CN115700542A - Optical fiber pipeline safety early warning algorithm based on deep learning - Google Patents
Optical fiber pipeline safety early warning algorithm based on deep learning Download PDFInfo
- Publication number
- CN115700542A CN115700542A CN202110826427.7A CN202110826427A CN115700542A CN 115700542 A CN115700542 A CN 115700542A CN 202110826427 A CN202110826427 A CN 202110826427A CN 115700542 A CN115700542 A CN 115700542A
- Authority
- CN
- China
- Prior art keywords
- deep learning
- channel
- sublayer
- optical fiber
- network
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to an optical fiber pipeline safety early warning algorithm based on deep learning, which can effectively improve early warning accuracy and has high environmental adaptability, and is characterized by comprising the following steps: denoising an optical fiber signal acquired by a distributed optical fiber vibration sensing system to obtain a vibration ripple signal; calculating the characteristics of the vibration ripple signals of each spatial position point, wherein the signal characteristics of all the position points form space-time characteristic images in the time space dimension; obtaining local space-time characteristic images of the intrusion position points and the intrusion position points; constructing a channel-I conv2D network, constructing a channel-II conv1D-LSTM network, fusing the characteristics output by the two-channel network, fully utilizing the one-dimensional vibration ripple signal characteristics and the two-dimensional local space-time image characteristics of the intrusion point, and adding a full connection layer, a Dropout layer and a classification layer after the network fusion to form a two-channel deep learning network; and carrying out signal classification and identification, monitoring the invasion behavior of destroying the pipeline safety, and effectively shielding the interference.
Description
The technical field is as follows:
the invention relates to the technical field of distributed optical fiber vibration sensing signal processing, in particular to an optical fiber pipeline safety early warning algorithm based on deep learning, which can effectively improve early warning accuracy and has high environmental adaptability.
The invention content is as follows:
destructive events such as third-party construction and the like often occur around an oil and gas transmission pipeline in the petroleum and petrochemical industry, once oil and gas are leaked, serious economic losses can be caused to the country and the society, and the life and property safety of people is directly threatened. The traditional manual inspection method cannot meet the current pipeline safety monitoring requirement, in recent years, a distributed optical fiber vibration sensing system is widely applied to the field of pipeline safety of oil and gas industry, and compared with the traditional monitoring method, the optical fiber pipeline safety early warning technology has the advantages of long-distance monitoring, accurate positioning and quick response.
The intrusion alarm algorithm is the key for realizing the pipeline safety monitoring of the distributed optical fiber vibration sensing system, and the accurate optical fiber pipeline safety monitoring still faces the challenge due to the complexity of interference noise and the diversity of deployment environments. The algorithm aims at accurately positioning and quickly responding to destructive behaviors such as artificial excavation, mechanical construction and the like, and simultaneously shielding various environmental interferences such as vehicle passing and the like. How to improve the alarm accuracy of the optical fiber pipeline safety monitoring system and improve the algorithm environmental adaptability is a key problem to be solved urgently at present.
In the current optical fiber pipeline safety monitoring technology, a threshold value method or a traditional machine learning algorithm is poor in environmental adaptability and low in algorithm accuracy in practical application. Recently, deep learning algorithms are increasingly applied to the field of optical fiber pipeline safety monitoring due to good generalization performance of the algorithms, but most of the algorithms currently use the existing deep learning network to learn signal characteristics, and how to construct an efficient deep learning network to more fully mine and utilize behavior characteristics of different events still needs further research. In addition, the existing algorithm focuses on mining the time-frequency domain mixed features or feature image information of signals by using a deep learning network, and the insufficient mining of different event behavior features causes poor environmental adaptability of the algorithm and low alarm accuracy, and particularly has high false alarm rate of passing interference.
The invention content is as follows:
aiming at the defects and shortcomings in the prior art, the invention provides the deep learning-based optical fiber pipeline safety early warning which can effectively improve the early warning accuracy and has high environmental adaptability.
The invention is achieved by the following measures:
the deep learning-based optical fiber pipeline safety early warning algorithm is characterized by comprising the following steps of:
step 1: denoising an optical fiber signal acquired by a distributed optical fiber vibration sensing system to obtain a vibration ripple signal; calculating the characteristics of the vibration ripple signals of each spatial position point, wherein the signal characteristics of all the position points form a space-time characteristic image in the time-space dimension;
step 2: carrying out image detection on the time-space characteristic images to obtain intrusion position points and local time-space characteristic images of the intrusion position points;
and 3, step 3: constructing a first channel conv2D network, inputting a local space-time characteristic image of an intrusion position into the network, constructing a second channel conv1D-LSTM network, inputting a vibration ripple signal into the network, fusing the characteristics output by the two-channel network, fully utilizing the one-dimensional vibration ripple signal characteristics and the two-dimensional local space-time image characteristics of an intrusion point, and adding a full connection layer, a Dropout layer and a classification layer after the network fusion to form a two-channel deep learning network;
and 4, step 4: and carrying out signal classification and identification by using the model obtained by the network training, monitoring the invasion behavior of destroying the pipeline safety, and effectively shielding the interference.
In the step 1, the original optical fiber signal collected by the distributed optical fiber vibration sensing system is represented in the form of a two-dimensional matrix containing time domain information and space information: d t×l ＝(dij) t×l (i =1,2, t; j =1,2, l), where t represents the time dimension and l represents the number of spatial location points; firstly, obtaining a vibration ripple signal by adopting a wavelet denoising mode; and then calculating the zero-crossing rate characteristic of the vibration ripple signal of each spatial position point, wherein the zero-crossing rate characteristics of all the position points form a space-time characteristic image in a time-space dimension.
In step 3, the two-channel deep learning network comprises a parallel spatio-temporal image feature extraction layer, a signal vibration ripple feature extraction layer, a feature fusion layer, a full connection layer, a Dropout layer and a classification layer, wherein the channel one network comprises 3 two-dimensional convolution blocks (conv 2D), each convolution block comprises a convolution sublayer, a batch regularization sublayer, an activation function sublayer, a convolution sublayer, a batch regularization sublayer, an activation function sublayer and a pooling sublayer which are connected in sequence; the channel two network comprises 2 one-dimensional convolution blocks (conv 1D) and a long-short term memory network (LSTM layer), wherein each convolution block comprises a convolution sublayer, an activation function sublayer, a convolution sublayer, an activation function sublayer and a pooling sublayer which are connected in sequence; the dual-channel network is respectively flattened, subjected to feature fusion and sequentially connected with a full connection layer, a Dropout layer and a classification layer.
In the step 3, a Relu function is used in an activation function sublayer of the dual-channel deep learning network; maximum pooling (Max Pooling) is adopted for the pooling sublayer; selecting a global average pooling substitute by the flattening operation; the Dropout layer is used for preventing overfitting, neurons are inactivated randomly according to the probability of 0.5 in each iterative training process, the model is prevented from being excessively dependent on certain local characteristics, and the generalization performance of the model is improved; the classification layer uses the softmax function.
The step 3 of the invention also comprises the training of a deep learning model, and the specific process is as follows:
step 3-1: initializing parameters of the dual-channel deep learning model, wherein the parameters comprise a weight parameter w and a bias b;
step 3-2: inputting the characteristics of the training samples into the dual-channel deep learning model for forward propagation to obtain a sample prediction label;
step 3-3: calculating the loss values of the predicted label and the real label by using a cross entropy loss function, wherein the calculation formula is as follows:
wherein, x represents a sample, n represents the total number of the sample, and a and y are respectively a sample prediction label and a real label;
step 3-4: calculating the gradient of each learning parameter by utilizing the loss value back propagation, dynamically adjusting the learning rate of each parameter by utilizing an Adam optimization algorithm, updating the model parameters in a gradient descending manner, and minimizing a cross entropy loss function through continuous iteration;
step 3-5: and judging whether the dual-channel deep learning model is converged or not by using the loss value, if so, ending the training process, and otherwise, skipping to the step 3-2.
According to the method, a two-channel deep learning network is constructed, vibration ripple characteristics of signals are mined by using a Conv1D-LSTM network, and space-time image characteristics of optical fiber signals are mined by using a Conv2D network, so that the recognition accuracy of different behaviors is greatly improved; the optical fiber pipeline safety monitoring algorithm in the scheme can effectively monitor invasion behaviors threatening pipeline safety such as mechanical excavation and artificial excavation, can shield environmental interference such as vehicle passing and water pumps, is high in model universality and stability, and solves the technical problems of poor generalization and high false alarm rate of the existing algorithm model in different environments.
Description of the drawings:
FIG. 1 is a flow chart of the present invention.
FIG. 2 is a diagram of a dual channel deep learning network architecture in accordance with the present invention.
Fig. 3 is a schematic diagram of vibration ripples of filter signals with different behaviors in embodiment 1 of the present invention, where fig. 3 (a) is a schematic diagram of a mechanical excavation signal, fig. 3 (b) is a schematic diagram of an artificial excavation signal, fig. 3 (c) is a schematic diagram of a passing disturbance signal, and fig. 3 (d) is a schematic diagram of a water pump disturbance signal.
Fig. 4 is a spatiotemporal feature image of intrusion positions of different behaviors in embodiment 1 of the present invention, where fig. 4 (a) is a spatiotemporal feature image corresponding to a mechanical excavation signal, fig. 4 (b) is a spatiotemporal feature image corresponding to an artificial excavation signal, fig. 4 (c) is a spatiotemporal feature image corresponding to a passing-vehicle interference signal, and fig. 4 (d) is a spatiotemporal feature signal corresponding to a water pump interference signal.
Fig. 5 is cross entropy loss and precision variation curves in the model training process in embodiment 1 of the present invention, where fig. 5 (a) is a cross entropy loss curve and fig. 5 (b) is a training set and validation set precision variation curves.
FIG. 6 is a schematic diagram of an algorithm model test number confusion matrix in embodiment 1 of the present invention.
The specific implementation mode is as follows:
the invention is further described below with reference to the accompanying drawings and examples.
Example 1:
the embodiment provides an optical fiber pipeline safety early warning algorithm based on deep learning, the flow of the algorithm is shown as the attached figure 1, and the method specifically comprises the following steps:
the method comprises the steps that firstly, denoising original signals collected by a distributed optical fiber vibration sensing system to obtain vibration ripple signals; and calculating one or more characteristics of the vibration ripple signal of each spatial position point, wherein the signal characteristics of all the position points are represented as space-time characteristic images in the time-space dimension, and finally obtaining one or more space-time characteristic images.
And secondly, carrying out image detection on the space-time characteristic image to obtain an intrusion position point and a local space-time characteristic image of the intrusion position.
And thirdly, inputting a first construction channel conv2D network into an intrusion position space-time characteristic image, inputting a second construction channel conv1D-LSTM network into a vibration ripple signal, fusing the dual-channel characteristic results, adding a full connection layer, a Dropout layer and a classification layer to form a dual-channel deep learning network, and fully utilizing the intrusion point filtering signal vibration ripple characteristic and the intrusion position space-time image characteristic.
And fourthly, carrying out signal classification and identification by using the model obtained by training, monitoring invasion behaviors of damaging pipeline safety such as mechanical excavation and artificial excavation, and effectively shielding interference of passing vehicles and the like.
In the first step of this example, the raw signal collected by the distributed optical fiber vibration sensing system can be represented in the form of a two-dimensional matrix containing time domain information and spatial information: d t×l ＝(dij) t×l (i =1,2.. T; j =1,2.. L), where t represents the time dimension and l represents the number of spatial location points. Firstly, wavelet denoising is carried out on an original signal of each position point, and direct current components and high-frequency noise are filtered to obtain a vibration ripple signal. The vibration ripple signals of mechanical excavation, man-made excavation, vehicle passing interference and water pump interference are shown in figures 3 (a) to 3 (d);
and calculating the zero crossing rate of the vibration ripple signal of each spatial position point, wherein the zero crossing rate characteristics of all the position points form a space-time characteristic image in a time-space dimension. And carrying out image detection operation on the space-time characteristic image to obtain an intrusion position space-time image, and taking the central point of the space-time image as an intrusion position point. FIGS. 4 (a) -4 (d) are respectively time-space characteristic images of mechanical excavation, man-made excavation, vehicle-passing interference and water pump interference.
Then, a dual-channel deep learning network model is constructed, and the network structure is shown in fig. 2:
the double-channel deep learning network comprises a spatio-temporal image feature extraction layer, a signal vibration ripple feature extraction layer, a feature fusion layer, a full connection layer, a Dropout layer and a classification layer which are parallel.
The first channel network comprises 3 two-dimensional convolution blocks (conv 2D), wherein each convolution block comprises a convolution sublayer, a batch regularization sublayer, an activation function sublayer, a convolution sublayer, a batch regularization sublayer, an activation function sublayer and a pooling sublayer which are sequentially connected.
The second channel network comprises 2 one-dimensional convolution blocks (conv 1D) and a long-short term memory network (LSTM layer), wherein each convolution block comprises a convolution sublayer, an activation function sublayer, a convolution sublayer, an activation function sublayer and a pooling sublayer which are connected in sequence; the dual-channel network is respectively flattened, subjected to characteristic fusion and sequentially connected with a full connection layer, a Dropout layer and a classification layer; relu function is used in the activation function sublayer; the pooling sublayer adopts maximum pooling (Max Pooling); the flattening operation selects a global average pooling replacement; the Dropout layer is used for preventing overfitting, neurons are inactivated randomly according to the probability of 0.5 in each iterative training process, the model is prevented from being excessively dependent on certain local characteristics, and the generalization performance of the model is improved; the classification layer uses a softmax function.
Then, deep learning model training is carried out, and the specific process is as follows: (1) Initializing parameters of the dual-channel deep learning model, wherein the parameters comprise a weight parameter w and a bias b; (2) Inputting the characteristics of the training samples into the dual-channel deep learning model for forward propagation to obtain a sample prediction label; (3) Calculating the loss values of the predicted label and the real label by using a cross entropy loss function, wherein the calculation formula is as follows:
wherein, x represents a sample, n represents the total number of samples, and a and y are a sample prediction label and a true label respectively. Calculating the gradient of each learning parameter by utilizing the loss value back propagation, dynamically adjusting the learning rate of each parameter by utilizing an Adam optimization algorithm, updating the model parameters in a gradient descending manner, and minimizing a cross entropy loss function through continuous iteration;
(4) And (4) judging whether the dual-channel deep learning model is converged or not by using the loss value, if so, ending the training process, and otherwise, skipping to the step (2).
Fig. 5 shows cross-entropy loss and accuracy during training. The horizontal axes of fig. 5 (a) and (b) each represent the number of iterations, and the vertical axes represent the cross-entropy loss and the precision, respectively, where the blue curve is the cross-entropy loss and the precision of the training samples, and the orange curve is the cross-entropy loss and the precision of the validation samples (10% of the samples are divided as the validation set, and do not participate in the training). It can be seen that in the network training process, the training precision and the verification precision are stably improved, and meanwhile, the cross entropy loss is rapidly and stably converged.
Calling a deep learning model obtained by training to perform signal real-time classification recognition and judge and alarm: inputting the vibration ripples of all the intrusion center points and the space-time characteristic images corresponding to the vibration ripples into a dual-channel deep learning network model, wherein the vibration ripple signals of the intrusion center points are used as the input of a conv1D-LSTM network, the space-time characteristic images are used as the input of a conv2D network, calling the deep learning model obtained through training for signal classification and identification, and outputting a real-time model identification result.
And finally, judging the real-time identification result of the model, outputting alarm information of a corresponding position if the model is identified as mechanical excavation or artificial excavation, and otherwise, not outputting the alarm.
Model training is carried out by adopting the two-channel deep learning network structure constructed by the method, and the model training precision and the verification precision are respectively 99.36% and 98.95%. 500 test samples were selected for each class for model accuracy testing, and the confusion matrix is shown in fig. 6. The category label 0 is mechanical excavation, 1 is artificial excavation, 2 is vehicle passing interference, and 3 is water pump interference, and the test result algorithm model has a false alarm rate of 1% and a false alarm rate of 0.7%.
The algorithm model can accurately monitor intrusion behaviors damaging pipeline safety such as mechanical excavation, artificial excavation and the like, can effectively shield interference behaviors such as vehicle passing, water pump continuous interference and the like, can adapt to various environmental scenes such as roads, wastelands, farmlands, ponds and the like and optical cable burial depth, and has high algorithm accuracy and universality.
Claims (5)
1. The optical fiber pipeline safety early warning algorithm based on deep learning is characterized by comprising the following steps of:
step 1: denoising an original signal acquired by a distributed optical fiber vibration sensing system to obtain a vibration ripple signal; calculating the characteristics of the vibration ripple signals of each space position, wherein the signal characteristics of all position points are represented as space-time characteristic images in the time space dimension, and the space-time characteristic images are obtained;
and 2, step: carrying out image detection on the time-space characteristic image to obtain an intrusion position point and an intrusion position time-space characteristic image;
and 3, step 3: constructing a first channel conv2D network input as an intrusion position space-time characteristic image, constructing a second channel conv1D-LSTM network input as a signal vibration ripple signal, fusing a dual-channel characteristic result, adding a full connection layer, a Dropout layer and a classification layer to form a dual-channel deep learning network, and fully utilizing the intrusion point filtering signal vibration ripple characteristic and the intrusion position space-time image characteristic;
and 4, step 4: and carrying out signal classification and identification by using the model obtained by training, monitoring the invasion behavior of destroying the pipeline safety, and effectively shielding the interference.
2. The deep learning-based optical fiber pipeline safety early warning algorithm according to claim 1, wherein the original signal collected by the distributed optical fiber vibration sensing system in step 1 is represented in the form of a two-dimensional matrix containing time domain information and spatial information: d t×l ＝(dij) t×l (i =1,2, t; j =1,2, l), where t represents the time dimension and l represents the number of spatial location points; firstly, performing wavelet denoising on an original signal of each position point, and filtering a direct current component and high-frequency noise to obtain a vibration ripple signal; and calculating the zero crossing rate of the vibration ripple signal of each spatial position point, wherein the zero crossing rate characteristics of all the position points form a space-time characteristic image in the time-space dimension.
3. The optical fiber pipeline safety early warning algorithm based on deep learning in the step 3 is characterized in that in the step 3, a dual-channel deep learning network comprises a parallel spatio-temporal image feature extraction layer, a signal vibration ripple feature extraction layer, a feature fusion layer, a full connection layer, a Dropout layer and a classification layer, wherein the channel one network comprises 3 two-dimensional convolution blocks, each convolution block comprises a convolution sublayer, a batch regularization sublayer, an activation function sublayer, a convolution sublayer, a batch regularization sublayer, an activation function sublayer and a pooling sublayer which are connected in sequence; the second channel network comprises 2 one-dimensional convolution blocks and a long-short term memory network, wherein each convolution block comprises a convolution sublayer, an activation function sublayer, a convolution sublayer, an activation function sublayer and a pooling sublayer which are connected in sequence; the dual-channel network is respectively flattened, subjected to feature fusion and sequentially connected with a full connection layer, a Dropout layer and a classification layer.
4. The deep learning-based optical fiber pipeline safety early warning algorithm according to claim 3, wherein a Relu function is used in an activation function sublayer of the two-channel deep learning network in the step 3; the pooling sublayer selects maximum pooling; selecting a global average pooling substitute by the flattening operation; the Dropout layer is used for preventing overfitting, neurons are inactivated randomly according to the probability of 0.5 in each iterative training process, the model is prevented from being over dependent on certain local characteristics, and the generalization performance of the model is improved; the classification layer uses a softmax function.
5. The optical fiber pipeline safety early warning algorithm based on deep learning as claimed in claim 3, wherein step 3 further comprises training of a deep learning model, and the specific process is as follows:
step 3-1: initializing parameters of the dual-channel deep learning model, wherein the parameters comprise a weight parameter w and a bias b;
step 3-2: inputting the characteristics of the training samples into the dual-channel deep learning model for forward propagation to obtain a sample prediction label;
step 3-3: calculating the loss values of the predicted label and the real label by using a cross entropy loss function, wherein the calculation formula is as follows:
wherein x represents a sample, n represents the total number of samples, and a and y are a sample prediction label and a real label respectively;
step 3-4: calculating the gradient of each learning parameter by utilizing the loss value back propagation, dynamically adjusting the learning rate of each parameter by utilizing an Adam optimization algorithm, updating the model parameters in a gradient descending manner, and minimizing a cross entropy loss function through continuous iteration;
step 3-5: and judging whether the dual-channel deep learning model is converged or not by using the loss value, if so, ending the training process, and otherwise, skipping to the step 3-2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110826427.7A CN115700542A (en) | 2021-07-21 | 2021-07-21 | Optical fiber pipeline safety early warning algorithm based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110826427.7A CN115700542A (en) | 2021-07-21 | 2021-07-21 | Optical fiber pipeline safety early warning algorithm based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115700542A true CN115700542A (en) | 2023-02-07 |
Family
ID=85120692
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110826427.7A Pending CN115700542A (en) | 2021-07-21 | 2021-07-21 | Optical fiber pipeline safety early warning algorithm based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115700542A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116186642A (en) * | 2023-04-27 | 2023-05-30 | 山东汇英信息科技有限公司 | Distributed optical fiber sensing event early warning method based on multidimensional feature fusion |
| CN117132601A (en) * | 2023-10-27 | 2023-11-28 | 山东飞博赛斯光电科技有限公司 | Pipeline mode identification method and system based on distributed optical fiber sensing |
| CN117312828A (en) * | 2023-09-28 | 2023-12-29 | 光谷技术有限公司 | Public facility monitoring method and system |
-
2021
- 2021-07-21 CN CN202110826427.7A patent/CN115700542A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116186642A (en) * | 2023-04-27 | 2023-05-30 | 山东汇英信息科技有限公司 | Distributed optical fiber sensing event early warning method based on multidimensional feature fusion |
| CN116186642B (en) * | 2023-04-27 | 2023-09-08 | 山东汇英光电科技有限公司 | Distributed optical fiber sensing event early warning method based on multidimensional feature fusion |
| CN117312828A (en) * | 2023-09-28 | 2023-12-29 | 光谷技术有限公司 | Public facility monitoring method and system |
| CN117132601A (en) * | 2023-10-27 | 2023-11-28 | 山东飞博赛斯光电科技有限公司 | Pipeline mode identification method and system based on distributed optical fiber sensing |
| CN117132601B (en) * | 2023-10-27 | 2024-01-23 | 山东飞博赛斯光电科技有限公司 | Pipeline mode identification method and system based on distributed optical fiber sensing |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115700542A (en) | Optical fiber pipeline safety early warning algorithm based on deep learning | |
| US11562224B2 (en) | 1D-CNN-based distributed optical fiber sensing signal feature learning and classification method | |
| CN108709633B (en) | Distributed optical fiber vibration sensing intelligent safety monitoring method based on deep learning | |
| CN112287816B (en) | Dangerous work area accident automatic detection and alarm method based on deep learning | |
| CN107016357B (en) | Video pedestrian detection method based on time domain convolutional neural network | |
| CN110853074B (en) | Video target detection network system for enhancing targets by utilizing optical flow | |
| CN114692681B (en) | SCNN-based distributed optical fiber vibration and acoustic wave sensing signal identification method | |
| CN112562255B (en) | Intelligent image detection method for cable channel smoke and fire conditions in low-light-level environment | |
| CN110826429A (en) | Scenic spot video-based method and system for automatically monitoring travel emergency | |
| CN106815576B (en) | Target tracking method based on continuous space-time confidence map and semi-supervised extreme learning machine | |
| CN113313037A (en) | Method for detecting video abnormity of generation countermeasure network based on self-attention mechanism | |
| CN108757426B (en) | Fault diagnosis method for water injection plunger pump in petroleum mine field | |
| CN116629465B (en) | Smart power grids video monitoring and risk prediction response system | |
| CN112738014A (en) | Industrial control flow abnormity detection method and system based on convolution time sequence network | |
| CN112949579A (en) | Target fusion detection system and method based on dense convolution block neural network | |
| CN114926825A (en) | Vehicle driving behavior detection method based on space-time feature fusion | |
| CN112541480A (en) | Online identification method and system for tunnel foreign matter invasion event | |
| Kadim et al. | Deep-learning based single object tracker for night surveillance. | |
| CN115964670A (en) | Frequency spectrum anomaly detection method | |
| Kim et al. | Video anomaly detection using cross u-net and cascade sliding window | |
| CN110111358B (en) | Target tracking method based on multilayer time sequence filtering | |
| CN110043808B (en) | Water supply network leakage monitoring and early warning method based on time series analysis | |
| Al-Shammri et al. | A combined method for object detection under rain conditions using deep learning | |
| CN114595448B (en) | Industrial control anomaly detection method, system and equipment based on correlation analysis and three-dimensional convolution and storage medium | |
| CN115951418A (en) | Magnetic anomaly detection method based on deep reinforcement learning |
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 |