A data set sample collection device for identifying cable damage based on deep neural network
By designing a data set sample collection device for ADSS optical cable damage detection, using mobile cameras and environmental simulation devices, the problem of collecting multi-environmental conditions and background sample sets is solved, and the detection accuracy and generalization capabilities of deep neural networks are improved, so as to achieve The accuracy and reliability of automatic damage recognition of line patrol robots.
Patent Information
- Application Number
- CN202010763990.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-08-01
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2040-08-01
AI Technical Summary
The prior art is difficult to obtain a sufficient number of cable damage image sample sets with multiple environmental conditions and backgrounds, resulting in low accuracy of deep neural networks in ADSS optical cable damage detection, especially for sample sets taken by specific angles of line patrol robots. .
Design a data set sample collection device, including a bracket, a mobile camera and an environmental simulation device, quickly generate cable damage image data sets with different backgrounds and light conditions through a virtual environment, and realize the camera using a lead screw slide mechanism driven by a stepper motor. Move along the cable length direction and simulate natural light environments through LCD screen and adjustable light sources.
In the case of a small number of cable damage samples, quickly generate a large number of image sample sets to improve the detection accuracy of deep neural networks, avoid overfitting, enhance generalization capabilities, and realize the accuracy and reliability of automatic damage recognition during line patrols. .
Smart Images

Figure CN111914824B_ABST
Abstract
Description
Technical field:
[0001] The present invention relates to a data set sample acquisition device for identifying cable damage based on a deep neural network. Background technology:
[0002] ADSS optical cable is a full-dielectric, metal-free, non-conductive communication cable. It is widely used in communication facilities of power systems. Its safety is of great significance to the normal and stable operation of power systems. ADSS is built with transmission lines and hung in the middle of power towers, close to the safe distance of the conductors. It is vulnerable to electrical corrosion. Because there are no metal parts inside the ADSS optical cable, it cannot prevent small animals from biting it, which can easily cause damage to the optical cable. Therefore, automatic damage detection can be performed based on deep learning image target recognition technology by mounting automatic identification devices on patrol robots or drones, which can effectively improve detection efficiency, save manpower, realize intelligent power grids, and ensure the safety of ADSS optical cables.
[0003] Target detection technology based on deep neural networks has achieved great success in the field of image processing in recent years. By building a high-performance convolutional neural network and pre-collecting a sufficient number of target image samples for neural network training, automatic recognition and positioning of targets can be achieved, which can be used for automatic detection of various types of damage to ADSS optical cables.
[0004] The quantity and quality of image data sets collected have a great impact on the training effect of deep neural networks and the accuracy of ADSS optical cable damage detection. In actual environments, the cable defect images taken by line patrol robots may contain various complex backgrounds and light of different weather, intensity and angles. In order to accurately identify cable damage in different environments, a large number of cable damage image sample sets containing multiple ambient light conditions and multiple backgrounds are required for deep neural network training. However, since such image data is usually not public and lacks targeted collection and collection, it is difficult to obtain a sufficient number of image samples for deep learning training. In particular, the collection of various defect image sample sets taken at specific angles by line patrol robots is one of the difficulties in developing an automatic cable damage detection system. Summary of the invention:
[0005] In order to overcome the defects of the prior art, the purpose of the present invention is to propose a device that uses a small number of cable damage samples to build a virtual environment, quickly generate a large number of cable damage image data sets with different background images and different lighting conditions in a short time, and realize automatic calibration of cable damage targets, so as to provide enough image sample sets for the training of deep neural networks.
[0006] The present invention solves the technical problem by adopting the following technical solution:
[0007] A data set sample collection device for identifying cable damage based on a deep neural network, comprising:
[0008] A bracket and a damaged cable arranged horizontally with the bracket as support;
[0009] An image acquisition device is provided above the cable, and the image acquisition device has a camera that can move along the length direction of the cable;
[0010] It also includes an environment simulation device that can be used by the camera to capture simulated environmental scenes.
[0011] Preferably, further, the bracket comprises a base and a vertical rod fixed on the base, and a clamp is provided on the vertical rod for clamping the cable.
[0012] Preferably, further, the camera of the image acquisition device is arranged on a linear reciprocating mechanism of a lead screw slider, specifically comprising: a guide rail and a lead screw arranged parallel to the cable, and a slider cooperating with the lead screw, the camera is installed on the slider, the slider is driven by the lead screw to slide along the guide rail, the lead screw is driven by a stepper motor, and the base of the stepper motor is installed on the vertical rod.
[0013] Preferably, further, the environment simulation device comprises: a liquid crystal screen located below the cable capable of changing the playback image, and a lighting source capable of simulating natural light with adjustable color temperature and light softness.
[0014] Preferably, further, the LCD screen is arranged on an LCD screen bracket, both ends of the LCD screen bracket are hinged on columns of a sliding block, the sliding block is mounted on a slide rail, and the slide rail is parallel to the cable.
[0015] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0016] 1. The present invention proposes a data set sample acquisition device for identifying cable damage based on a deep neural network. It can quickly generate a large number of cable damage image sample sets that can be used for deep neural network training when there are only a small number of cable samples containing damage, thereby solving the current problem of a small number of cable damage samples.
[0017] 2. The environmental simulation device proposed in the present invention can simulate the real shooting environment, so that the cable damage in the sample set has rich background images and lighting conditions, thereby making the deep neural network detection more accurate, avoiding overfitting, and improving the generalization ability of the deep neural network, so that the automatic identification of damage during line inspection is more accurate and reliable. Description of the drawings:
[0018] Figure 1It is a schematic diagram of a data set sample collection device for identifying cable damage based on a deep neural network of the present invention;
[0019] Figure 2 This is a schematic diagram of the image acquisition device proposed in the present invention.
[0020] Numbers in the figure: 1 bracket, 2 image acquisition device, 3 environment simulation device, 4 fixture, 5 cable, 6 LCD screen, 7 lighting source, 8 slide rail, 9 computer, 10 cable damage, 11 camera, 12 bracket base, 13 vertical rod, 14 guide rail, 15 screw rod, 16 stepper motor.
[0021] The present invention will be further described below through specific implementation modes in conjunction with the accompanying drawings. Specific implementation method:
[0022] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0023] Embodiment: The data set sample collection device for identifying cable damage based on a deep neural network of the present invention comprises:
[0024] A bracket 1 and a cable 5 with a damage 10 arranged horizontally with the bracket as support; an image acquisition device 2 is arranged above the cable, and the image acquisition device has a camera 11 that can move along the length direction of the cable; and an environment simulation device 3 that can be captured by the camera to simulate the environment scene. The bracket 1 includes a base 12 and a vertical rod 13 fixed on the base 12, and a clamp 4 is arranged on the vertical rod for clamping the cable 5. The camera 11 of the image acquisition device 4 is arranged on a linear reciprocating mechanism of a lead screw slider, specifically including: a guide rail 14 arranged parallel to the cable, a lead screw 15, and a slider matched with the lead screw 15, the camera 11 is installed on the slider, the slider is driven by the lead screw to slide along the guide rail, the lead screw 15 is driven by a stepping motor 16, and the base of the stepping motor is installed on the vertical rod 13. The environment simulation device 3 includes: a liquid crystal screen 6 located below the cable that can change the playback picture and a lighting source 7 that can simulate natural light and has adjustable color temperature and light softness. The LCD screen 6 is arranged on the LCD screen bracket, and the two ends of the LCD screen bracket are hinged on the columns of the sliding block. The sliding block is mounted on the slide rail 8, and the slide rail 8 is parallel to the cable. In a specific implementation, the LCD screen 6 can adjust the angle and translate along the slide rail 8, and the illumination light source 7 simulates the ambient light by adjusting the color temperature, light softness and angle. The computer 9 controls the LCD screen 6 to quickly switch the background picture, and the image acquisition device 2 acquires the image of the cable damage 10 with different background pictures, and automatically marks the cable damage 10 in the image according to the position of the camera 11.
[0025] Working principle: When in use, prepare several sections of cables 5 with different types of damage, install the cables 5 with cable damage 10 on the fixture 4 in turn, adjust the lighting through the lighting source 7, the computer 9 controls the LCD screen 6 to quickly switch the background picture, and the camera 11 collects images of cable damage 10 with different background pictures, and automatically marks the cable damage 10 in the image. After the camera 11 completes the image collection at one position, it moves a step along the guide rail 14 and collects again, thereby collecting a large number of cable damage 10 image sample sets with different lighting conditions, different backgrounds, and different shooting angles for training deep neural networks.
[0026] It should be noted that some parts not described in detail in the present invention belong to the well-known technology in the art, or can be directly purchased from the market, and those skilled in the art can obtain them without creative work. The specific connection method is extremely widely used in this art or in daily life, and will not be described in detail here.
[0027] In addition, those skilled in the art may combine and combine the different embodiments or examples and the features of the different embodiments or examples described in this specification without contradiction. Although the embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and cannot be understood as limiting the present invention. Those skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.
Claims
1. A data set sample collection device for identifying cable damage based on deep neural network, Features include: A support (1) and a cable (5) with damage (10) arranged horizontally with the support as support; An image acquisition device (2) is provided above the cable, the image acquisition device having a camera (11) that can move along the length direction of the cable; It also includes an environment simulation device (3) capable of being captured by the camera to simulate the environment scene; The bracket (1) comprises a base (12) and a vertical rod (13) fixed on the base (12), and a clamp (4) is provided on the vertical rod for clamping the cable (5); The environment simulation device (3) comprises: a liquid crystal screen (6) located below the cable and capable of changing the playback image, and a lighting source (7) capable of simulating natural light and capable of adjusting the color temperature and light softness; The liquid crystal screen (6) is arranged on a liquid crystal screen bracket, the two ends of the liquid crystal screen bracket are hinged on the columns of the sliding block, the sliding block is mounted on a sliding rail (8), and the sliding rail (8) is parallel to the cable.
2. According to claim 1, a data set sample collection device for identifying cable damage based on a deep neural network, It is characterized in that The camera (11) of the image acquisition device (2) is arranged on a lead screw slider linear reciprocating mechanism, and specifically comprises: a guide rail (14) arranged parallel to the cable, a lead screw (15), and a slider cooperating with the lead screw (15), the camera (11) being mounted on the slider, the slider being driven by the lead screw to slide along the guide rail, the lead screw (15) being driven by a stepping motor (16), and the base of the stepping motor being mounted on the vertical rod (13).
Citation Information
Patent Citations
Movable substation and working method for unmanned aerial vehicle electric transmission line inspection
CN103941745A
Single image super-resolution reconstruction method base on block rotation and sharpness
CN106097253A
Movable type camera cradle head automatic-control track device
CN202746863U
Maintenance / Repair training system for electric cable stringing and facilities
JP2000050445A
