KR102167545B1 - Apparatus for detecting electric supply equipment - Google Patents

Abstract

The present invention discloses a power distribution facility detection device. The distribution facility detection apparatus of the present invention includes: a photographed image input unit for receiving a photographed image photographed of the distribution equipment; An object detection framework module that detects and learns an object by encoding information on an object to be detected in the captured image input from the captured image input unit and extracting a candidate region of interest; A prediction criticism module that predicts and criticizes the area of the object and the area of the object from the information encoded by the information on the object detected by the object detection framework module; And an output unit for outputting the object detected by the object detection framework module.

Description

Distribution facility detection device{APPARATUS FOR DETECTING ELECTRIC SUPPLY EQUIPMENT}

The present invention relates to a power distribution facility detection device, and more particularly, when detecting a power distribution facility based on an image captured in a driving environment, a problem that is not detected due to the masking phenomenon caused by road facilities is covered by deep learning. It relates to a distribution facility detection device capable of robustly detecting a distribution facility even in a hidden phenomenon by using a map predicting

In general, since the safety of a distribution line may be damaged by environmental changes or damage due to equipment failures around the distribution line, continuous inspection, removal of safety hazards, and management of equipment replacement are required.

Various methods of instantaneous inspection are used to prevent failure of power facilities related to distribution lines.The instantaneous method of distribution lines inspected with the naked eye through a distribution line patrol source, and a live line electric power facility by riding a live line bucket truck in a live line condition. A method for inspecting each live wire that measures the shared voltage using a visual and fork-type suspension insulator shared voltage measuring device, and a thermal imaging camera measurement method to prevent breakdown of distribution lines by measuring heat generated when deterioration in power facilities progresses. , An instantaneous method using RFI (RADIO FREQUENCY INTERFERENCE) is used.

However, in reality, it is impossible to measure an abnormality of the facility with the naked eye because the electric power facility exists in the process and the distance is more than 10m from the ground.

In addition, since the live wire operator checks the live wire by using a live wire bucket truck, it is possible to visually see a facility at a close distance, but in the case of a small area between the insulator and the insulator or minute cracks on the other side from the inspector. Very difficult to check.

Therefore, it is a reality that an electric power accident occurs even in an area that has been inspected for each live line due to poor inspection. Also, the inspection for each live line takes a lot of time and consumes a relatively higher cost, so it is inefficient.

On the other hand, since the thermal imaging camera measurement method measures heat generated when deterioration proceeds in the distribution line facility to prevent failure, extraction is not possible in the case of a facility that does not generate heat.

In fact, in high-voltage lines of power facilities, in most cases, heat is not generated from cracks in insulators.

In addition, it cannot be said that it is an effective detection method because heat is not generated even in the case of contact with foreign objects or loosening of the bolt.

In addition, the instantaneous method using RFI is difficult to distinguish from the noise from the power facility because the surrounding frequency noise is so severe in the case of an urban area or a densely concentrated area of a factory, and even if it is distinguished, it is impossible to accurately distinguish a facility with a possibility of failure. Do. In addition, since it is impossible to analyze the waveform, it is difficult to grasp the type or degree of the failure.

Therefore, in order to solve this problem, there is a method to check the defect by receiving the ultrasonic signal generated from the power facility, but since the angle must be adjusted while the ultrasonic detection device is held, it is difficult to perform accurate measurement due to vibration and even when setting the direction. The disadvantage is that the detection work is very cumbersome and difficult since the operator must turn the body and adjust the direction.

The background technology of the present invention is disclosed in Korean Patent Application Publication No. 2011-0070969 (published on June 27, 2011, ultrasonic detection device for checking defects in power facilities of overhead distribution lines).

Recently, by monitoring the distribution facilities through object detection based on the images taken of the distribution facilities seen on the road while driving, the monitoring and diagnosis methods that depend on the experience of the existing patrolmen and patrollers have been avoided, and the monitoring through object detection has been conducted. Diagnosis technology is being introduced.

The field of object detection is one of the most actively researched fields in the field of computer vision. It automatically classifies the type of object to be searched for in an image, and automatically identifies the location and area of the separated object. -box).

However, when object detection is performed based on images captured in the driving environment, problems such as occlusion and scale variance caused by automobiles, pedestrians, trees, etc., depending on the road environment, occur. do.

In particular, there is a problem in that the distribution equipment covered by the screening phenomenon is frequently missed when performing detection.

The present invention was conceived to improve the above problems, and an object of the present invention according to one aspect is that when detecting a power distribution facility based on an image captured in a driving environment, it is not detected due to the obscuration caused by road facilities. It is to provide a distribution facility detection device capable of robustly detecting distribution facilities even in the case of screening by using a map predicting the screening phenomenon based on deep learning for the problem.

An apparatus for detecting a distribution facility according to an aspect of the present invention includes: a photographed image input unit for receiving a photographed image photographed of the distribution equipment; An object detection framework module that detects and learns an object by encoding information on an object to be detected in the captured image input from the captured image input unit and extracting a candidate region of interest; A prediction criticism module that predicts and criticizes the area of the object and the area of the object from the information encoded by the information on the object detected by the object detection framework module; And an output unit for outputting the object detected by the object detection framework module.

In the present invention, the object detection framework module includes: a feature point encoding module for encoding a feature of an object from a photographed image into a feature map; A region of interest extraction module for extracting a region of interest of the object from the feature map encoded by the feature point encoding module; And an object prediction module for predicting the type and area of the object by applying the ROI of the object extracted by the ROI extraction module to the feature point encoding module.

In the present invention, the region of interest extraction module is characterized in that the region of interest is extracted in the form of a bounding box.

In the present invention, the prediction criticism module includes: an object bounding box prediction module that predicts a bounding box region and predicts a masked region of the bounding box region through spatial analysis of a feature map encoded by the feature point encoding module; And a bounding box criticism module that compares and determines the bounding box region predicted by the object bounding box prediction module with an actual correct answer bounding box map.

In the present invention, the object bounding box prediction module includes: a global bounding box region prediction module that predicts an object bounding box region and predicts a mask region by using a feature map encoded by the feature point encoding module; And a region bounding box region prediction module that predicts a bounding box region of an object and predicts a hidden region using the feature points of the region of interest extracted by the region of interest extraction module.

In the present invention, the global bounding box region prediction module is characterized in that it has a structure of any one of a convolutional neural network (CNN) and a convolutional auto encoder.

In the present invention, the regional bounding box region prediction module is characterized in that it has a convolutional neural network (CNN) structure.

In the present invention, the bounding box criticism module includes: a global criticism module for comparing and criticizing a global bounding box predicted by the global bounding box region prediction module and an actual global correct answer map; And a region criticism module that compares and criticizes the region bounding box predicted by the region bounding box region prediction module with an actual region correct answer map.

In the present invention, the bounding box critique module is characterized in that it has a structure of any one of a convolutional neural network (CNN) and a residual network (ResNet).

The present invention is characterized in that the object detection framework module detects an object by performing adversarial learning on the predictive criticism module in a plug-in method.

The distribution facility detection device according to an aspect of the present invention predicts a masking phenomenon based on deep learning for a problem that is not detected due to the masking phenomenon caused by road facilities when detecting a distribution facility based on an image captured in a driving environment. By using a map, it is possible to detect power distribution facilities robustly even in masking phenomena, enabling instantaneous monitoring of power distribution facilities using vehicles.

1 is a block diagram showing a distribution facility detection apparatus according to an embodiment of the present invention.
2 is a conceptual diagram illustrating an apparatus for detecting a distribution facility according to an embodiment of the present invention.
3 is a block diagram showing an object detection framework module of a distribution facility detection apparatus according to an embodiment of the present invention.
4 is a block diagram showing a predictive criticism module of a distribution facility detection apparatus according to an embodiment of the present invention.
5 is an exemplary view showing a bounding box and a hidden area for explaining a distribution facility detection apparatus according to an embodiment of the present invention.

Hereinafter, an apparatus for detecting a distribution facility according to the present invention will be described with reference to the accompanying drawings. In this process, the thickness of the lines or the size of components shown in the drawings may be exaggerated for clarity and convenience of description. In addition, terms to be described later are terms defined in consideration of functions in the present invention and may vary according to the intention or custom of users or operators. Therefore, definitions of these terms should be made based on the contents throughout the present specification.

1 is a block diagram showing a distribution facility detection apparatus according to an embodiment of the present invention, FIG. 2 is a conceptual diagram for explaining a distribution facility detection apparatus according to an embodiment of the present invention, and FIG. A block diagram showing an object detection framework module of a distribution facility detection apparatus according to an embodiment, and FIG. 4 is a block diagram showing a prediction criticism module of a distribution facility detection apparatus according to an embodiment of the present invention, and FIG. 5 Is an exemplary view showing a bounding box and a shielding area for explaining a distribution facility detection apparatus according to an embodiment of the present invention.

As shown in FIGS. 1 and 2, the apparatus for detecting a distribution facility according to an embodiment of the present invention includes a photographed image input unit 10, an object detection framework module 20, and a predictive criticism module. An OBB-Critic Network) 30 and an output unit 40 may be included.

The photographed image input unit 10 receives a photographed image photographed of a power distribution facility in a driving environment and inputs it to the object detection framework module 20.

The object detection framework module 20 may detect and learn the object by encoding information on an object to be detected from the captured image input from the captured image input unit 10 and extracting a candidate ROI.

Here, the object detection framework module may include a feature encoding module 210, an ROI extraction module 220, and an object prediction module 230, as shown in FIG. 3.

The feature point encoding module 210 learns a deep network hierarchically extracting feature information of an object, which is a distribution facility, from a captured image, and encodes the feature of the object into a feature map. That is, by learning to activate the area of the object to be detected in the captured image, the feature of the object is encoded as a feature map.

The region of interest extraction module 220 learns a deep network for extracting the region of interest of an object to be detected from a feature map encoded by the feature point encoding module 210. That is, the location where the distribution facility is determined to exist can be analyzed through the deep network, and the location of the ROI can be extracted in the form of a bounding box to extract the ROI of the object.

The object prediction module 230 applies the ROI of the object extracted by the ROI extraction module 220 to the feature point encoding module 210 to classify the type of the object, which is a distribution facility, and finally Objects can be predicted by learning and analyzing deep networks that predict areas.

In this way, the object detection framework module 20 may detect the object by analyzing the type and area of each object from the captured image based on deep learning through a deep network and extracting features.

The predictive criticism module 30 may predict and criticize the area of the object and the area to be hidden from information on which information on the object detected by the object detection framework module 20 is encoded.

Here, the prediction critique module 30 may include an object bounding box prediction module 310 and a bounding box critique module 320 as shown in FIG. 4.

The object bounding box prediction module 310 predicts the bounding box region of an object based on the feature map encoded by the feature point encoding module 210 and learns a deep network for classifying the object type, and the feature point encoding module 210 Through spatial analysis of the encoded feature map, the bounding box region can be predicted and the hidden region of the bounding box region can be predicted.

As for the bounding box area and the masking area, as shown in FIG. 5, an area in which the bounding box of one object does not overlap the bounding box area of another object is assigned to each bounding box map as 1, otherwise Assign it to 0. The area where the object's bounding box overlaps is determined as an occlusion area and assigned as 1, otherwise it is assigned as 0.

In addition, the bounding box criticism module 320 compares the bounding box region predicted by the object bounding box prediction module 310 with an actual ground truth bounding box map to determine whether it is close to the bounding box map. If you learn to predict a result close to the actual correct answer, you can judge it as real, otherwise it can be judged as false.

Thereafter, the bounding box critique module 320 is added to the deep learning-based object detection framework module 20 in a "Plug-and-Play" manner to be learned.

Here, the object bounding box prediction module 310 may include a global object bounding box area prediction module 312 and a local object bounding box area prediction module 314.

The global object bounding box region prediction module 312 may be configured as a deep network that predicts an object bounding box region and predicts a blinding region using a feature map encoded by the feature point encoding module 210.

In this case, the global bounding box region prediction module may have a structure of any one of a convolutional neural network (CNN) and a convolution auto encoder useful for analyzing spatial information of an object and learning features.

In addition, the local object bounding box region prediction module 314 predicts the bounding box region of an object and predicts the hidden region using the feature points of the region of interest extracted from the region of interest extraction module 220. It can be configured as a network.

In this case, the regional bounding box region prediction module 314 may be configured with a convolutional neural network (CNN) structure useful for analyzing spatial information of an object and learning features.

Meanwhile, the bounding box critique module 320 may include a global critique module 322 and a local critique module 324.

The global critique module 322 may be configured as a deep network that compares the global bounding box predicted by the global bounding box region prediction module 312 with an actual global ground truth map to criticize whether the answer is an actual correct answer.

In addition, the region criticism module 324 may be configured as a deep network that compares the region bounding box predicted by the region bounding box region prediction module 314 with an actual region correct answer (Ground Truth) map to criticize whether the answer is actually correct.

In this case, the bounding box critique module 320 may be configured in any one of a convolutional neural network (CNN) and a residual network (ResNet) useful for analyzing spatial information of an object and learning features.

Here, the object detection framework module 20 may detect an object by performing adversarial learning on the prediction criticism module 30 in a plug-in method.

First, the bounding box critique module 320 may be trained. The loss function used for learning the bounding box critique module 320 may be expressed as Equation 1.

Here, L _{Global -Critic} and L _{Local -Critic} mean a global critique module 322 and a local critique module 324, respectively, and λ _{Global -Critic} and λ _{Local -Critic} refer to a global critique module 322 and a local critique module. Means hyper-parameter of 324

In addition, the loss functions of the global criticism module 322 and the local criticism module 324 can be expressed as Equations 2 and 3, respectively.

는 비평 네트워크의 binary logistic 손실함수이고, x_i 는 i 번째 입력영상이고, y_i ^Global 와 y_i ^Local 는 i 번째 전역 및 지역 맵이다. 또한 S^t(x_i)는 i 번째 전역 바운딩 박스를 의미하고, t 는 high layer 또는 low layer의 바운딩 박스 맵을 의미한다. 이 2개의 손실 함수를 최소화 하도록 전역 비평 모듈(322) 및 지역 비평 모듈(324)을 학습한다.From here,

Is the binary logistic loss function of the critical network, x _i is the i-th input image, and y _i ^Global and y _i ^Local are the i-th global and regional maps. In addition, S ^t (x _i ) refers to the i-th global bounding box, and t refers to the high layer or low layer bounding box map. The global critique module 322 and the local critique module 324 are trained to minimize these two loss functions.

Next, the loss function used when learning the object bounding box prediction module 310 and learning the object detection framework module 20 can be expressed as Equation 4 below.

Here, L _Estimator means the object bounding box prediction module 310, and L _OD means the object detection framework module 20. λ _OD means a hyper-parameter of the object detection framework module 20.

In addition, the loss functions of the object bounding box prediction module 310 and the object detection framework module 20 may be expressed as Equations 5 and 6, respectively.

Here, L _{RPN - CLS} and L _{RPN - REG} are classification and regression loss functions of the ROI extraction module 220, respectively. L _CLS and L _REG are classification and regression loss functions of the object prediction module 230 that predict classification and region, respectively. L _Global and L _Local are loss functions of the global bounding box area prediction module 312 and the local bounding box area prediction module 314, respectively.

At this time, the L1 loss function was used as the regression loss function, and the global bounding box region prediction module 312 and the regional bounding box region prediction module 314 were used for the other loss functions using a cross-entropy loss function. The object detection framework module 20 is learned.

The output unit 40 may output the object detected by the object detection framework module 20.

As described above, according to the power distribution facility detection apparatus according to an embodiment of the present invention, when detecting a power distribution facility based on an image captured in a driving environment, deep learning about a problem that is not detected due to the obscuration by road facilities. Based on the map predicting the masking phenomenon, it is possible to detect the distribution equipment robustly even in the case of the masking phenomenon, so that instantaneous monitoring of the distribution equipment using vehicles can be made possible.

Although the present invention has been described with reference to the embodiments shown in the drawings, this is only illustrative, and those of ordinary skill in the field to which the technology pertains, various modifications and other equivalent embodiments are possible. I will understand.

Therefore, the true technical protection scope of the present invention should be determined by the following claims.

10: photographed image input unit 20: object detection framework module
30: predictive criticism module 40: output unit
210: feature point encoding module 220: region of interest extraction module
230: object prediction module 310: object bounding box prediction module
312: Global bounding box area prediction module
314: region bounding box region prediction module
320: bounding box critique module 322: global critique module
324: regional critique module

Claims (10)

A photographed image input unit for receiving a photographed image photographed of the distribution facility;
An object detection framework module for detecting and learning an object by encoding information on an object to be detected from the captured image input from the captured image input unit and extracting a candidate region of interest;
A predictive criticism module for predicting and criticizing an object region and a hidden region from information encoded with information on the object detected by the object detection framework module; And
Including; an output unit for outputting the object detected by the object detection framework module,
The object detection framework module,
A feature point encoding module for encoding a feature of an object from the captured image into a feature map;
A region of interest extraction module for extracting a region of interest of an object from a feature map encoded by the feature point encoding module; And
An object prediction module for predicting the type and area of the object by applying the ROI of the object extracted by the ROI extraction module to the feature point encoding module,
The predictive criticism module,
An object bounding box prediction module for predicting a bounding box region through spatial analysis of a feature map encoded by the feature point encoding module and predicting a masked region of the bounding box region; And
And a bounding box criticism module that compares and determines the bounding box region predicted by the object bounding box prediction module with an actual correct answer bounding box map, and comprises
The object bounding box prediction module,
A global bounding box region prediction module for predicting the object bounding box region and predicting the mask region using the feature map encoded by the feature point encoding module; And
And a region bounding box region prediction module for predicting the bounding box region of an object using the feature points of the region of interest extracted by the region of interest extraction module and predicting the masked region; .
delete The apparatus of claim 1, wherein the region of interest extraction module extracts the region of interest in the form of a bounding box.
delete delete The apparatus of claim 1, wherein the global bounding box region prediction module has a structure of any one of a convolutional neural network (CNN) and a convolution auto encoder.
The apparatus of claim 1, wherein the region bounding box region prediction module has a convolutional neural network (CNN) structure.
The method of claim 1, wherein the bounding box critique module,
A global criticism module for comparing and criticizing a global bounding box predicted by the global bounding box region prediction module with an actual global correct answer map; And
And a region criticism module for comparing and criticizing the region bounding box predicted by the region bounding box region prediction module with an actual region correct answer map.
The apparatus of claim 1, wherein the bounding box critique module has a structure of any one of a convolutional neural network (CNN) and a residual network (ResNet).
The apparatus of claim 1, wherein the object detection framework module detects an object by performing adversarial learning on the predictive criticism module in a plug-in method.

Images