CN110705542A - Crane intrusion detection mechanism under power transmission scene based on HDNet - Google Patents
Crane intrusion detection mechanism under power transmission scene based on HDNet Download PDFInfo
- Publication number
- CN110705542A CN110705542A CN201910303273.6A CN201910303273A CN110705542A CN 110705542 A CN110705542 A CN 110705542A CN 201910303273 A CN201910303273 A CN 201910303273A CN 110705542 A CN110705542 A CN 110705542A
- Authority
- CN
- China
- Prior art keywords
- crane
- network
- hdnet
- power transmission
- image
- 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
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Closed-Circuit Television Systems (AREA)
Abstract
The invention provides a crane intrusion detection mechanism under an HDNet-based power transmission scene, which utilizes traditional image geometric transformation and GAN to perform generative data enhancement on a data set; designing a new target detection network HDNet, which comprises a crane candidate area generation sub-network HRDNet and a crane target classification sub-network HCNet; and the invalid nodes are deleted by adopting a channel pruning strategy, so that model compression is realized, the model compression can be stably operated on an embedded platform, the target detection efficiency is improved, and the availability and the robustness of crane intrusion operation detection in a power transmission scene are ensured.
Description
Technical Field
The invention relates to the field of deep learning target detection, in particular to a crane detection mechanism under a power transmission scene based on HDNet.
Background
A crane detection mechanism under a power transmission scene based on HDNet is based on a deep learning target detection method, a novel convolutional neural network is designed, and the neural network is compressed. The closest techniques to the present invention are:
(1) the YOLO algorithm: the YOLO algorithm is one of typical representatives of the deep learning target detection one-stage algorithm, and the real-time target detection effect of the efficient end-to-end deep learning system is achieved. The YOLO algorithm first scales the image to a uniform size and then feeds the image into a convolutional neural network, which segments the input image into a plurality of grid cells, each grid cell being responsible for detecting those targets whose center points fall within the grid. Each detected target comprises coordinates and confidence degree provided by variable borrowing, and a bounding box of a correct target is obtained through regression calculation by utilizing back propagation. However, the efficiency of YOLO is subject to loss of accuracy, and when there are small targets or dense targets in an image, the detection effect of YOLO is very undesirable.
(2) The fast R-CNN algorithm: in 2016, the fast R-CNN provided by the two-stage method integrates the structure of feature extraction, candidate region generation, boundary frame regression and classification into a network, so that the comprehensive performance is greatly improved, and the detection speed in the two-stage method is particularly obvious. However, the detection efficiency is still far from that of the one-stage method, and the detection accuracy is still to be further improved.
In order to make up for the defects of the traditional deep learning target detection method in speed and precision and the difficult problem that an overlarge deep learning model is difficult to operate on a lightweight embedded computing platform, the method makes full use of the advantages of deep learning in the field of target detection, and further improves the speed and efficiency of crane intrusion detection in a power transmission scene by the traditional target detection method.
Disclosure of Invention
In order to solve the defects and shortcomings in the prior art, the invention provides a crane intrusion real-time detection mechanism in a power transmission scene based on HDNet, which utilizes traditional image geometric transformation and GAN to perform generative data enhancement on a data set; designing a new target detection network HDNet, which comprises a crane candidate area generation sub-network HRDNet and a crane target classification sub-network HCNet; and the invalid nodes are deleted by adopting a channel pruning strategy, so that model compression is realized, the model compression can be stably operated on an embedded platform, the target detection efficiency is improved, and the availability and the robustness of crane intrusion operation detection in a power transmission scene are ensured. The technical scheme of the invention is as follows:
based on a traditional data enhancement method, obtaining a new image by adopting methods such as horizontal turning, rotation, random cutting and the like;
generating a crane image in a new power transmission scene by adopting DCGAN based on GAN, and performing generative data enhancement on a data set;
step (3), awakening the camera through a sensor, loading a neural network model, zooming the acquired image to a uniform size, and inputting the image into an HDNet;
step (4), the HDNet is divided into a precise crane candidate area extraction network HRDNet and a candidate area precise classification network HCNet;
step (5), in HRDNet, dividing the image into 9 x 9 grids, performing regression on the candidate region in each grid, and finally taking the candidate region with the confidence coefficient greater than a certain threshold value as a target region to be classified;
step (6), reclassifying the candidate regions in the HCNet to obtain a final detection result;
deleting redundant channel neurons by using a Frobenius norm large for convolution kernels on each channel level, realizing pruning of branches with small effect on a neural network, and obtaining a model with a smaller volume;
step (8), retraining the model obtained by pruning by using the training set so as to make up for the precision loss caused by pruning;
and (9) comprehensively analyzing the detection result and sending the detection result to a server terminal in a wireless manner.
The invention has the beneficial effects that:
(1) the method expands the data set by combining the traditional geometric transformation and the GAN, thereby solving the problem of insufficient training samples;
(2) a new scheme of target monitoring is provided, namely, the detection gravity center is generated in a more accurate candidate area, and a smaller classification network is used for judging again;
(3) and a channel pruning scheme is adopted for model compression aiming at the limitation of part of lightweight computing platforms, so that the calculated amount of target monitoring is greatly reduced.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a schematic diagram of a crane intrusion detection model in an HDNet-based power transmission scene according to the present invention; FIG. 2 is a diagram of a HDNet network architecture; FIG. 3 is a diagram of a network architecture for HRDNet; fig. 4 is a diagram showing the structure of an HCNet network.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1, a cache model diagram of a crane intrusion detection mechanism in an HDNet-based power transmission scenario combines a sensor and a model compression algorithm to reduce energy consumption so that the system can operate on a lightweight computing platform, and HDNet realizes high-precision and high-efficiency detection of a target. As shown in fig. 2, HDNet is divided into two parts, i.e., HRNet and HCNet, and the two sub-network structures are shown in fig. 3 and 4, and a crane target in an image is detected by processing the two sub-networks.
The following describes in detail a specific flow of a crane intrusion detection mechanism in an HDNet-based power transmission scenario:
based on a traditional data enhancement method, obtaining a new image by adopting methods such as horizontal turning, rotation, random cutting and the like;
generating a crane image in a new power transmission scene by adopting DCGAN based on GAN, and performing generative data enhancement on a data set;
step (3), awakening the camera through a sensor, loading a neural network model, zooming the acquired image to a uniform size, and inputting the image into an HDNet;
step (4), the HDNet is divided into a precise crane candidate area extraction network HRDNet and a candidate area precise classification network HCNet;
step (5), in HRDNet, dividing the image into 9 x 9 grids, performing regression on the candidate region in each grid, and finally taking the candidate region with the confidence coefficient greater than a certain threshold value as a target region to be classified;
step (6), reclassifying the candidate regions in the HCNet to obtain a final detection result;
deleting redundant channel neurons by using a Frobenius norm large for convolution kernels on each channel level, realizing pruning of branches with small effect on a neural network, and obtaining a model with a smaller volume;
step (8), retraining the model obtained by pruning by using the training set so as to make up for the precision loss caused by pruning;
and (9) comprehensively analyzing the detection result and sending the detection result to a server terminal in a wireless manner.
According to the crane intrusion detection mechanism based on the HDNet in the power transmission scene, the traditional image geometric transformation and the GAN are utilized to perform generative data enhancement on a data set. A crane detection network HDNet is designed, regions where a crane is likely to exist are generated by a crane candidate region sub-network HRNet, and reclassification is performed by HCNet. And deleting invalid nodes by adopting a channel pruning strategy, realizing model compression, and recovering the accuracy of the model through retraining again, so that the model can stably run on an embedded platform, the efficiency of target detection is improved, and the availability and the robustness of crane intrusion operation detection in a power transmission scene are ensured.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (1)
1. A crane intrusion detection mechanism based on an HDNet (high-density data network) transmission scene utilizes traditional image geometric transformation and GAN (generic object network) to perform generative data enhancement on a data set; designing a new target detection network HDNet, which comprises a crane candidate area generation sub-network HRDNet and a crane target classification sub-network HCNet; and the invalid nodes are deleted by adopting a channel pruning strategy, so that model compression is realized, the model compression can be stably operated on an embedded platform, the target detection efficiency is improved, and the availability and the robustness of crane intrusion operation detection in a power transmission scene are ensured. The method comprises the following steps:
based on a traditional data enhancement method, obtaining a new image by adopting methods such as horizontal turning, rotation, random cutting and the like;
generating a crane image in a new power transmission scene by adopting DCGAN based on GAN, and performing generative data enhancement on a data set;
step (3), awakening a camera through a sensor, loading a neural network model, zooming the acquired image to a uniform size, and inputting the image into an HDNet;
step (4), the HDNet is divided into a precise crane candidate area extraction network HRDNet and a candidate area precise classification network HCNet;
step (5), in HRDNet, dividing the image into 9 x 9 grids, performing regression on the candidate region in each grid, and finally taking the candidate region with the confidence coefficient greater than a certain threshold value as a target region to be classified;
step (6), reclassifying the candidate regions in the HCNet to obtain a final detection result;
deleting redundant channel neurons by using a Frobenius norm large for convolution kernels on each channel level, realizing pruning of branches with small effect on a neural network, and obtaining a model with a smaller volume;
step (8), retraining the model obtained by pruning by using the training set so as to make up for the precision loss caused by pruning;
and (9) comprehensively analyzing the detection result and sending the detection result to a server terminal in a wireless manner.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910303273.6A CN110705542A (en) | 2019-04-15 | 2019-04-15 | Crane intrusion detection mechanism under power transmission scene based on HDNet |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910303273.6A CN110705542A (en) | 2019-04-15 | 2019-04-15 | Crane intrusion detection mechanism under power transmission scene based on HDNet |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110705542A true CN110705542A (en) | 2020-01-17 |
Family
ID=69193025
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910303273.6A Pending CN110705542A (en) | 2019-04-15 | 2019-04-15 | Crane intrusion detection mechanism under power transmission scene based on HDNet |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110705542A (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101968832A (en) * | 2010-10-26 | 2011-02-09 | 东南大学 | Coal ash fusion temperature forecasting method based on construction-pruning mixed optimizing RBF (Radial Basis Function) network |
CN106886795A (en) * | 2017-02-17 | 2017-06-23 | 北京维弦科技有限责任公司 | Object identification method based on the obvious object in image |
CN108734225A (en) * | 2018-06-08 | 2018-11-02 | 浙江大学 | A kind of transmission line construction subject image detection method based on deep learning |
CN109063694A (en) * | 2018-09-12 | 2018-12-21 | 北京科技大学 | A kind of video object detection recognition method |
CN109472229A (en) * | 2018-10-30 | 2019-03-15 | 福州大学 | Shaft tower Bird's Nest detection method based on deep learning |
-
2019
- 2019-04-15 CN CN201910303273.6A patent/CN110705542A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101968832A (en) * | 2010-10-26 | 2011-02-09 | 东南大学 | Coal ash fusion temperature forecasting method based on construction-pruning mixed optimizing RBF (Radial Basis Function) network |
CN106886795A (en) * | 2017-02-17 | 2017-06-23 | 北京维弦科技有限责任公司 | Object identification method based on the obvious object in image |
CN108734225A (en) * | 2018-06-08 | 2018-11-02 | 浙江大学 | A kind of transmission line construction subject image detection method based on deep learning |
CN109063694A (en) * | 2018-09-12 | 2018-12-21 | 北京科技大学 | A kind of video object detection recognition method |
CN109472229A (en) * | 2018-10-30 | 2019-03-15 | 福州大学 | Shaft tower Bird's Nest detection method based on deep learning |
Non-Patent Citations (4)
Title |
---|
JOSEPH REDMON等: ""YOLO9000:Better, Faster, Stronger"", 《ARXIV》 * |
JOSEPH REDMON等: ""You Only Look Once:Unified, Real-Time Object Detection"", 《ARXIV》 * |
SHAOQING REN等: ""Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks"", 《ARXIV》 * |
YIHUI HE等: ""Channel Pruning for Accelerating Very Deep Neural Networks"", 《ARXIV》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111160108B (en) | Anchor-free face detection method and system | |
CN108846826B (en) | Object detection method, object detection device, image processing apparatus, and storage medium | |
CN110991311B (en) | Target detection method based on dense connection deep network | |
Liu et al. | Research on automatic garbage detection system based on deep learning and narrowband internet of things | |
CN109214353B (en) | Training method and device for rapid detection of face image based on pruning model | |
CN110322453A (en) | 3D point cloud semantic segmentation method based on position attention and auxiliary network | |
CN111753682B (en) | Hoisting area dynamic monitoring method based on target detection algorithm | |
CN109242019B (en) | Rapid detection and tracking method for optical small target on water surface | |
CN110717496B (en) | Complex scene tree detection method based on neural network | |
CN112528845B (en) | Physical circuit diagram identification method based on deep learning and application thereof | |
CN113222210B (en) | Typhoon short-term prediction method and device based on multi-mode generation confrontation | |
CN114972312A (en) | Improved insulator defect detection method based on YOLOv4-Tiny | |
CN102903121A (en) | Fusion algorithm based on moving target tracking | |
CN114548606B (en) | Construction method of cyclone strength prediction model and cyclone strength prediction method | |
CN116012879A (en) | Pedestrian detection method, system, equipment and medium for improving YOLOv4 network | |
Yang et al. | Research on subway pedestrian detection algorithms based on SSD model | |
Huang et al. | Mobile phone component object detection algorithm based on improved SSD | |
CN113902792A (en) | Building height detection method and system based on improved RetinaNet network and electronic equipment | |
Shi et al. | A new multiface target detection algorithm for students in class based on bayesian optimized YOLOv3 model | |
CN111931551B (en) | Face detection method based on lightweight cascade network | |
Ouyang et al. | Aerial target detection based on the improved YOLOv3 algorithm | |
Chen et al. | [Retracted] Surface Defect Detection Method Based on Improved Attention Mechanism and Feature Fusion Model | |
Xie et al. | CSPPartial-YOLO: A lightweight YOLO-based method for typical objects detection in remote sensing images | |
CN110705542A (en) | Crane intrusion detection mechanism under power transmission scene based on HDNet | |
CN116884211A (en) | Traffic sequence prediction method and device, electronic equipment and storage medium |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20200117 |