CN116363512A - Method and device for detecting stability of surrounding rock - Google Patents
Method and device for detecting stability of surrounding rock Download PDFInfo
- Publication number
- CN116363512A CN116363512A CN202310300308.7A CN202310300308A CN116363512A CN 116363512 A CN116363512 A CN 116363512A CN 202310300308 A CN202310300308 A CN 202310300308A CN 116363512 A CN116363512 A CN 116363512A
- Authority
- CN
- China
- Prior art keywords
- abnormal target
- detection network
- image
- visible light
- target detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/176—Urban or other man-made structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Image Processing (AREA)
Abstract
The invention provides a method and a device for detecting stability of surrounding rock, wherein the method comprises the following steps: obtaining a visible light image and an infrared thermal imaging image of the surrounding rock of the sample tunnel, and respectively marking the visible light image and the infrared thermal imaging image; pre-training an abnormal target detection network based on the marked visible light image and the marked infrared thermal imaging image; based on the predicted position and the corresponding type of the abnormal target output by the pre-trained abnormal target detection network, the position and the corresponding type of the marked abnormal target, and the pre-trained abnormal target detection network is updated with the parameters of the abnormal target detection network based on reinforcement learning by using a reward function; and (3) acquiring a visible light image and an infrared thermal imaging image of the surrounding rock of the tunnel to be detected, inputting an abnormal target detection network for updating parameters of the abnormal target detection network, acquiring an abnormal target prediction result, and determining whether the tunnel is stable or not based on the abnormal target prediction result. The accuracy of surrounding rock stability detection can be improved.
Description
Technical Field
The invention relates to the technical field of geological monitoring, in particular to a method and a device for detecting stability of surrounding rock.
Background
The tunnel has important function in transportation, and surrounding rock stability is a main factor influencing the safety of tunnel engineering, and the personal safety and property safety can be seriously threatened by surrounding rock loosening, collapse and the like. Taking the tunnel face instability with the characteristics of extremely short occurrence process, oversized occurrence scale, great safety influence and the like as an example, when preliminary treatment operations such as primary support and the like are improper in the excavation process or the soil body of the excavation face faces sudden loads, large-area tunnel face rock mass instability tends to occur at one moment, and then intermittent stranding and scale sand soil sliding off can be accompanied, and secondary and tertiary soil body collapse can be accompanied. Therefore, the stability of the excavated soil body and surrounding rock at the excavation face in the track traffic underground excavation construction process is effectively monitored, abnormal changes of the stability of the tunnel surrounding rock are timely found, early warning and tracking records are timely issued, and the method has great significance for timely checking dangerous cases.
At present, a target detection algorithm based on deep learning is generally adopted to detect the stability of surrounding rock. For example, by generating a suggestion frame (region prediction), extracting image features from the suggestion frame, discriminating the target category to which the image features corresponding to the suggestion frame belong based on a region classifier, and performing classification prediction on the target at each position of the image features, thereby obtaining a surrounding rock stability detection result of the target. However, in the method, when the situation of complex scenes such as instability of surrounding rocks of the tunnel and secondary damage is faced, the conditions of missed detection, re-detection and false detection are very easy to occur, and the accuracy and timeliness of the stability detection of the surrounding rocks of the tunnel are greatly reduced.
Disclosure of Invention
In view of the above, the present invention aims to provide a method and a device for detecting the stability of surrounding rock, so as to improve the accuracy of detecting the stability of the surrounding rock.
In a first aspect, an embodiment of the present invention provides a method for detecting stability of surrounding rock, including:
obtaining a visible light image and an infrared thermal imaging image of the surrounding rock of the sample tunnel, and respectively marking the visible light image and the infrared thermal imaging image;
pre-training an abnormal target detection network based on the marked visible light image and the marked infrared thermal imaging image;
based on the predicted position and the corresponding type of the abnormal target output by the pre-trained abnormal target detection network, the position and the corresponding type of the marked abnormal target, and the pre-trained abnormal target detection network is updated with the parameters of the abnormal target detection network based on reinforcement learning by using a reward function;
and acquiring a visible light image and an infrared thermal imaging image of the surrounding rock of the tunnel to be detected, inputting an abnormal target detection network for updating parameters of the abnormal target detection network, acquiring an abnormal target prediction result of the surrounding rock of the tunnel to be detected, and determining the stability of the surrounding rock of the tunnel to be detected based on the abnormal target prediction result.
With reference to the first aspect, an embodiment of the present invention provides a first possible implementation manner of the first aspect, where after the obtaining a visible light image and an infrared thermal imaging image of the surrounding rock of the sample tunnel, before labeling the visible light image and the infrared thermal imaging image, respectively, the method further includes:
and preprocessing the obtained visible light image and infrared thermal imaging image.
With reference to the first possible implementation manner of the first aspect, the embodiment of the present invention provides a second possible implementation manner of the first aspect, wherein the preprocessing includes: image rotation, image blurring, image sharpening, image sharpness, image size transformation, or any combination thereof.
With reference to the first aspect, an embodiment of the present invention provides a third possible implementation manner of the first aspect, where the pre-training the abnormal target detection network based on the visible light image and the infrared thermal imaging image includes:
respectively inputting the visible light image and the infrared thermal imaging image into a channel exchange feature extraction module of an abnormal target detection network to obtain corresponding image features;
carrying out feature fusion on the image features corresponding to the obtained visible light image and the image features corresponding to the infrared thermal imaging image by utilizing a visible-infrared feature fusion module to obtain fusion features;
And inputting the fusion characteristics into an abnormal target regression module, and training an abnormal target detection network based on the abnormal target prediction result and the marked visible light image and infrared thermal imaging image output by the abnormal target regression module.
With reference to the third possible implementation manner of the first aspect, the embodiment of the present invention provides a fourth possible implementation manner of the first aspect, wherein training the abnormal target detection network based on the abnormal target prediction result and the marked visible light image and the infrared thermal imaging image output by the abnormal target regression module includes:
obtaining an abnormal target prediction position, a position prediction confidence coefficient and a class probability distribution corresponding to an abnormal target output by an abnormal target regression module, and calculating a difference value among the position of the abnormal target, the detection confidence coefficient of the abnormal target and the class probability distribution corresponding to the abnormal target, which are marked corresponding to an input image, by adopting a loss function of yolov 4;
based on the difference value, back propagation is performed to update the abnormal target detection network parameters;
judging whether the maximum iteration times are reached, if so, acquiring the abnormal target detection network parameters with the minimum loss function values in the iteration process, and obtaining an abnormal target detection network with the pre-trained completion; and if not, executing the step of acquiring the predicted position of the abnormal target, the confidence coefficient of the position prediction and the class probability distribution corresponding to the abnormal target, which are output by the abnormal target regression module.
With reference to the first aspect and any one of the first possible implementation manner to the fourth possible implementation manner of the first aspect, the embodiment of the present invention provides a fifth possible implementation manner of the first aspect, wherein the performing, with a reward function, updating parameters of the pre-trained anomaly target detection network based on the reinforcement learning anomaly target detection network based on the pre-trained anomaly target prediction position, the category corresponding to the anomaly target, and the labeled anomaly target position, the category corresponding to the anomaly target includes:
obtaining an abnormal target prediction position and a class corresponding to the abnormal target output by an abnormal target detection network based on pre-training, and a marked abnormal target position and a marked abnormal target class reward value by using a reward function;
calculating abnormal target detection network parameters of the abnormal target detection network based on pre-training based on the reward value by adopting a strategy gradient optimization algorithm;
calculating the parameter gradient of the abnormal target detection network parameter, and updating the abnormal target detection network parameter according to the parameter gradient back propagation;
judging whether the maximum iteration times are reached, if so, acquiring the abnormal target detection network parameter with the highest rewarding function value in the iteration process, and obtaining the abnormal target detection network parameter as a final abnormal target detection network; if not, executing the step of utilizing the reward function to acquire the predicted position of the abnormal target, the class corresponding to the abnormal target, the position of the marked abnormal target and the reward value of the class corresponding to the abnormal target, which are output by the pre-trained abnormal target detection network.
With reference to the first aspect and any one of the first possible implementation manner to the fourth possible implementation manner of the first aspect, an embodiment of the present invention provides a sixth possible implementation manner of the first aspect, where labeling the visible light image includes:
determining the position of an abnormal target of surrounding rock instability deformation of a working face in the visible light image from the acquired visible light image;
and acquiring detection confidence of the abnormal target and class probability distribution corresponding to the abnormal target according to the position of the abnormal target determined in each visible light image.
In a second aspect, an embodiment of the present invention further provides a device for detecting stability of surrounding rock, including:
the marking module is used for acquiring a visible light image and an infrared thermal imaging image of the surrounding rock of the sample tunnel and marking the visible light image and the infrared thermal imaging image respectively;
the pre-training module is used for pre-training the abnormal target detection network based on the marked visible light image and the infrared thermal imaging image;
the parameter updating module is used for updating parameters of the abnormal target detection network based on reinforcement learning by utilizing a reward function on the pre-trained abnormal target detection network based on the predicted position of the abnormal target, the type corresponding to the abnormal target, the marked position of the abnormal target and the type corresponding to the abnormal target;
The prediction module is used for acquiring a visible light image and an infrared thermal imaging image of the surrounding rock of the tunnel to be detected, inputting an abnormal target detection network for updating parameters of the abnormal target detection network, acquiring an abnormal target prediction result of the surrounding rock of the tunnel to be detected, and determining the stability of the surrounding rock of the tunnel to be detected based on the abnormal target prediction result.
In a third aspect, embodiments of the present application provide a computer device including a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the above method when executing the computer program.
In a fourth aspect, embodiments of the present application provide a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of the method described above.
According to the method and the device for detecting the stability of the surrounding rock, provided by the embodiment of the invention, the visible light image and the infrared thermal imaging image of the surrounding rock of the sample tunnel are respectively marked by acquiring the visible light image and the infrared thermal imaging image; pre-training an abnormal target detection network based on the marked visible light image and the marked infrared thermal imaging image; based on the predicted position and the corresponding type of the abnormal target output by the pre-trained abnormal target detection network, the position and the corresponding type of the marked abnormal target, and the pre-trained abnormal target detection network is updated with the parameters of the abnormal target detection network based on reinforcement learning by using a reward function; and acquiring a visible light image and an infrared thermal imaging image of the surrounding rock of the tunnel to be detected, inputting an abnormal target detection network for updating parameters of the abnormal target detection network, acquiring an abnormal target prediction result of the surrounding rock of the tunnel to be detected, and determining the stability of the surrounding rock of the tunnel to be detected based on the abnormal target prediction result. Therefore, through the visible light image and the infrared thermal imaging image, reinforcement learning is performed by taking the rewarding function as the objective function, and small sample detection of the tunnel surrounding rock stability abnormal target can be effectively relieved, so that the accuracy of surrounding rock stability detection is improved.
In order to make the above objects, features and advantages of the present invention more comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the embodiments will be briefly described below, it being understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and other related drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 shows a schematic flow chart of a method for detecting stability of surrounding rock according to an embodiment of the present invention;
fig. 2 shows a schematic structural diagram of an apparatus for detecting stability of surrounding rock according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a computer device 300 according to an embodiment of the present application.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be made by a person skilled in the art without making any inventive effort, are intended to be within the scope of the present invention.
At present, a public data set is rarely arranged for a tunnel surrounding rock instability scene, and the tunnel surrounding rock instability scene has extremely low occurrence probability, extremely high occurrence speed and extremely high damage, so that tunnel surrounding rock instability belongs to the problem of small sample target detection, and when the small sample target detection of the tunnel surrounding rock instability is faced by the existing target detection algorithm based on deep learning, missing detection, re-detection and false detection are easy to occur, and the accuracy of tunnel stability detection is greatly reduced. In the embodiment of the invention, the tunnel construction images in the infrared thermal imaging wave band can provide complementary temperature sensing information in addition to the tunnel construction images in the visible light wave band, so that the multi-mode image characteristics of the visible light image and the infrared thermal imaging image are synthesized, and the accuracy of tunnel stability detection is improved; further, reinforcement learning is performed by designing a reward function as an objective function, so that the difficulty in detecting a small sample target of instability of tunnel surrounding rock is relieved, and accurate judgment of the position and the category of an abnormal target of stability of the tunnel surrounding rock is realized.
The embodiment of the invention provides a method and a device for detecting stability of surrounding rock, and the method and the device are described below through the embodiment.
Fig. 1 shows a flow chart of a method for detecting stability of surrounding rock according to an embodiment of the present invention. As shown in fig. 1, the method includes:
in the embodiment of the invention, a tunnel surrounding rock stability abnormal target detection data set is constructed according to the visible light image and the infrared thermal imaging image which are marked.
In the embodiment of the invention, the visible light image and the infrared thermal imaging image of the tunnel surrounding rock are acquired according to cameras which are arranged in unmanned construction, normal construction and rare tunnel surrounding rock instability scenes of a plurality of tunnel construction sites meeting the standard.
In an embodiment of the present invention, as an optional embodiment, labeling a visible light image includes:
determining the position of an abnormal target of surrounding rock instability deformation of a working face in the visible light image from the acquired visible light image;
and acquiring detection confidence of the abnormal target and class probability distribution corresponding to the abnormal target according to the position of the abnormal target determined in each visible light image.
In the embodiment of the invention, each visible light image is analyzed in a manual mode, the position of an abnormal target of surrounding rock instability deformation of the working face and soil and stone falling possibly caused by deformation are obtained, and the position of the abnormal target, the detection confidence of the abnormal target and the class probability distribution corresponding to the abnormal target are marked in the visible light image. As an alternative embodiment, the labeling target area for position labeling is the minimum circumscribed horizontal rectangle of the actual area to which the abnormal target belongs. The categories include the locations where the face is subjected to instability deformation of surrounding rock and the falling of earth and stones possibly caused by deformation.
In the embodiment of the invention, the tunnel surrounding rock stability abnormal target detection data set is composed of a visible light image, an infrared thermal imaging image and corresponding abnormal target marks.
In an embodiment of the present invention, as an optional embodiment, after obtaining the visible light image and the infrared thermal imaging image of the surrounding rock of the sample tunnel, before labeling the visible light image and the infrared thermal imaging image respectively, the method further includes:
and preprocessing the obtained visible light image and infrared thermal imaging image.
In an embodiment of the present invention, as an alternative embodiment, the preprocessing includes, but is not limited to: one or more of image rotation, image blurring, image sharpening, image sharpness, image size transformation.
In an embodiment of the present invention, as an alternative embodiment, preprocessing is accomplished by invoking python image processing library (PIL, python Image Library). Wherein, the liquid crystal display device comprises a liquid crystal display device,
the image is rotated at random angles, as an alternative embodiment, the rotation angle range is [ -30,30]; the sharpening degree of the image is adjusted, the value range of the sharpening degree is [0.1,1.9], the sharpening degree is 0 to represent a blurred image, the sharpening degree is 1 to represent an original image, and the sharpening degree is 2 to represent a sharpened image; the image sharpness is to carry out histogram operation on the image and is used for enhancing the contrast of the image; the image size conversion is used to size-convert the images so that the images have the same size. As an alternative embodiment, the transformed size is 224×224.
in an embodiment of the present invention, as an optional embodiment, the abnormal object detection network includes: the system comprises a channel exchange feature extraction module, a visible-infrared feature fusion module and an abnormal target regression module.
In the embodiment of the invention, the abnormal target detection network predicts the abnormal target in the input image by regression through the input visible light image and the infrared thermal imaging image.
In the embodiment of the invention, the image input to the abnormal target detection network is set as follows:
wherein x is rgb X is visible light image t For infrared thermographic image, x m The method comprises the steps of obtaining images corresponding to wave bands, wherein the wave bands comprise visible light wave bands and infrared wave bands;
W in 、H in 、C in the width, height, and number of channels of the input image, respectively.
In the embodiment of the present invention, as an optional embodiment, the pre-training abnormal target detection network based on the marked visible light image and the infrared thermal imaging image includes:
s11, respectively inputting the visible light image and the infrared thermal imaging image into a channel exchange feature extraction module of an abnormal target detection network to obtain corresponding image features;
in the embodiment of the invention, the channel exchange feature extraction module is adopted to extract the image features, and as an optional embodiment, the following formula is used to extract the features:
Wherein, the liquid crystal display device comprises a liquid crystal display device,
for the corresponding features (sequence, feature images) of the band images, -for the band images>
A feature extraction module for channel switching;
W middle 、H middle 、C middle the width, height, number of channels of the extracted features, respectively.
In the embodiment of the invention, the channel exchange feature extraction module is utilized to extract the features of the input visible light image and the input infrared thermal imaging image. As an alternative embodiment, the channel switching feature extraction module is a neural network, and is composed of basic elements such as a convolution layer, a channel switching layer, a ReLU activation function and the like, and the input image sequentially passes through L 1 And the convolution layer, the channel exchange layer and the ReLU activation function are processed by the pooling operation layer, and then imaging feature extraction results of corresponding wave bands are output.
In the embodiment of the present invention, for the convolutional layer of layer 1, the channel switching feature extraction module performs an operation using the following formula:
wherein, the liquid crystal display device comprises a liquid crystal display device,
the input visual characteristics and the output visual characteristics of the first convolution layer are respectively m wave bands (if m=rgb, the visible light wave band is represented, and if m=t, the infrared thermal imaging wave band is represented), and the channel exchange layer does not change the characteristic dimension; />
And the parameters of the channel exchange characteristic extraction modules corresponding to the m wave bands of the first convolution layer are shared except the channel exchange layers.
In the embodiment of the invention, the channel exchange layer performs input processing by using the following formula:
wherein, the liquid crystal display device comprises a liquid crystal display device,
the input visual characteristics and the output visual characteristics of a first layer and a c channel of m wave bands are respectively;
In the embodiment of the invention, if m is a visible light wave band, m' is an infrared wave band; if m is the infrared band, then m' is the visible band.
In the embodiment of the invention, the learnable parameter reduction factor in the channel exchange layer can be used for inputting x in the evaluation training process m，l，c And output of
Degree of correlation between the two, if the parameter reduction factor gamma can be learned m，l，c Near 0, the loss function is about x m，l，c Will be near 0, characterizing x m，l，c Will lose the effect on the final prediction result, i.e. x m，l，c Is considered as a redundant component in the corresponding band profile. That is, if a feature map in the visible light band has little influence on final abnormal target detection, in the embodiment of the present invention, the feature map is replaced with a corresponding feature map in the infrared thermal imaging band, and the same applies to the infrared thermal imaging band. After channel exchange, gradient update is separated from the original channel and is replaced to be connected with the exchanged channel.
In the embodiment of the invention, the image features extracted by the channel exchange feature extraction module are depth features of the input image, the depth features are different from color features, texture features and the like in image processing, and the physical meaning of each dimension is not clear, but the features can better assist in high-precision realization of subsequent abnormal target detection tasks.
S12, carrying out feature fusion on the image features corresponding to the obtained visible light image and the image features corresponding to the infrared thermal imaging image by utilizing a visible-infrared feature fusion module to obtain fusion features;
in the embodiment of the present invention, as an alternative embodiment, feature fusion is performed using the following formula:
in the formula g fusion () In order to be a visible-infrared feature fusion module,
for the image features corresponding to the visible light image,
is an image feature corresponding to the infrared thermal imaging image.
In the embodiment of the present invention, as an alternative embodiment, the fusion feature is obtained using the following formula:
in the embodiment of the invention, the visible-infrared characteristic fusion module g fusion (. Cndot.) simple mean calculation is used to act on the extracted depth features of the visible light band
Infrared thermal imaging band depth profile>
Output IR-visible fusion feature +. >
In the embodiment of the invention, as another optional embodiment, fusion characteristic processing can be performed by using a splicing operator and a neural network to obtain fusion characteristics.
S13, inputting the fusion characteristics into an abnormal target regression module, and training an abnormal target detection network based on the abnormal target prediction result output by the abnormal target regression module and the marked visible light image and infrared thermal imaging image.
In the embodiment of the invention, the output of the abnormal target regression module comprises: the position of the abnormal target, the detection confidence of the abnormal target and the class probability distribution corresponding to the abnormal target.
In the embodiment of the invention, the abnormal target regression module predicts the abnormal target by using the following formula:
wherein g detection (. Cndot.) is an outlier regression module,
for each position of the abnormal target, the position includes four degrees of freedom (D 1 =4), i.e. the central point space coordinates of the horizontal rectangle, the length, width of the horizontal rectangle, +.>
Confidence of detection for abnormal target (D 2 ＝1)，/>
For the class probability distribution (D) corresponding to the outlier target 3 The number of categories representing abnormal objects), N is the number of abnormal objects detected from the input image.
In the embodiment of the present invention, as an optional embodiment, obtaining confidence and class probability distribution of an abnormal target may be implemented by using yolo architecture.
In the embodiment of the invention, the abnormal target regression module regressively outputs the predicted position box of the abnormal target based on the fusion characteristic of infrared light and visible light position Confidence of position prediction p (box) position ) Class probability distribution p (box) corresponding to abnormal target class ). As a oneIn an alternative embodiment, the anomaly target regression module employs a network architecture of yolov 4. As another alternative, an Oriented-RCNN, KLD, R3Det network architecture may also be employed.
In an embodiment of the present invention, as an alternative embodiment, the abnormal objects include, but are not limited to: the surrounding rock is unstable and deformed, and the surrounding rock possibly falls off from soil and rocks caused by deformation.
In the embodiment of the invention, the fusion characteristic is input into the abnormal target regression module to obtain the output abnormal target prediction result, and then the output abnormal target prediction result is compared with the abnormal target marked by the visible light image or the infrared image corresponding to the actual fusion characteristic, namely, the loss function L of yolov4 is adopted yolo (θ) evaluating sample abnormal target prediction values (including regression output abnormal target prediction position box position Confidence of position prediction p (box) position ) Class probability distribution p (box) corresponding to abnormal target class ) The difference value between the abnormal target regression module and the actual labeling value to adjust the network parameters of the abnormal target regression module, so that the output abnormal target prediction result is similar to the actual result, thereby completing the pre-training of the abnormal target regression module.
In the embodiment of the invention, as an optional embodiment, the channel exchange feature extraction module can keep most of unique global features of two imaging bands (a visible light band and an infrared band) in a self-adaptive mode, and can fuse key unique information of each of the two imaging bands to obtain depth features, so that effective detection of a tunnel surrounding rock stability abnormal target can be assisted, and the accuracy of abnormal target detection and alarm can be improved in a robust manner.
In the embodiment of the present invention, as an optional embodiment, training the abnormal target detection network based on the abnormal target prediction result and the marked visible light image and the infrared thermal imaging image output by the abnormal target regression module includes:
s21, obtaining an abnormal target prediction position, a position prediction confidence coefficient and a class probability distribution corresponding to an abnormal target output by an abnormal target regression module, and calculating a difference value among the position of the abnormal target, the detection confidence coefficient of the abnormal target and the class probability distribution corresponding to the abnormal target, which are marked corresponding to an input image, by adopting a loss function of yolov 4;
in the embodiment of the invention, a loss function L of yolov4 is adopted yolo (θ) evaluating the abnormal target prediction value output by the abnormal target regression module (including regression output abnormal target prediction position box) position Confidence of position prediction p (box) position ) Class probability distribution p (boc) corresponding to abnormal target class ) A difference value between the actual labeling value).
In the embodiment of the present invention, as another alternative embodiment, the loss function may also be used: yolov1, yolov2, yolov3, focal Loss, smoothed first norm, smoothed second norm, IOU penalty function, mutual information. Initial values of network parameters for the anomaly object detection network include, but are not limited to, random initialization and fixed value initialization.
S22, back propagation is carried out based on the difference value so as to update abnormal target detection network parameters;
s23, judging whether the maximum iteration times are reached, if so, acquiring the abnormal target detection network parameters with the minimum loss function values in the iteration process, and obtaining an abnormal target detection network with the pre-trained function values; and if not, executing the step of acquiring the predicted position of the abnormal target, the confidence coefficient of the position prediction and the class probability distribution corresponding to the abnormal target, which are output by the abnormal target regression module.
In the embodiment of the invention, based on a pre-trained abnormal target detection network, abnormal detection output corresponding to a sample in a data set is obtained, and the pre-trained abnormal target network is subjected to reinforcement learning-based parameter updating according to a reward function.
In the embodiment of the invention, the reward function is used for encouraging the finally obtained abnormal target detection network to relieve the difficulty of detecting the small sample of the abnormal target, and the accurate judgment of the position and the category of the abnormal target is realized.
In an embodiment of the present invention, as an alternative embodiment, the bonus process is performed using the following formula:
wherein, from analog to p (box) position ) Indicating whether the abnormal target detection frame is correct or not (1 represents correct, 0 represents error); class-p (box) class ) And the category to which the abnormal target belongs is represented.
In the embodiment of the invention, if the pre-trained abnormal target detection network erroneously detects the original normal area as the tunnel surrounding rock stability abnormal area, the reward function outputs a negative number-r 1 (r 1 >0) The system is used for punishing abnormal target detection network parameters; if the abnormal target detection network predicts errors for the class of the abnormal target, the reward function will output a negative number-r 2 (r 2 > 0) for punishing abnormal target detection network parameters; if the prediction of the abnormal target detection frame and the class is correct, the bonus function will output a positive number +r 3 (r 3 > 0) for encouraging abnormal target detection of network parameters.
In the embodiment of the present invention, as an optional embodiment, based on the predicted position of the abnormal target, the type corresponding to the abnormal target, the position of the marked abnormal target, and the type corresponding to the abnormal target output by the pre-trained abnormal target detection network, updating parameters of the pre-trained abnormal target detection network based on reinforcement learning by using a reward function, the method includes:
s31, acquiring an abnormal target prediction position and a class corresponding to the abnormal target output by an abnormal target detection network based on pre-training, and a marked abnormal target position and a marked abnormal target class rewarding value by using a rewarding function;
s32, calculating abnormal target detection network parameters of the abnormal target detection network based on pre-training based on the rewarding value by adopting a strategy gradient optimization algorithm;
in the embodiment of the invention, the abnormal target detection network parameter is an expected value calculated on the rewarding function value by adopting a strategy gradient optimization algorithm:
in the embodiment of the invention, as another optional embodiment, the strategy gradient optimization algorithm can also adopt reinforcement learning optimization algorithms such as PPO, A2C, A C and the like.
S33, calculating a parameter gradient of the abnormal target detection network parameter, and updating the abnormal target detection network parameter according to the parameter gradient back propagation;
in an embodiment of the present invention, as an alternative embodiment, the parameter gradient is calculated using the following formula:
wherein, xi > 0 is the preset parameter updating step length,
representing the gradient of the abnormal target detection network parameter theta.
S34, judging whether the maximum iteration times are reached, if so, acquiring the abnormal target detection network parameter with the highest rewarding function value in the iteration process, and obtaining the abnormal target detection network parameter as a final abnormal target detection network; if not, executing the step of utilizing the reward function to acquire the predicted position of the abnormal target, the class corresponding to the abnormal target, the position of the marked abnormal target and the reward value of the class corresponding to the abnormal target, which are output by the pre-trained abnormal target detection network.
In the embodiment of the invention, the abnormal target detection network for updating the parameters of the abnormal target detection network is a final abnormal target detection network, and the tunnel surrounding rock stability abnormal event is detected and alarm warning is carried out through the obtained abnormal target detection network.
In the embodiment of the present invention, as an optional embodiment, a criterion for determining stability of the tunnel surrounding rock to be detected based on the abnormal target prediction result is: determining the number of high-confidence abnormal targets according to the abnormal target prediction result, and if the determined number is non-zero, alarming to warn that small-scale instability deformation has occurred in the current tunnel surrounding rock construction range, so that the risk of subsequent large-scale secondary accidents is caused, and emergency measures are needed to avoid the surrounding rock stability expansion risk; if the determined number is zero, the alarm is not given, and the stability of the surrounding rock is continuously detected.
In the embodiment of the invention, the parameter value of the abnormal target detection network obtained according to back propagation is reprocessed with the reward function and then is used as a new parameter value of the abnormal target detection network, and the retrained abnormal target detection network is retrained.
According to the embodiment of the invention, the detection of the tunnel surrounding rock stability abnormal target is realized by utilizing an intelligent algorithm, so that related personnel are not required to be on site, or a monitoring camera is relied on to carry out remote evaluation, the tunnel surrounding rock stability abnormal event can be found and reported at the first time under the unattended condition, and precious time is provided for rescue at the first time. Further, through synthesizing the multi-mode tunnel construction image characteristics of visible light wave band and infrared thermal imaging wave band, can combine visible light and temperature information, rationally judge the unusual target of tunnel surrounding rock stability to improve the accuracy of target detection. Meanwhile, reinforcement learning is performed by taking the reward function as the objective function, so that the difficulty in detecting a small sample of an abnormal target of the stability of the surrounding rock of the tunnel can be effectively relieved, and the accurate judgment of the position and the category of the abnormal target can be realized.
Fig. 2 shows a schematic structural diagram of an apparatus for detecting stability of surrounding rock according to an embodiment of the present invention. As shown in the figure 2 of the drawings,
the labeling module 201 is configured to obtain a visible light image and an infrared thermal imaging image of the surrounding rock of the sample tunnel, and label the visible light image and the infrared thermal imaging image respectively;
in this embodiment of the present invention, as an optional embodiment, the labeling module 201 is specifically configured to:
determining the position of an abnormal target of surrounding rock instability deformation of a working face in the visible light image from the acquired visible light image;
and acquiring detection confidence of the abnormal target and class probability distribution corresponding to the abnormal target according to the position of the abnormal target determined in each visible light image.
In this embodiment of the present invention, as another optional embodiment, the labeling module 201 is specifically further configured to:
and preprocessing the obtained visible light image and infrared thermal imaging image.
In an embodiment of the present invention, the preprocessing includes, but is not limited to: image rotation, image blurring, image sharpening, image sharpness, image size transformation, or any combination thereof.
The pre-training module 202 is configured to pre-train the abnormal target detection network based on the marked visible light image and the infrared thermal imaging image;
In an embodiment of the present invention, as an optional embodiment, the pre-training module 202 is specifically configured to:
respectively inputting the visible light image and the infrared thermal imaging image into a channel exchange feature extraction module of an abnormal target detection network to obtain corresponding image features;
carrying out feature fusion on the image features corresponding to the obtained visible light image and the image features corresponding to the infrared thermal imaging image by utilizing a visible-infrared feature fusion module to obtain fusion features;
and inputting the fusion characteristics into an abnormal target regression module, and training an abnormal target detection network based on the abnormal target prediction result and the marked visible light image and infrared thermal imaging image output by the abnormal target regression module.
In the embodiment of the present invention, as an optional embodiment, training the abnormal target detection network based on the abnormal target prediction result and the marked visible light image and the infrared thermal imaging image output by the abnormal target regression module includes:
obtaining an abnormal target prediction position, a position prediction confidence coefficient and a class probability distribution corresponding to an abnormal target output by an abnormal target regression module, and calculating a difference value among the position of the abnormal target, the detection confidence coefficient of the abnormal target and the class probability distribution corresponding to the abnormal target, which are marked corresponding to an input image, by adopting a loss function of yolov 4;
Based on the difference value, back propagation is performed to update the abnormal target detection network parameters;
judging whether the maximum iteration times are reached, if so, acquiring the abnormal target detection network parameters with the minimum loss function values in the iteration process, and obtaining an abnormal target detection network with the pre-trained completion; and if not, executing the step of acquiring the predicted position of the abnormal target, the confidence coefficient of the position prediction and the class probability distribution corresponding to the abnormal target, which are output by the abnormal target regression module.
The parameter updating module 203 is configured to update parameters of the abnormal target detection network based on reinforcement learning by using a reward function, based on the predicted position of the abnormal target and the class corresponding to the abnormal target output by the pre-trained abnormal target detection network, and the position of the marked abnormal target and the class corresponding to the abnormal target;
in this embodiment of the present invention, as an optional embodiment, the parameter updating module 203 is specifically configured to:
obtaining an abnormal target prediction position and a class corresponding to the abnormal target output by an abnormal target detection network based on pre-training, and a marked abnormal target position and a marked abnormal target class reward value by using a reward function;
Calculating abnormal target detection network parameters of the abnormal target detection network based on pre-training based on the reward value by adopting a strategy gradient optimization algorithm;
calculating the parameter gradient of the abnormal target detection network parameter, and updating the abnormal target detection network parameter according to the parameter gradient back propagation;
judging whether the maximum iteration times are reached, if so, acquiring the abnormal target detection network parameter with the highest rewarding function value in the iteration process, and obtaining the abnormal target detection network parameter as a final abnormal target detection network; if not, executing the step of utilizing the reward function to acquire the predicted position of the abnormal target, the class corresponding to the abnormal target, the position of the marked abnormal target and the reward value of the class corresponding to the abnormal target, which are output by the pre-trained abnormal target detection network.
The prediction module 204 is configured to obtain a visible light image and an infrared thermal imaging image of a surrounding rock of a tunnel to be detected, input an abnormal target detection network for updating parameters of the abnormal target detection network, obtain an abnormal target prediction result of the surrounding rock of the tunnel to be detected, and determine stability of the surrounding rock of the tunnel to be detected based on the abnormal target prediction result.
In the embodiment of the present invention, as an optional embodiment, the prediction module 204 determines the number of abnormal targets with high confidence according to the prediction result of the abnormal targets, if the determined number is non-zero, the alarm alerts that the current tunnel surrounding rock construction range has small-scale instability deformation, and there is a risk of subsequently causing large-scale secondary accidents, and emergency measures need to be taken to avoid the risk of expanding the surrounding rock stability; if the determined number is zero, the alarm is not given, and the stability of the surrounding rock is continuously detected.
As shown in fig. 3, an embodiment of the present application provides a computer device 300 for performing the method for detecting the stability of surrounding rock in fig. 1, where the device includes a memory 301, a processor 302 connected to the memory 301 through a bus, and a computer program stored on the memory 301 and capable of running on the processor 302, where the steps of the method for detecting the stability of surrounding rock are implemented when the processor 302 executes the computer program.
Specifically, the above memory 301 and the processor 302 can be general-purpose memories and processors, and are not particularly limited herein, and the above method for detecting the stability of the surrounding rock can be performed when the processor 302 runs a computer program stored in the memory 301.
Corresponding to the method for detecting the stability of the surrounding rock in fig. 1, the embodiment of the present application further provides a computer readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, performs the steps of the above method for detecting the stability of the surrounding rock.
In particular, the storage medium can be a general-purpose storage medium, such as a removable disk, a hard disk, etc., on which a computer program is executed, capable of performing the above-described method of detecting the stability of the surrounding rock.
In the embodiments provided herein, it should be understood that the disclosed systems and methods may be implemented in other ways. The system embodiments described above are merely illustrative, e.g., the division of the elements is merely a logical functional division, and there may be additional divisions in actual implementation, and e.g., multiple elements or components may be combined or integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be through some communication interface, system or unit indirect coupling or communication connection, which may be in electrical, mechanical or other form.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional unit in the embodiments provided in the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
It should be noted that: like reference numerals and letters in the following figures denote like items, and thus once an item is defined in one figure, no further definition or explanation of it is required in the following figures, and furthermore, the terms "first," "second," "third," etc. are used merely to distinguish one description from another and are not to be construed as indicating or implying relative importance.
Finally, it should be noted that: the foregoing examples are merely specific embodiments of the present application, and are not intended to limit the scope of the present application, but the present application is not limited thereto, and those skilled in the art will appreciate that while the foregoing examples are described in detail, the present application is not limited thereto. Any person skilled in the art may modify or easily conceive of the technical solution described in the foregoing embodiments, or make equivalent substitutions for some of the technical features within the technical scope of the disclosure of the present application; such modifications, changes or substitutions do not depart from the spirit and scope of the corresponding technical solutions. Are intended to be encompassed within the scope of this application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (10)
1. A method of detecting stability of a surrounding rock, comprising:
obtaining a visible light image and an infrared thermal imaging image of the surrounding rock of the sample tunnel, and respectively marking the visible light image and the infrared thermal imaging image;
pre-training an abnormal target detection network based on the marked visible light image and the marked infrared thermal imaging image;
Based on the predicted position and the corresponding type of the abnormal target output by the pre-trained abnormal target detection network, the position and the corresponding type of the marked abnormal target, and the pre-trained abnormal target detection network is updated with the parameters of the abnormal target detection network based on reinforcement learning by using a reward function;
and acquiring a visible light image and an infrared thermal imaging image of the surrounding rock of the tunnel to be detected, inputting an abnormal target detection network for updating parameters of the abnormal target detection network, acquiring an abnormal target prediction result of the surrounding rock of the tunnel to be detected, and determining the stability of the surrounding rock of the tunnel to be detected based on the abnormal target prediction result.
2. The method of claim 1, wherein after the acquiring the visible light image and the infrared thermographic image of the sample tunnel surrounding rock, before labeling the visible light image and the infrared thermographic image, respectively, the method further comprises:
and preprocessing the obtained visible light image and infrared thermal imaging image.
3. The method of claim 2, wherein the preprocessing comprises: image rotation, image blurring, image sharpening, image sharpness, image size transformation, or any combination thereof.
4. The method of claim 1, wherein the pre-training the abnormal object detection network based on the annotated visible light image and the infrared thermographic image comprises:
respectively inputting the visible light image and the infrared thermal imaging image into a channel exchange feature extraction module of an abnormal target detection network to obtain corresponding image features;
carrying out feature fusion on the image features corresponding to the obtained visible light image and the image features corresponding to the infrared thermal imaging image by utilizing a visible-infrared feature fusion module to obtain fusion features;
and inputting the fusion characteristics into an abnormal target regression module, and training an abnormal target detection network based on the abnormal target prediction result and the marked visible light image and infrared thermal imaging image output by the abnormal target regression module.
5. The method of claim 4, wherein training the abnormal target detection network based on the abnormal target prediction result and the annotated visible light image and the infrared thermal imaging image output by the abnormal target regression module comprises:
obtaining an abnormal target prediction position, a position prediction confidence coefficient and a class probability distribution corresponding to an abnormal target output by an abnormal target regression module, and calculating a difference value among the position of the abnormal target, the detection confidence coefficient of the abnormal target and the class probability distribution corresponding to the abnormal target, which are marked corresponding to an input image, by adopting a loss function of yolov 4;
Based on the difference value, back propagation is performed to update the abnormal target detection network parameters;
judging whether the maximum iteration times are reached, if so, acquiring the abnormal target detection network parameters with the minimum loss function values in the iteration process, and obtaining an abnormal target detection network with the pre-trained completion; and if not, executing the step of acquiring the predicted position of the abnormal target, the confidence coefficient of the position prediction and the class probability distribution corresponding to the abnormal target, which are output by the abnormal target regression module.
6. The method according to any one of claims 1 to 5, wherein the performing, with a reward function, the reinforcement learning-based updating of the abnormal target detection network parameters of the pre-trained abnormal target detection network based on the abnormal target prediction position and the class corresponding to the abnormal target output by the pre-trained abnormal target detection network, and the labeled abnormal target position and the class corresponding to the abnormal target includes:
obtaining an abnormal target prediction position and a class corresponding to the abnormal target output by an abnormal target detection network based on pre-training, and a marked abnormal target position and a marked abnormal target class reward value by using a reward function;
calculating abnormal target detection network parameters of the abnormal target detection network based on pre-training based on the reward value by adopting a strategy gradient optimization algorithm;
Calculating the parameter gradient of the abnormal target detection network parameter, and updating the abnormal target detection network parameter according to the parameter gradient back propagation;
judging whether the maximum iteration times are reached, if so, acquiring the abnormal target detection network parameter with the highest rewarding function value in the iteration process, and obtaining the abnormal target detection network parameter as a final abnormal target detection network; if not, executing the step of utilizing the reward function to acquire the predicted position of the abnormal target, the class corresponding to the abnormal target, the position of the marked abnormal target and the reward value of the class corresponding to the abnormal target, which are output by the pre-trained abnormal target detection network.
7. The method of any one of claims 1 to 5, wherein labeling the visible light image comprises:
determining the position of an abnormal target of surrounding rock instability deformation of a working face in the visible light image from the acquired visible light image;
and acquiring detection confidence of the abnormal target and class probability distribution corresponding to the abnormal target according to the position of the abnormal target determined in each visible light image.
8. An apparatus for detecting stability of a surrounding rock, comprising:
the marking module is used for acquiring a visible light image and an infrared thermal imaging image of the surrounding rock of the sample tunnel and marking the visible light image and the infrared thermal imaging image respectively;
The pre-training module is used for pre-training the abnormal target detection network based on the marked visible light image and the infrared thermal imaging image;
the parameter updating module is used for updating parameters of the abnormal target detection network based on reinforcement learning by utilizing a reward function on the pre-trained abnormal target detection network based on the predicted position of the abnormal target, the type corresponding to the abnormal target, the marked position of the abnormal target and the type corresponding to the abnormal target;
the prediction module is used for acquiring a visible light image and an infrared thermal imaging image of the surrounding rock of the tunnel to be detected, inputting an abnormal target detection network for updating parameters of the abnormal target detection network, acquiring an abnormal target prediction result of the surrounding rock of the tunnel to be detected, and determining the stability of the surrounding rock of the tunnel to be detected based on the abnormal target prediction result.
9. A computer device, comprising: a processor, a memory and a bus, said memory storing machine readable instructions executable by said processor, said processor and said memory communicating via the bus when the computer device is running, said machine readable instructions when executed by said processor performing the steps of the method of detecting the stability of a surrounding rock as claimed in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that it has stored thereon a computer program which, when executed by a processor, performs the steps of the method of detecting the stability of a surrounding rock according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310300308.7A CN116363512A (en) | 2023-03-24 | 2023-03-24 | Method and device for detecting stability of surrounding rock |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310300308.7A CN116363512A (en) | 2023-03-24 | 2023-03-24 | Method and device for detecting stability of surrounding rock |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116363512A true CN116363512A (en) | 2023-06-30 |
Family
ID=86917126
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310300308.7A Pending CN116363512A (en) | 2023-03-24 | 2023-03-24 | Method and device for detecting stability of surrounding rock |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116363512A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117456461A (en) * | 2023-11-07 | 2024-01-26 | 北京城市轨道交通咨询有限公司 | Method and device for detecting violations of fire operation |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110956617A (en) * | 2019-11-18 | 2020-04-03 | 湖北工业大学 | Brain nuclear magnetic resonance abnormal image visualization method based on circulation attention model |
| CN113011446A (en) * | 2019-12-20 | 2021-06-22 | 中国科学院沈阳自动化研究所 | Intelligent target identification method based on multi-source heterogeneous data learning |
| CN113850242A (en) * | 2021-11-30 | 2021-12-28 | 北京中超伟业信息安全技术股份有限公司 | Storage abnormal target detection method and system based on deep learning algorithm |
-
2023
- 2023-03-24 CN CN202310300308.7A patent/CN116363512A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110956617A (en) * | 2019-11-18 | 2020-04-03 | 湖北工业大学 | Brain nuclear magnetic resonance abnormal image visualization method based on circulation attention model |
| CN113011446A (en) * | 2019-12-20 | 2021-06-22 | 中国科学院沈阳自动化研究所 | Intelligent target identification method based on multi-source heterogeneous data learning |
| CN113850242A (en) * | 2021-11-30 | 2021-12-28 | 北京中超伟业信息安全技术股份有限公司 | Storage abnormal target detection method and system based on deep learning algorithm |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117456461A (en) * | 2023-11-07 | 2024-01-26 | 北京城市轨道交通咨询有限公司 | Method and device for detecting violations of fire operation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | Multiscale U-shaped CNN building instance extraction framework with edge constraint for high-spatial-resolution remote sensing imagery | |
| US8447069B2 (en) | Apparatus and method for moving object detection | |
| US9607228B2 (en) | Parts based object tracking method and apparatus | |
| Hadjidemetriou et al. | Automated detection of pavement patches utilizing support vector machine classification | |
| CN111931582A (en) | Image processing-based highway traffic incident detection method | |
| CN111242026B (en) | Remote sensing image target detection method based on spatial hierarchy perception module and metric learning | |
| CN116052026B (en) | Unmanned aerial vehicle aerial image target detection method, system and storage medium | |
| CN116363512A (en) | Method and device for detecting stability of surrounding rock | |
| KR20240051005A (en) | Method for landslide crack detection based deep learning and Method for landslide monitoring therewith and Apparatus thereof | |
| WO2019073024A1 (en) | Lane sensing method | |
| Park et al. | Small and overlapping worker detection at construction sites | |
| CN116524017B (en) | Underground detection, identification and positioning system for mine | |
| CN110929782B (en) | River channel abnormity detection method based on orthophoto map comparison | |
| CN114332644B (en) | Large-view-field traffic density acquisition method based on video satellite data | |
| CN116740988A (en) | Full-time bridge self-adaptive ship collision prevention early warning method of visible light and thermal imaging camera | |
| CN114022760B (en) | Railway tunnel barrier monitoring and early warning method, system, equipment and storage medium | |
| CN111507126B (en) | Alarm method and device of driving assistance system and electronic equipment | |
| Pham et al. | Automatic detection and measurement of ground crack propagation using deep learning networks and an image processing technique | |
| Chen et al. | An image restoration and detection method for picking robot based on convolutional auto-encoder | |
| Lincy et al. | Road Pothole Detection System | |
| CN117787690A (en) | Hoisting operation safety risk identification method and identification device | |
| Lucks et al. | Superpixel-wise Assessment of Building Damage from Aerial Images. | |
| CN114882020B (en) | Product defect detection method, device, equipment and computer readable medium | |
| Huang et al. | MAST: An Earthquake Triggered Landslides Extraction Method Combining Morphological Analysis Edge Recognition with Swin-Transformer Deep Learning Model | |
| Zhang et al. | Vehicle detection and tracking in remote sensing satellite vidio based on dynamic association |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |