process of calculating the difference between the soil color mean of each sub-area and the soil color mean of the entire area in S33 includes: subtracting the soil color mean of the entire area from the soil color mean of each sub-area to obtain a difference, and taking the absolute value of the difference as the difference.

[0078] In this embodiment, the gap threshold is set based on experiments or experience.

[0079] In this embodiment, S4 includes the following sub-steps:

[0080] S41. averaging the soil color coefficients of all soil color distribution maps in the normal set to obtain a normal soil color mean;

[0081] S42. Calculate the overall importance factor based on the difference between the soil color coefficient of each pixel point in the abnormal concentrated soil color distribution map and the normal soil color mean value, and arrange the overall importance factor of each pixel point according to the corresponding pixel point position to obtain the overall importance distribution map.

[0082] The present invention calculates the mean value of the soil color coefficient of all soil color distribution maps in the normal set to obtain the soil color coefficient of the mountain surface that can represent the area where landslide displacement has not occurred. A soil color distribution map labeled A is randomly selected from the abnormal set. According to the difference between the soil color coefficient of each pixel point on the soil color distribution map A and the mean value of the normal soil color, the overall importance factor is calculated. This factor directly reflects the deviation of the soil characteristics of each position in the abnormal area from the normal state of the whole mountain. The larger the difference, the more significant the difference between the soil characteristics corresponding to the pixel point and the normal state, and the more likely it is that the inner soil layer is exposed due to mountain displacement.

[0083] In this embodiment, S5 includes the following sub-steps:

[0084] S51, extracting a normal concentrated soil color distribution map that is spatially in contact with the abnormal concentrated soil color distribution map to obtain a neighborhood soil color distribution map;

[0085] S52, averaging the soil color coefficients of all neighborhood soil color distribution maps to obtain a neighborhood soil color mean;

[0086] S53. Calculate the neighborhood importance factor based on the difference between the soil color coefficient of each pixel point in the abnormal concentration soil color distribution map and the corresponding neighborhood soil color mean value, and arrange the neighborhood importance factor of each pixel point according to the corresponding pixel point position to obtain a neighborhood importance distribution map.

[0087] The present invention extracts the soil color distribution map of the normal concentration that is spatially in contact with the soil color distribution map of the abnormal concentration, and uses this as the benchmark value of the surrounding environment of the abnormal area. The neighborhood importance factor is calculated based on the difference between the soil color coefficient of each pixel point in the abnormal concentration and the mean soil color of the neighborhood. This factor quantifies the degree of deviation of soil characteristics at each position in the abnormal area relative to the adjacent normal area. Compared with global comparison, neighborhood comparison can better reflect the differences in soil characteristics caused by local microtopography changes or small-scale displacements. For example, when the difference between the abnormal area and the neighborhood is significant, it indicates that the soil characteristics of the area have undergone a mutation compared with the surrounding environment, which is very likely to be newly exposed soil caused by recent mountain displacement.

[0088] exist Figure 3 Among them, B1, B2, B3, B4, B5, B6, B7, and B8 are eight soil color distribution maps in the normal concentration that are spatially in contact with the abnormally concentrated soil color distribution map A.

[0089] In this embodiment, the specific process of calculating the overall importance factor in S42 includes: subtracting the normal soil color mean from the soil color coefficient of each pixel to obtain an overall difference, extracting the maximum soil color coefficient and the minimum soil color coefficient of all soil color distribution maps in the mountain monitoring area, subtracting the minimum soil color coefficient from the maximum soil color coefficient to obtain the maximum difference, and using the ratio of the overall difference to the maximum difference as the overall importance factor;

[0090] The specific process of calculating the neighborhood importance factor in S53 includes: using the soil color coefficient of each pixel point to subtract the neighborhood soil color mean to obtain the neighborhood difference, and taking the ratio of the neighborhood difference to the maximum difference as the neighborhood importance factor.

[0091] In this embodiment, the formula for calculating the overall importance factor is: , where γ i,w is the overall importance factor of the i-th pixel, S i is the earth color coefficient of the i-th pixel, S w,avg is the normal soil color mean, S max is the maximum earth color coefficient, S min is the minimum earth color coefficient.

[0092] The formula for calculating the neighborhood importance factor is: , where γ i,n is the neighborhood importance factor of the i-th pixel, S n,avg is the neighborhood soil color mean.

[0093] In this embodiment, a constant may be selected to normalize the numerator of the denominator of the formula for calculating the overall importance factor and the formula for calculating the neighborhood importance factor.

[0094] like Figure 4 As shown, the monitoring neural network model in S6 includes: importance image fusion enhancement unit, first convolutional layer, second convolutional layer, attention feature fusion unit AFF, deep feature acquisition unit and fully connected layer;

[0095] The first input of the importance image fusion enhancement unit is used to input the overall importance distribution map, the second input is used to input the abnormal concentration soil color distribution map, the third input is used to input the neighborhood importance distribution map, and the output is connected to the input of the first convolution layer and the input of the second convolution layer respectively.

[0096] The input end of the attention feature fusion unit AFF is connected to the output end of the first convolutional layer and the output end of the second convolutional layer respectively, and its output end is connected to the input end of the deep feature acquisition unit;

[0097] The input end of the fully connected layer is connected to the output end of the deep feature acquisition unit, and its output end serves as the output end of the monitoring neural network model.

[0098] The convolution kernel size of the first convolution layer is 3×3, and the convolution kernel size of the second convolution layer is 5×5.

[0099] The present invention integrates three types of data: overall importance (global deviation), soil color distribution (original features), and neighborhood importance (local deviation) through the importance image fusion enhancement unit. The first and second convolutional layers process the fused features in parallel to improve the richness of feature representation. The attention feature fusion unit AFF allows the model to focus on risk-significant areas. The deep feature acquisition unit further extracts deep features and outputs the landslide displacement risk score through the fully connected layer.

[0100] The full English name of the attention feature fusion module AFF is Attentional Feature Fusion.

[0101] In this embodiment, when the soil color distribution map A is input into the monitoring neural network model, the overall importance distribution map input to the model is the overall importance distribution map obtained by the difference between the soil color distribution map A and the normal soil color mean, and the neighborhood importance distribution map input to the model is the neighborhood importance distribution map obtained by the soil color distribution map A and the neighborhood soil color mean.

[0102] like Figure 5 As shown, the importance image fusion enhancement unit includes: a first shallow feature acquisition module, a second shallow feature acquisition module, a third shallow feature acquisition module and an image feature enhancement module;

[0103] The input end of the first shallow feature acquisition module serves as the first input end of the importance image fusion enhancement unit; the input end of the second shallow feature acquisition module serves as the second input end of the importance image fusion enhancement unit; the input end of the third shallow feature acquisition module serves as the third input end of the importance image fusion enhancement unit;

[0104] The input end of the image feature enhancement module is connected to the output end of the first shallow feature acquisition module, the output end of the second shallow feature acquisition module and the output end of the third shallow feature acquisition module respectively, and its output end serves as the output end of the importance image fusion enhancement unit.

[0105] In this embodiment, the expression of the image feature enhancement module is: , where Z is the image fusion enhancement feature, C1 is the output of the first shallow feature acquisition module, C2 is the output of the second shallow feature acquisition module, and C3 is the output of the third shallow feature acquisition module. For element-wise multiplication, Add element-wise.

[0106] In the present invention, the first shallow feature acquisition module is used to extract the shallow feature C1 of the overall importance distribution map, the second shallow feature acquisition module is used to extract the shallow feature C2 of the abnormal concentration soil color distribution map, and the third shallow feature acquisition module is used to extract the shallow feature C3 of the neighborhood importance distribution map. The shallow feature C1 and the shallow feature C2 are element-wise multiplied, and the shallow feature C2 of the soil color distribution map is enhanced by the shallow feature C1 of the overall importance distribution map. The shallow feature C3 of the neighborhood importance distribution map is element-wise multiplied with the shallow feature C2, and the shallow feature C2 of the soil color distribution map is enhanced by the shallow feature C3 of the neighborhood importance distribution map. This allows the model to focus more on the true displacement-related soil color change characteristics, filter out irrelevant soil color fluctuations such as vegetation interference, and then add them element-wise to fuse the enhanced features of the two aspects.

[0107] like Figure 6 As shown, the deep feature acquisition unit includes: maximum pooling layer, average pooling layer, Concat layer and the third convolution layer;

[0108] The input of the maximum pooling layer is connected to the input of the average pooling layer and serves as the input of the deep feature acquisition unit; the input of the Concat layer is connected to the output of the maximum pooling layer and the output of the average pooling layer respectively, and its output is connected to the input of the third convolutional layer; the output of the third convolutional layer serves as the output of the deep feature acquisition unit.

[0109] In this embodiment, the convolution kernel size of the third convolutional layer is 7×7.

[0110] The present invention extracts the most significant and representative features through the maximum pooling layer, extracts the mean features through the average pooling layer to reflect the overall feature level, uses the Concat layer to splice the output of the maximum pooling layer and the output of the average pooling layer, and then further performs convolution operation on the spliced ​​features through the third convolution layer.

[0111] In this embodiment, if Figure 7 As shown, the first shallow feature acquisition module, the second shallow feature acquisition module and the third shallow feature acquisition module all include: a fourth convolutional layer, a fifth convolutional layer and a sixth convolutional layer connected in sequence, the convolution kernel size of the fourth convolutional layer is 1×1, the convolution kernel size of the fifth convolutional layer is 3×3, and the convolution kernel size of the sixth convolutional layer is 1×1.

[0112] In this embodiment, when the landslide displacement risk score is between 80 and 100, there is large-scale landslide displacement in the area corresponding to the soil color distribution map; when the landslide displacement risk score is between 60 and 80, there is moderate landslide displacement in the area corresponding to the soil color distribution map; when the landslide displacement risk score is between 40 and 60, there is mild landslide displacement in the area corresponding to the soil color distribution map; when the landslide displacement risk score is between 0 and 40, there is no obvious landslide displacement in the area corresponding to the soil color distribution map.

[0113] The method first divides the mountain monitoring area into multiple sub-areas. Using drones, the method collects mountain surface images from each sub-area and extracts the soil color coefficient for each pixel to highlight soil characteristics. Subsequently, all soil color distribution maps are divided into a normal set and an abnormal set. Maps in the abnormal set indicate areas where landslide displacement is likely to occur.

[0114] On this basis, the analysis was further deepened through two dimensions: first, the difference between the soil color distribution map in the abnormal concentration and the average soil color of the normal set in the entire area was calculated to generate an overall importance distribution map, which can intuitively reflect the degree of deviation of the soil color coefficient of the abnormal area relative to the overall average state of the mountain; second, the difference between the soil color coefficient of the soil color distribution map in the abnormal concentration and the neighborhood soil color average of the neighborhood soil color distribution map was calculated to obtain a neighborhood importance distribution map, which is used to reflect the difference in the soil color coefficient of the abnormal area relative to the average state of the surrounding normal neighborhood.

[0115] Finally, a monitoring neural network model was used to process the soil color distribution map, the overall importance distribution map, and the neighborhood importance distribution map. The features of the soil color distribution map were enhanced using the corresponding features of the overall importance distribution map and the neighborhood importance distribution map. This multi-dimensional analysis method breaks through the limitations of existing technologies that rely solely on image color feature assessment. It can utilize the differences between displaced areas and normal areas and neighborhoods, effectively addressing the inaccuracy of traditional assessment methods and improving the accuracy of landslide displacement risk assessment.

[0116] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A detection method for mountain displacement monitoring using an unmanned aerial vehicle, characterized in that: The following steps are involved: S1. Divide the mountain monitoring area into multiple sub-areas and use a drone to collect a mountain surface image for each sub-area; S2. extracting the soil color coefficient of each pixel point on each mountain surface image to obtain a soil color distribution map; S3, divide all soil color distribution maps into normal set and abnormal set; S4. Obtain an overall importance distribution map based on the difference between the soil color coefficient of the soil color distribution map in the abnormal concentration and the average soil color value of the entire area in the normal concentration; S5. Obtain a neighborhood importance distribution map based on the difference between the soil color coefficient of the abnormal concentration soil color distribution map and the neighborhood soil color mean of the neighborhood soil color distribution map; S6. Use the monitoring neural network model to process the abnormal concentration soil color distribution map, the overall importance distribution map, and the neighborhood importance distribution map to obtain the landslide displacement risk score; The S2 comprises the following sub-steps: S21. In each mountain surface image, subtract the G channel value from the R channel value of each pixel to obtain the RG channel difference; S22. In each mountain surface image, add the R channel value and the G channel value of each pixel to obtain the RG channel sum; S23. For the same pixel point, take the ratio of the RG channel difference to the RG channel sum as the earth color coefficient; S24. Arrange the soil color coefficient of each pixel point according to the corresponding pixel point position to obtain a soil color distribution map.

2. The method for monitoring mountain displacement using a drone according to claim 1, characterized in that: The S3 includes the following sub-steps: S31. Average the soil color coefficients in all soil color distribution maps of the mountain monitoring area to obtain the average soil color value of the entire area; S32, averaging the soil color coefficients in each soil color distribution map to obtain a sub-area soil color mean; S33, calculating the difference between the soil color mean of each sub-area and the soil color mean of the entire area; S34. When the gap is greater than the gap threshold, the corresponding soil color distribution map is classified into an abnormal set; S35. Classify other soil color distribution maps into the normal set.

3. The method for monitoring mountain displacement using an unmanned aerial vehicle according to claim 1, wherein: The S4 comprises the following sub-steps: S41. averaging the soil color coefficients of all soil color distribution maps in the normal set to obtain a normal soil color mean; S42. Calculate the overall importance factor based on the difference between the soil color coefficient of each pixel point in the abnormal concentrated soil color distribution map and the normal soil color mean value, and arrange the overall importance factor of each pixel point according to the corresponding pixel point position to obtain the overall importance distribution map.

4. The method for monitoring mountain displacement using an unmanned aerial vehicle according to claim 3, wherein: The S5 comprises the following sub-steps: S51, extracting a normal concentrated soil color distribution map that is spatially in contact with the abnormal concentrated soil color distribution map to obtain a neighborhood soil color distribution map; S52, averaging the soil color coefficients of all neighborhood soil color distribution maps to obtain a neighborhood soil color mean; S53. Calculate the neighborhood importance factor based on the difference between the soil color coefficient of each pixel point in the abnormal concentration soil color distribution map and the corresponding neighborhood soil color mean value, and arrange the neighborhood importance factor of each pixel point according to the corresponding pixel point position to obtain a neighborhood importance distribution map.

5. The method for monitoring mountain displacement using an unmanned aerial vehicle according to claim 3 or 4, characterized in that: The specific process of calculating the overall importance factor in S42 includes: subtracting the normal soil color mean from the soil color coefficient of each pixel to obtain an overall difference, extracting the maximum soil color coefficient and the minimum soil color coefficient of all soil color distribution maps in the mountain monitoring area, subtracting the minimum soil color coefficient from the maximum soil color coefficient to obtain the maximum difference, and using the ratio of the overall difference to the maximum difference as the overall importance factor; The specific process of calculating the neighborhood importance factor in S53 includes: using the soil color coefficient of each pixel point to subtract the neighborhood soil color mean to obtain the neighborhood difference, and taking the ratio of the neighborhood difference to the maximum difference as the neighborhood importance factor.

6. The method for monitoring mountain displacement using an unmanned aerial vehicle according to claim 1, wherein: The monitoring neural network model in S6 includes: an importance image fusion enhancement unit, a first convolutional layer, a second convolutional layer, an attention feature fusion unit AFF, a deep feature acquisition unit and a fully connected layer; The first input of the importance image fusion enhancement unit is used to input the overall importance distribution map, the second input is used to input the abnormal concentration soil color distribution map, the third input is used to input the neighborhood importance distribution map, and the output is connected to the input of the first convolution layer and the input of the second convolution layer respectively. The input end of the attention feature fusion unit AFF is connected to the output end of the first convolutional layer and the output end of the second convolutional layer respectively, and its output end is connected to the input end of the deep feature acquisition unit; The input end of the fully connected layer is connected to the output end of the deep feature acquisition unit, and its output end serves as the output end of the monitoring neural network model.

7. The method for monitoring mountain displacement using an unmanned aerial vehicle according to claim 6, wherein: The importance image fusion enhancement unit includes: a first shallow feature acquisition module, a second shallow feature acquisition module, a third shallow feature acquisition module and an image feature enhancement module; The input end of the first shallow feature acquisition module serves as the first input end of the importance image fusion enhancement unit; the input end of the second shallow feature acquisition module serves as the second input end of the importance image fusion enhancement unit; the input end of the third shallow feature acquisition module serves as the third input end of the importance image fusion enhancement unit; The input end of the image feature enhancement module is connected to the output end of the first shallow feature acquisition module, the output end of the second shallow feature acquisition module and the output end of the third shallow feature acquisition module respectively, and its output end serves as the output end of the importance image fusion enhancement unit.

8. The method for monitoring mountain displacement using an unmanned aerial vehicle according to claim 7, wherein: The expression of the image feature enhancement module is: , where Z is the image fusion enhancement feature, C1 is the output of the first shallow feature acquisition module, C2 is the output of the second shallow feature acquisition module, and C3 is the output of the third shallow feature acquisition module. For element-wise multiplication, is element-wise addition.

9. The method for monitoring mountain displacement using an unmanned aerial vehicle according to claim 6, wherein: The deep feature acquisition unit includes: a maximum pooling layer, an average pooling layer, a Concat layer and a third convolutional layer; The input of the maximum pooling layer is connected to the input of the average pooling layer and serves as the input of the deep feature acquisition unit; the input of the Concat layer is connected to the output of the maximum pooling layer and the output of the average pooling layer respectively, and its output is connected to the input of the third convolutional layer; the output of the third convolutional layer serves as the output of the deep feature acquisition unit.

Images