A non-contact intelligent detection method for the working state of an energy dissipator in a flexible slope protection system

Abstract

The present invention relates to the technical field of slope geological disaster prevention and control, and specifically discloses a non-contact intelligent detection method for the working state of an energy dissipator in a slope flexible protection system. The steps include: conducting a pseudo-static tensile test, a full-scale impact test, and field engineering measurements on the energy dissipator to establish an image data set of the working process of the energy dissipator; training a deep neural network for salient object detection using the data set to achieve automatic segmentation of the binary image of the working state of the energy dissipator; using morphological image processing technology to obtain the skeleton of the energy dissipator and using contour detection technology to extract the inner and outer contours of the skeleton of the energy dissipator and automatically calculate the remaining energy dissipation capacity. The method of the present invention solves the problem of non-contact intelligent identification of the working state of the key energy dissipating components of the flexible protection system through artificial intelligence technology, with high accuracy, and realizes the integrated application of visual monitoring of the appearance and morphology of the key components of the protection project and the discrimination of the internal performance.

Description

Technical Field

[0001] The present invention relates to the technical field of slope geological disaster prevention and protection, and specifically relates to a non-contact intelligent detection method for the working state of an energy dissipator of a slope flexible protection system. Background Art

[0002] Flexible protection systems are widely used in the prevention of strong impact natural disasters such as rockfalls, debris flows, sand and wind flows, and snow and wind flows. When the system works, it often faces multiple impacts and needs to be maintained regularly. At present, the post-disaster residual protection ability of the system is mainly determined by manual inspection. When the system works, it mainly consumes impact energy through energy dissipators. The residual protection ability of the system is closely related to the working state of the energy dissipators. One of the most crucial tasks in manual inspection is to determine the working state of the energy dissipators and evaluate their remaining energy dissipation capacity, so as to judge whether the structure needs to be repaired, strengthened or replaced. Due to the highly non-linear characteristics of the operation of energy dissipators, it is often associated with large errors when inspectors evaluate their remaining energy dissipation capacity by manually measuring the stretching amount of the energy dissipators. At the same time, the working environment of flexible protection systems is mostly dangerous areas such as high mountains and steep cliffs. It is very dangerous to collect data at close range after a disaster. Therefore, there is an urgent need for a high-precision non-contact detection method.

[0003] Salient object detection technology is a branch of the object detection field, mainly used to segment the object from the background and output a binary image, that is, the object pixel value is 1 and the background pixel value is 0. Morphological image processing technology is a branch of the computer vision field, mainly used to process binary images to obtain their features. By using a trained deep neural network for salient object detection, the binary image of the energy dissipator can be automatically segmented from the picture. Further, morphological image processing operations such as filtering, medial axis transformation, opening operation, and contour detection are used to obtain the inner and outer contours of the skeleton of the binary image of the energy dissipator. The remaining energy dissipation capacity of the energy dissipator can be automatically calculated through the inner and outer contours of the skeleton of the binary image of the energy dissipator, and precise, non-contact intelligent detection of the working state of the energy dissipator can be achieved. It is of great significance for improving the accuracy of disaster damage assessment of flexible protection systems and reducing safety risks. Summary of the Invention

[0004] To solve the problems existing in the prior art, the present invention provides a non-contact intelligent detection method for the working state of an energy dissipator of a slope flexible protection system, which realizes precise, non-contact intelligent detection of the working state of the energy dissipator, thereby reducing the indirect losses caused by the slow emergency response of the flexible protection structure, and solving the problems of poor accuracy and high risk of the existing manual detection method mentioned in the above background art.

[0005] To achieve the above object, the present invention provides the following technical solution: A non-contact intelligent detection method for the working state of an energy dissipator of a slope flexible protection system, comprising the following steps:

[0006] S1. Conduct the simulated static tensile test, full-scale impact test, and field engineering measurement of the energy dissipator to establish an image dataset of the working process of the energy dissipator;

[0007] S2. Use the dataset to train a deep neural network for salient object detection to achieve automatic segmentation of the binary image of the working state of the energy dissipator;

[0008] S3. Use morphological image processing technology to obtain the skeleton of the energy dissipator and use contour detection technology to extract the inner and outer contours of the skeleton of the energy dissipator to automatically calculate the remaining energy dissipation capacity.

[0009] Preferably, in step S1, it specifically includes the following:

[0010] S11. Conduct the simulated static tensile test of the energy dissipator to obtain the video of the whole process of its tensile deformation, and extract key frames to obtain images of the whole working process;

[0011] S12. Conduct the full-scale impact test of the protective structure and field engineering measurement to obtain the images of the states of the energy dissipator before and after working in multiple backgrounds;

[0012] S13. Perform salient object detection label annotation on the images of the energy dissipator obtained from the quasi-static tensile test, full-scale impact test, and field engineering measurement, and at the same time perform data augmentation and divide them into a training set and a validation set.

[0013] Preferably, in order to obtain sufficient working images of the energy dissipator, images are collected through methods such as conducting the quasi-static tensile test of the energy dissipator, full-scale impact test of the protective structure, and field engineering measurement. The test methods, loading equipment, loading speed, etc. should comply with the requirements of relevant specifications.

[0014] Preferably, in order to improve the generalization performance of the deep neural network model for salient object detection, tensile deformation images of the whole working process of the energy dissipator and multi-background images covering daily use scenarios should be obtained as much as possible.

[0015] Preferably, in step S2, it specifically includes the following:

[0016] S21. Determine the structure of the deep neural network for salient object detection and write the code;

[0017] S22. Use the training set to train the deep neural network for salient object detection and use the validation set for testing;

[0018] S23. Monitor the training error and validation error and adjust the network structure until the error meets the accuracy requirements to obtain a trained neural network model, and use the trained neural network model to achieve automatic segmentation of the binary image of the working state of the energy dissipator.

[0019] Preferably, in step S3, it specifically includes the following:

[0020] S31. Filter the binary image of the energy dissipator output by the deep neural network for salient object detection and perform medial axis transformation to obtain a clear skeleton image;

[0021] S32. Perform opening operation and contour detection on the skeleton image to obtain the lengths of the inner and outer contours of the energy dissipator skeleton;

[0022] S33. Calculate the remaining energy dissipation capacity of the energy dissipator according to the lengths of the inner and outer contours of the energy dissipator skeleton.

[0023] Preferably, in step S13, in order to perform semantic segmentation annotation on the energy dissipator image, the salient object detection label annotation is performed using annotation software such as Labelme, EISeg or LabelImg. To ensure the training effect, the standards are unified during annotation to ensure the accuracy and consistency of the labels. To improve the training accuracy of the deep neural network model for salient object detection, when forming the image dataset, rotation, scaling, and blurring operations are performed on the original images and labels to achieve image data augmentation.

[0024] Preferably, in step S21, in order to ensure the training effect of the deep neural network for salient object detection, the deep neural network for salient object detection is any one of the mature deep neural network structures for salient object detection such as U 2 -Net, F 3 -Net, BASNet, etc.

[0025] Preferably, in step S22, when training the deep neural network for salient object detection, the loss function L is calculated according to the following formula:

[0026]

[0027] where: M represents the number of feature maps, m represents the m-th feature map, represents the weight coefficient of the m-th sub-feature map, represents the loss value of the m-th sub-feature map, w fuse represents the total feature map weight coefficient after feature fusion, l fuse represents the total feature map loss value after feature fusion;

[0028] To evaluate the training effect of the deep neural network for salient object detection, when training the deep neural network for salient object detection, the performance of the model is determined by key evaluation indicators such as F β , MAE, and calculated according to the following formula:

[0029]

[0030] Where: β is the weight factor, Precision is the accuracy, Recall is the recall rate, H and W are the length and width of the image respectively, r and c represent the pixel coordinates in the length and width directions of the image, and P(r, c) and G(r, c) represent the model prediction value and the true label value respectively.

[0031] Preferably, in order to automatically calculate the remaining energy consumption capacity of the energy dissipator, the binary image output by the trained neural network is further processed by morphological image processing technology to obtain the energy dissipator skeleton image, which specifically includes three steps: filtering, medial axis transformation, and opening operation. In step S31, in order to improve the quality of the skeleton image output by the medial axis transformation, the Gaussian filter is selected and calculated according to the following formula to filter the binary image of the energy dissipator output by the deep neural network for saliency target detection using the Python language and the OpenCV software library. In order to obtain the energy dissipator skeleton image, the MedialAxis function in the Scikit-image library is then used for medial axis transformation:

[0032]

[0033] Where: x g and y g respectively represent the horizontal and vertical coordinate distances from the center point of the Gaussian kernel, and δ represents the variance.

[0034] Preferably, in step S32, in order to reduce the calculation error caused by the thickness of the skeleton, based on the Python language and the OpenCV software library, the opening operation is used to unify the thickness of the energy dissipator skeleton; in order to obtain the inner and outer contours of the energy dissipator skeleton image, the Canny algorithm is then used to detect the contours of the energy dissipator skeleton image, and the calculation is carried out according to the following formula:

[0035] θ[i, j] = arctan(P y [i, j] / P x [i, j])

[0036]

[0037] Where: M[i, j] and θ[i, j] respectively represent the gradient amplitude and direction of the pixel point in the image, and Px[i, j] and Py[i, j] respectively represent the partial derivatives of the pixel point in the x and y directions.

[0038] Preferably, in step S33, the remaining energy consumption capacity R of the energy dissipator is calculated through the inner and outer contour lengths output by the contour detection of its skeleton image, as follows:

[0039]

[0040] L0 = η(L R +L C );

[0041] Where: L R represents the length that has not been activated after the energy dissipator actually works, L C represents the elongation after the energy dissipator actually works, L0 represents the length at which the energy dissipator cannot be activated under the limit state, L Outer represents the outer contour length of the energy dissipator skeleton diagram, L Inner represents the inner contour length of the energy dissipator skeleton diagram, and η is the incomplete activation coefficient of the energy dissipator.

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

[0043] 1) The present invention proposes a non-contact intelligent detection method for the working state of the energy dissipator of the slope flexible protection system. For the first time, deep learning technology and morphological image processing technology are adopted in the detection of the energy dissipator of the flexible protection structure, realizing the non-contact intelligent detection of the working state of the energy dissipator, and providing a revolutionary intelligent alternative solution for the current manual detection method;

[0044] 2) The method of the present invention realizes the non-contact detection of the working state of the energy dissipator, without the need to be in close contact with the disaster source, greatly improving the safety of post-disaster manual detection; the method of the present invention realizes the automatic calculation of the remaining energy dissipation capacity of the energy dissipator, and improves the calculation accuracy through the method of precise proportional calculation;

[0045] 3) The non-contact intelligent detection method for the working state of the energy dissipator of the slope flexible protection system described in the present invention has clear logic, sufficient theoretical support, and high operability, providing a revolutionary intelligent alternative solution for the current industry situation in the protection field that strongly relies on manual detection. It promotes the process of "intelligentization", "automation", and "labor reduction" in the protection industry, explores the application of artificial intelligence technology in the protection industry, and is of great significance for promoting the progress of the industry. The present invention has substantial features and progress, has a very broad market application prospect, and is very suitable for popularization and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a schematic flow chart of the non-contact intelligent detection method for the working state of the energy dissipator of the slope flexible protection system of the present invention;

[0047] Figure 2 is a schematic diagram of the pseudo-static tensile test, full-scale impact test, and field engineering measurement of the energy dissipator in the embodiment of the present invention. a is the pseudo-static tensile test, b is the full-scale impact test, and c is the field engineering measurement;

[0048] Figure 3 is a schematic diagram of some images obtained from the pseudo-static tensile test of the energy dissipator in the embodiment of the present invention;

[0049] Figure 4Schematic diagram of some images obtained from the full-scale impact test of the embodiment of the present invention;

[0050] Figure 5 Schematic diagram of some images obtained from the field engineering measurement of the embodiment of the present invention;

[0051] Figure 6 Schematic diagram of some images and labels in the energy dissipator image dataset of the embodiment of the present invention, where a is the energy dissipator image and b is the label;

[0052] Figure 7 Schematic diagram of the structure of the saliency object detection deep neural network U2-Net adopted in the embodiment of the present invention;

[0053] Figure 8 Schematic diagram of the intelligent detection and segmentation results of some images in the validation set of the embodiment of the present invention;

[0054] Figure 9 Schematic diagram of the technical process of morphological image processing of the energy dissipator binary image in the embodiment of the present invention;

[0055] Figure 10 Schematic diagram of the automatic calculation method for the remaining energy dissipation capacity of the energy dissipator in the embodiment of the present invention;

[0056] Figure 11 Schematic diagram of the pseudo-static tensile test equipment for the energy dissipator in the embodiment of the present invention;

[0057] Figure 12 Schematic diagram of the key steps for intelligently detecting the working state of the energy dissipator by using the method of the present invention for the pseudo-static tensile test image of the energy dissipator in the embodiment of the present invention;

[0058] Figure 13 Schematic diagram of the comparison between the remaining energy dissipation capacity of the energy dissipator output by the pseudo-static tensile test and the intelligent detection result of the method of the present invention in the embodiment of the present invention. Detailed implementation manners

[0059] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0060] Please refer to Figures 1-13 , the embodiment of the present invention provides a technical solution: a non-contact intelligent detection method for the working state of the energy dissipator of a slope flexible protection system performs non-contact intelligent detection on the key images and the remaining energy dissipation capacity of the energy dissipator during a pseudo-static tensile test of a certain GS8002 model energy dissipator, as Figure 1 shown, and the steps are as follows:

[0061] Step 1: Conduct pseudo-static tensile tests, full-scale impact tests, and in-situ field measurements of the energy dissipator to establish an image dataset of the energy dissipator's working process, as Figure 2 shown.

[0062] a. Carry out a simulated static tensile test on the energy dissipator to obtain a video of the entire process of its tensile deformation, and extract key frames to obtain images of the entire working process;

[0063] b. Conduct full-scale impact tests on the protective structure and in-situ field measurements to obtain images of the energy dissipator's states before and after working in multiple backgrounds;

[0064] c. Perform significant object detection label annotation on the energy dissipator images obtained from the pseudo-static tensile test, full-scale impact test, and in-situ field measurements. At the same time, perform data augmentation and divide them into a training set and a validation set;

[0065] Step 2: Use the image dataset to train a deep neural network for significant object detection to achieve automatic segmentation of the energy dissipator

[0066] a. Determine the structure of the deep neural network for significant object detection and write the corresponding code;

[0067] b. Use the training set to train the deep neural network for significant object detection and use the validation set for testing;

[0068] c. Monitor the training error and validation error and adjust the network structure until the error meets the accuracy requirements.

[0069] Step 3: Use morphological image processing techniques to obtain the skeleton of the energy dissipator and use contour detection techniques to extract the inner and outer contours of the energy dissipator's skeleton to automatically calculate the remaining energy dissipation capacity

[0070] a. Write code to filter and perform medial axis transformation on the binary image of the energy dissipator output by the deep neural network for significant object detection to obtain a clear skeleton image;

[0071] b. Write code to perform opening operation and contour detection on the skeleton image to obtain the lengths of the inner and outer contours of the energy dissipator's skeleton;

[0072] c. Calculate the remaining energy dissipation capacity of the energy dissipator based on the lengths of the inner and outer contours of the energy dissipator's skeleton.

[0073] The specific operation details are as follows:

[0074] To obtain sufficient working images of the energy dissipator, 10 groups of pseudo-static tensile tests of the energy dissipator were carried out according to the specification standards, obtaining a video of the entire process of its tensile deformation, and extracting key frames to obtain images of the entire working process, as Figure 3As shown, a total of 240 images were obtained; twenty full-scale impact tests on the flexible protection structure were carried out according to the specification standards, and the images of the energy dissipator before and after working under the background of the full-scale impact test were obtained, such as Figure 4 shown, a total of 210; 5 field engineering measurements were carried out, and the images of the energy dissipator before and after working in multiple backgrounds in the actual project were obtained, such as Figure 5 shown, a total of 50. A total of 500 working images of the energy dissipator were obtained through three ways.

[0075] In order to perform semantic segmentation annotation on the energy dissipator images, the Labelme annotation software was selected for annotation. When annotating, the standards should be unified to ensure the accuracy and consistency of the labels. The energy dissipator and the background were separated, and code was written to convert the output json file of the annotation into a binary map label for training, such as Figure 6 shown.

[0076] In order to improve the training accuracy of the deep neural network model for salient object detection, when forming the image dataset, the images and labels were augmented by rotation, rotated by 30°, 60°, 90°, 120°, 150°, and 180° respectively. A total of 3500 images and their corresponding labels were obtained, and 80% of them were divided into the training set for model training, and the remaining 20% were divided into the validation set for model verification.

[0077] In order to ensure the training effect of the deep neural network model for salient object detection, the U 2 -Net deep neural network structure for salient object detection was selected for training. The network structure is as Figure 7 shown, and the publicly available code was modified for the training of this embodiment.

[0078] When training the deep neural network for salient object detection, the loss function L can be calculated according to the following formula:

[0079]

[0080] In the formula: M represents the number of feature maps. In the U 2 -Net structure adopted in this embodiment, M = 6, m represents the m-th feature map, represents the weight coefficient of the m-th sub-feature map, and all take the value of 1, represents the loss value of the m-th sub-feature map, which is calculated using the standard binary cross-entropy. w fuse represents the total feature map weight coefficient after feature fusion, taking the value of 1, and l fuse represents the total feature map loss value after feature fusion, which is calculated using the standard binary cross-entropy.

[0081] In order to evaluate the training effect of the deep neural network for salient object detection, the maximum F β, To determine the performance of the model using key evaluation indicators such as MAE, it can be calculated according to the following formula:

[0082]

[0083] In the formula: β is the weight factor, and the U 2 -Net structure β 2 = 0.3, Precision is the accuracy rate, Recall is the recall rate, H and W are the length and width of the picture respectively, r and c represent the pixel coordinates in the length and width directions of the picture, P(r, c) and G(r, c) represent the model prediction value and the true label value respectively.

[0084] Use the aforementioned validation set to test the performance of the trained saliency object detection neural network in this embodiment. The qualitative evaluation results of some images in the validation set are shown as Figure 8 shown. It can be seen that the trained model can accurately automatically segment the binary map of the energy dissipator; the quantitative evaluation is calculated according to the above indicators, and the maximum F β = 0.953, MAE = 0.0067, indicating that the model has a high accuracy.

[0085] To automatically calculate the remaining energy dissipation capacity of the energy dissipator, the binary map output by the trained neural network is further processed using morphological image processing techniques (specifically including filtering, medial axis transformation, and opening operation) to obtain the skeleton map of the energy dissipator, and contour detection is performed on the skeleton map, as Figure 9 shown.

[0086] To improve the quality of the skeleton map output by the medial axis transformation, Gaussian filtering is selected to filter the binary map of the energy dissipator output by the saliency object detection deep neural network using the Python language and the OpenCV software library. The calculation is performed according to the following formula:

[0087]

[0088] In the formula: x g and y g respectively represent the horizontal and vertical coordinate distances from the center point of the Gaussian kernel, and δ represents the variance with a value of 0.8.

[0089] To obtain the skeleton map of the energy dissipator, the MedialAxis function in the Python language and the Scikit-image library is used for medial axis transformation.

[0090] To reduce the calculation error caused by the thickness of the skeleton, the Python language and the opening operation in the OpenCV software library are used to process the skeleton map to unify the thickness of the skeleton.

[0091] To obtain the inner and outer contours of the energy dissipator skeleton diagram, based on the Python language and the OpenCV software library, the Canny algorithm is used to detect the contours of the energy dissipator skeleton diagram, and the calculation is carried out according to the following formula:

[0092] θ[i,j] = arctan(P y [i,j] / P x [i,j])

[0093]

[0094] In the formula: M[i,j] and θ[i,j] respectively represent the gradient amplitude and direction of the pixel points in the image, and P x [i,j] and P y [i,j] respectively represent the partial derivatives of the pixel points in the x and y directions.

[0095] To automatically calculate the remaining energy dissipation capacity R of the energy dissipator, it can be calculated through the inner and outer contour lengths output by the contour detection of its skeleton diagram. As shown in Figure 10 , this calculation method is equivalent to directly calculating the ratio of the potentially starting length to the total length after the actual operation of the energy dissipator, as shown in the following formula:

[0096]

[0097] L0 = η(L R +L C )

[0098] In the formula: L R represents the length that has not started after the actual operation of the energy dissipator, L C represents the elongation after the actual operation of the energy dissipator, L0 represents the length at which the energy dissipator cannot start under the limit state. For the GS8002 type energy dissipator, L0 = 416.68 mm, and L Outer represents the outer contour length of the energy dissipator skeleton diagram, L Inner represents the inner contour length of the energy dissipator skeleton diagram, and η is the incomplete start coefficient of the energy dissipator. For the GS8002 type energy dissipator, η = 0.298.

[0099] To verify the effectiveness of the method of the present invention, a quasi-static tensile test of the energy dissipator is carried out to perform non-contact intelligent detection on the working state of the energy dissipator after the test. The test equipment is as shown in Figure 11 .

[0100] Select ten pictures with the remaining energy consumption capacity of the test equipment during the test process being 90%, 80%, 70%, 60%, 50%, 40%, 30%, 20%, and 10%, and number them R1 to R10. Input them into the trained saliency object detection neural network to obtain their binary images, and perform Gaussian filtering, medial axis transformation, opening operation, and contour detection on the obtained binary images to obtain the inner and outer contours of the skeleton image, as Figure 12 shown. Finally, write code to automatically calculate the remaining energy consumption capacity of the energy dissipator based on the lengths of the inner and outer contours of the energy dissipator skeleton image.

[0101] To illustrate the accuracy of the method of the present invention, the remaining energy consumption capacity of the energy dissipator detected by intelligent detection is compared with the output results of the test equipment as Figure 13 shown. The error between the intelligent detection and the test results is less than 3%, which proves the accuracy of the method of the present invention.

[0102] The method of the present invention solves the problem of non-contact intelligent identification of the working state of the key energy-consuming components of the flexible protection system through artificial intelligence technology, has high accuracy, and realizes the integrated application of visual monitoring of the appearance and internal performance discrimination of the key components of the protection project.

[0103] Although the present invention has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A non-contact intelligent detection method for the working status of an energy absorber of a slope flexible protection system, characterized in that: The steps include: S1. Conduct quasi-static tensile test, full-scale impact test and field engineering measurement of energy absorber to establish image data set of working process of energy absorber; S2, using the data set to train a deep neural network for salient target detection to achieve automatic segmentation of the binary image of the working state of the energy consumer; S3, using morphological image processing technology to obtain the skeleton of the energy absorber and using contour detection technology to extract the inner and outer contours of the skeleton of the energy absorber to automatically calculate the remaining energy consumption capacity; specifically including the following: S31. Filter and transform the binary image of the energy dissipator output by the deep neural network for salient target detection to obtain a clear skeleton image; use Python language and OpenCV software library to filter the binary image of the energy dissipator output by the deep neural network for salient target detection, and then use the Medial Axis function in the Scikit-image library to perform medial axis transformation; select Gaussian filtering during filtering, and calculate according to the following formula: Where: x g and g They represent the horizontal and vertical coordinate distances from the center point of the Gaussian kernel, and δ represents the variance; S32, performing an opening operation and contour detection on the skeleton image to obtain the inner and outer contour lengths of the energy consumer skeleton; based on the Python language and the OpenCV software library, an opening operation is used to unify the thickness of the energy consumer skeleton, and then the Canny algorithm is used to perform contour detection on the energy consumer skeleton image, and the calculation is performed according to the following formula: θ[i,j]=arctan(P y [i,j] / P x [i,j]) Where: M[i,j] and θ[i,j] represent the gradient magnitude and direction of the pixel in the image, respectively; Px[i,j] and Py[i,j] represent the partial derivatives of the pixel in the x and y directions, respectively; S33, calculating the remaining energy consumption capacity of the energy consumer according to the inner and outer contour lengths of the energy consumer skeleton.

2. The non-contact intelligent detection method for the working state of the energy absorber of the slope flexible protection system according to claim 1 is characterized by: In step S1, the specific steps include: S11. Conduct a quasi-static tensile test on the energy dissipator to obtain a video of the entire tensile deformation process, and extract key frames to obtain images of the entire working process; S12. Carry out full-scale impact tests on protective structures and field engineering measurements to obtain images of the energy absorber's status before and after working with multiple backgrounds; S13. The images of energy absorbers obtained from the quasi-static tensile test, full-scale impact test and field engineering measurements are annotated with salient target detection labels, and data is augmented and divided into training and validation sets.

3. The non-contact intelligent detection method for the working state of the energy absorber of the slope flexible protection system according to claim 1 is characterized by: In step S2, the specific steps include: S21. Determine the structure of the deep neural network for salient object detection and write code; S22, using the training set to train the deep neural network for salient object detection and using the validation set for testing; S23, monitoring the training error, verifying the error and adjusting the network structure until the error meets the accuracy requirement, obtaining a trained neural network model, and using the trained neural network model to realize automatic segmentation of the binary graph of the working state of the energy consumer.

4. The non-contact intelligent detection method for the working state of the energy absorber of the slope flexible protection system according to claim 2 is characterized by: In step S13, the notable target detection label annotation is performed by selecting Labelme, EISeg or LabelImg annotation software for annotation, and a unified standard is used during annotation to ensure the accuracy and consistency of the labels; when forming an image data set, the original image and label are rotated, scaled, and blurred to achieve image data amplification.

5. The non-contact intelligent detection method for the working state of the energy absorber of the slope flexible protection system according to claim 3 is characterized by: In step S21, the salient object detection deep neural network is U 2 -Net, F 3 -Net, BASNet or any other.

6. The non-contact intelligent detection method for the working state of the energy absorber of the slope flexible protection system according to claim 3 is characterized by: In step S22, when training the deep neural network for salient object detection, the loss function L is calculated as follows: Where: M represents the number of feature maps, m represents the mth feature map, represents the weight coefficient of the mth sub-feature map, represents the loss value of the mth sub-feature map, w fuse Represents the total feature map weight coefficient after feature fusion, l fuse Represents the total feature map loss value after feature fusion; When training a deep neural network for salient object detection, we use F β , MAE is a key evaluation index to determine the model performance, and is calculated as follows: Where: β is the weight factor, Precision is the accuracy, Recall is the recall, H and W are the length and width of the image respectively, r and c are the pixel coordinates in the length and width directions of the image respectively, P(r,c) and G(r,c) are the model prediction value and the true value of the label respectively.

7. The non-contact intelligent detection method for the working state of the energy absorber of the slope flexible protection system according to claim 1 is characterized by: The remaining energy consumption capacity R of the energy consumer is calculated by the inner and outer contour lengths output by the skeleton image contour detection, as follows: L0=η(L R +L C ); Where: L R Indicates the length of the energy consumer before it is started after it actually works, L C It indicates the elongation of the energy absorber after actual operation. L0 indicates the length of the energy absorber that cannot be started under the limit state. L Outer Represents the outer contour length of the energy consumer skeleton diagram, L Inner represents the inner contour length of the energy consumer skeleton graph, and η is the incomplete start-up coefficient of the energy consumer.

Images