CN109492651B - Intelligent identification method for equipment signal lamp - Google Patents

Intelligent identification method for equipment signal lamp Download PDF

Info

Publication number
CN109492651B
CN109492651B CN201811297199.3A CN201811297199A CN109492651B CN 109492651 B CN109492651 B CN 109492651B CN 201811297199 A CN201811297199 A CN 201811297199A CN 109492651 B CN109492651 B CN 109492651B
Authority
CN
China
Prior art keywords
signal lamp
rgb
dbn
image
dbn 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.)
Active
Application number
CN201811297199.3A
Other languages
Chinese (zh)
Other versions
CN109492651A (en
Inventor
吴珺
刘明峰
李文坤
田小川
侯路
郭顺森
韩然
李祥新
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Qingdao Power Supply Co of State Grid Shandong Electric Power Co Ltd
Original Assignee
Qingdao Power Supply Co of State Grid Shandong Electric Power Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Qingdao Power Supply Co of State Grid Shandong Electric Power Co Ltd filed Critical Qingdao Power Supply Co of State Grid Shandong Electric Power Co Ltd
Priority to CN201811297199.3A priority Critical patent/CN109492651B/en
Publication of CN109492651A publication Critical patent/CN109492651A/en
Application granted granted Critical
Publication of CN109492651B publication Critical patent/CN109492651B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06V10/40—Extraction of image or video features
    • G06V10/56—Extraction of image or video features relating to colour
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/30—Noise filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses an intelligent identification method of equipment signal lamps, which comprises the following steps: s1: preprocessing the acquired signal lamp image; s2: performing RGB maximum ratio feature extraction on the preprocessed signal lamp image; s3: training the DBN network, and finely adjusting the DBN network; s4: and inputting the extracted RGB maximum ratio features into the trimmed DBN network for recognition. The method provided by the invention can effectively identify the signal lamp, and the comprehensive identification rate of the signal lamp image reaches 98%.

Description

Intelligent identification method for equipment signal lamp
Technical Field
The invention relates to the technical field of equipment signal lamp detection, in particular to an intelligent identification method of an equipment signal lamp.
Background
Along with the continuous deepening of the informatization degree of the electric power system, the types and the number of the machine room equipment are more and more, and the network topology structure is more and more complex. The staff is difficult to grasp the equipment running condition accurately, is unfavorable for grasping the whole running condition of the information system, is more unfavorable for unifying and optimizing the allocation of information resources and guaranteeing the reliable operation of the information system. Therefore, the operation state of each machine room device can be mastered timely and accurately, and the operation of each system in the electric power enterprise is very important to the efficient and safe operation of each system.
At present, one core task of monitoring the operation condition of equipment in a machine room is to accurately identify signal lamp state images of each system server captured by an industrial camera in time so as to judge whether the relevant equipment normally operates. At present, in the aspect of identifying the state of a traffic signal lamp, most of common algorithms use different direction, shape and color information thereof for detection and identification. However, unlike traffic lights, server lights have no shape features available and are smaller in size and more dense, so sensing their status information is more difficult. And at present, the research on the aspect at home and abroad is only reported.
Therefore, how to provide a method for intelligently identifying a device signal lamp is a problem that needs to be solved urgently by the technical personnel in the field.
Disclosure of Invention
In view of this, the invention provides an intelligent identification method for a device signal lamp, which extracts a signal lamp state image feature RGBMR, and after extracting the feature data, evaluates and identifies the signal lamp image by using a DBN network model, so as to grasp the operating state of a machine room server.
In order to achieve the purpose, the invention adopts the following technical scheme:
an intelligent identification method for equipment signal lamps comprises the following steps:
s1: preprocessing the acquired signal lamp image;
s2: performing RGB maximum ratio feature extraction on the preprocessed signal lamp image;
s3: training the DBN network, and finely adjusting the DBN network;
s4: and inputting the extracted RGB maximum ratio features into the trimmed DBN network for recognition to obtain a recognition result.
Preferably, step S1 specifically includes: and denoising the acquired signal lamp image.
Preferably, the specific algorithm flow of step S2 includes:
s21: extracting and segmenting each region of interest containing a signal lamp from the preprocessed signal lamp image by using a Kalman tracking algorithm, wherein each region of interest only contains one signal lamp;
s22: calculating the number of pixels in each region of interest, and acquiring RGB (red, green and blue) tricolor brightness values of each pixel;
s23: respectively sorting the RGB three-primary-color brightness values of each pixel from large to small, reading the first 10% of the RGB three-primary-color brightness values as RGB maximum value sequences, and respectively recording the RGB three-primary-color brightness values as a sequence RM10、GM10And BM10；
S24: respectively calculate RM10、GM10And BM10The maximum mean values obtained are respectively denoted as RMa、GMaAnd BMa；
S25: calculating the maximum average values R respectivelyMa、GMaAnd BMaThe relative ratio of (A) and (B) as the RGB maximum ratio features are respectively recorded as MRr、MRgAnd MRbWherein, in the step (A),
Figure BDA0001851472150000021
Figure BDA0001851472150000022
Figure BDA0001851472150000023
preferably, the method for fine-tuning the DBN network includes: and utilizing a BP algorithm to finely adjust the trained DBN network.
According to the technical scheme, compared with the prior art, the invention discloses an intelligent identification method of the equipment signal lamp, which is characterized in that signal lamp state image features RGBMR (RGB maximum ratio features) are extracted, and a DBN network model is used for evaluating and identifying the signal lamp image after the feature data are extracted. Through a large number of experimental analyses, compared with a common image HSV space characteristic and BPNN network identification method, the method provided by the invention can identify the state of the signal lamp more accurately, so that the running state of the server of the machine room can be mastered accurately.
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 embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
FIG. 1 is a flow chart of an intelligent identification method for a device signal lamp provided by the invention;
FIG. 2 is a schematic diagram of a DBN structure provided by the present invention;
FIG. 3 is a diagram of a signal lamp identification experiment system provided by the present invention;
FIG. 4 is a diagram of an image of a server state taken by an industrial camera in accordance with the present invention;
FIG. 5 is a schematic diagram of an isolated green, red, and yellow signal light image sample provided by the present invention;
FIG. 6 is a diagram of an RGBMR average distribution diagram of a green-red-yellow lamp image provided by the present invention;
FIG. 7 is a diagram illustrating the recognition result of the signal lamp recognition method according to the present invention;
fig. 8 is a drawing illustrating an identification result corresponding to extracted HSV features provided by the present invention.
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.
The embodiment of the invention discloses an intelligent identification method for equipment signal lamps, which can be used for more accurately identifying the state of the signal lamps so as to accurately master the running state of a machine room server.
First, a flow of extracting the RGBMR features of the image will be described.
The signal lamps of the server equipment in the machine room mainly have three states of green lamps, red lamps and yellow lamps, which respectively correspond to three conditions of normal operation, failure and waiting for inspection of the equipment, so that the main way of judging the operation condition of the equipment is to accurately identify the on and off of the signal lamps with the 3 colors. However, on one hand, the server signal lamp is small and emits light, which can cause great interference to the shot image; on the other hand, a plurality of signal lamps are often distributed in a shot image, and focusing is not available during shooting, so that the imaging color of the shot image is very sensitive to the change of external light, and the state information of the shot image is difficult to perceive. In view of the great defects of the existing method for identifying the signal lamp by utilizing the RGB characteristic value, HSV and HSI color space characteristic value and the color histogram, the invention provides a new signal lamp image color characteristic-RGB maximum ratio (RGBMR) characteristic. After the signal lamp image is denoised, the RGBMR characteristic extraction algorithm flow is as follows:
s21: extracting and segmenting each region of interest containing a signal lamp from the preprocessed signal lamp image by using a Kalman tracking algorithm, wherein each region of interest only contains one signal lamp;
s22: calculating the number of pixels in each region of interest, and acquiring RGB (red, green and blue) tricolor brightness values of each pixel;
s23: respectively sorting the RGB three-primary-color brightness values of each pixel from large to small, reading the first 10% of the RGB three-primary-color brightness values as RGB maximum value sequences, and respectively recording the RGB three-primary-color brightness values as a sequence RM10、GM10And BM10；
S24: respectively calculate RM10、GM10And BM10The maximum mean values obtained are respectively denoted as RMa、GMaAnd BMa；
S25: calculating the maximum average values R respectivelyMa、GMaAnd BMaThe relative ratio of (A) and (B) as the RGB maximum ratio features are respectively recorded as MRr、MRgAnd MRbWherein, in the step (A),
Figure BDA0001851472150000041
Figure BDA0001851472150000042
Figure BDA0001851472150000043
the DBN model used in the present invention is described in detail below.
The DBN is a deep learning model based on probability generation, can effectively capture important information from raw data through various nonlinear transformation and approximate complex nonlinear functions, and is suitable for classification and evaluation.
The DBN is constructed by stacking a series of constrained Boltzmann machines (RBMs) layer by layer, as shown in FIG. 2, layer 1 (input layer V) and layer 2 (hidden layer H)1) Form RBM1Layer 2 (hidden layer H)1) And layer 3 (hidden layer H)2) Form RBM2Go and go in this way, wherein hl，hkRespectively representing the l and k hidden units of the hidden layer in which it is located, w1、w2Representing the weight coefficients between layers. Each RBM is composed of an implicit layer and a visual layer, each layer is composed of binary random units, and the units are only connected with units in different layers and are not connected with the units in the same layer.
The energy of one connection of the visual unit and the hidden unit can be defined as:
Figure BDA0001851472150000051
where w represents the weight value between the visual layer and the hidden layer, and vectors a and b are the hidden units h, respectivelyjAnd a visual element viAnd θ ═ { w, b, a } represents the model parameters.
The number of input nodes, the number of hidden layers and the number of output nodes are the most important parameters of the DBN model. In the present invention, the architecture of the DBN model is defined as follows:
DBN[param1；param21,……,param2j；param3]
wherein param1 represents the number of input nodes, param2iRepresents the number of hidden nodes of the ith hidden layer, and param3 represents the number of output nodes.
The unique structure of the DBN enables it to be trained by training a series of RBMs using the Contrast Divergence (CD) algorithm. The primary training process can be summarized as: each RBM layer is trained using the activation probabilities of the subnet RBMs as input training data, and its output is used as input for the next RBM layer. After unsupervised pre-training, the first layer of RBMs is populated with raw input data and is used as a real-valued input GB-RBM (Gaussian-Bernoulli RBM, a model of RBMs), while the other layer is binary or Bernoulli-Bernoulli RBM (Bernoulli-Bernoulli limited Boltzmann, a model of RBMs). Finally, the update rule giving the parameters is as follows:
W←W+εw(<vihj>0-<vihj>1)
a←a+εa(<hj>0-<hj>1)
b←b+εb(<vi>0-<vi>1)
here epsilonw,εaAnd εbRespectively representing the learning rates of the weight w, the hidden layer bias a and the visible layer bias b.
Wherein, the activation probability refers to the conditional activation probability p (hv) of the hidden layer unit and the conditional activation probability p (v | h) of the visible layer unit, which represent the probability that the neural unit is activated and generates output, and they are respectively:
Figure BDA0001851472150000061
Figure BDA0001851472150000062
wherein sig (x) is 1/(1+ e)-x) Namely the sigmoid function.
For server signal lamp image recognition, after generative pre-training, the performance of the DBN is improved by combining other key steps capable of effectively fine-tuning weight values, such as discrimination, learning and the like. One proven very effective way to make a discriminant fine-tuning is to add a variable layer after the last RBM layer, with the parameters in this variable layer representing the expected tag values. Therefore, a BP algorithm similar to the standard Back Propagation Neural Network (BPNN) was introduced to adjust all DBN network weights.
A large amount of images which are acquired by an industrial camera when the industrial camera patrols a machine room and contain red, green and yellow state signal lamps are used as analysis objects. Considering that the green light is in a normal state and the number of the green light is the largest, the RGBMR characteristics for extracting a larger number of green light images form a training data set NLAnd carrying out unsupervised training on the initialized DBN model to obtain a standard DBN model. Inputting a large number of training data sets N consisting of green light images RGBMR characteristicsLIn the DBN model, after a large amount of data training, each parameter in the DBN model adapts to RGBMR characteristics of the blue light picture, and output values thereof are concentrated in a range. When a set test data set consisting of RGBMR characteristics of red and yellow light pictures is input, the output value of the DBN model deviates from the output value of the blue light picture to different degrees because the parameters of the DBN model are not trained by the data.
The specific structure and non-linear learning process of a DBN makes it very efficient to extract its essential features from a large amount of data. After a standard DBN model is obtained, a certain number of RGBMR characteristics of green, red and yellow signal lamp images are respectively extracted to form a test data set MLWill MLAnd inputting the data into a trained standard DBN model to perform evaluation and classification, so that the signal lamp state of the image corresponding to each group of data can be identified.
The identification method provided by the present invention is further described below with reference to specific experiments.
(1) Extracting RGBMR data sets
The signal lamp identification experimental system comprises a rail fixed at the top of a machine room, a rail car, a holder installed on the rail car, an industrial camera and the like, as shown in fig. 3.
In the experiment, send the automatic instruction of patrolling and examining to the cloud platform, it freely removes and oscilaltion along the track, drives the state image that industry camera shot each server signal lamp. In the experiment, the industrial camera captures 89 server images with different distances, light intensities and brightnesses as shown in fig. 4, and each image contains several to tens of signal lamps in different states. A Kalman tracking algorithm is adopted to extract an image area containing a signal lamp in an image and cut the image area separately, the colors, the brightness, the distance, the size and the like of the separated signal lamp images are different and accord with the actual routing inspection condition, and partial images are shown in figure 5.
Then, 200 green light images are randomly selected from the separated single signal light images to serve as training samples, and 100 green light images, red light images and yellow light images are selected to serve as testing samples. RGBMR characteristics of extracted training samples form a 200 x 3 training data set NLExtracting RGBMR characteristics of the test sample constitutes a 300 x 3 test data set ML. As can be seen from FIG. 6, the RGBMR values of the green light image are greatly different, and the MR thereof is largerr、MRgAnd MRbThe mean values are 0.311, 0.484 and 0.205, respectively; the difference of each value of the red light image is very small, and is about 0.33; MR of yellow light imager、MRgMean values are very close, all around 0.36, but their MRsbThe mean value is significantly smaller, 0.269. As can be seen, the RGBMR value distribution situations of the three types of images are greatly different, and the three types of images can be distinguished more easily by using the characteristic.
(2) DBN establishment and evaluation identification
In the invention, the number of input nodes of the DBN model corresponds to the dimension of the RGBMR data set, and the value is 3. Since the model is used for evaluation recognition of the image, the output node is set to 1. The ability of a DBN to obtain useful information from input data is determined by the number of hidden nodes, too few of which are generally not able to shape the data, and too many of which are hiddenThe nodes may then cause problems with overfitting and even eventually degradation of the evaluation performance. Therefore, N isLAnd MLRespectively used as a training data set and a testing data set to be input into the DBN model, and after a series of experimental research and analysis, the DBN model with stable, clear and reasonable evaluation results is obtained and is constructed as
DBN1[3；100,100,50,10；1]。
The number of DBN model input nodes corresponds here to the dimension of the RGBMR data set, which has a value of 3. Since the model is used for evaluation recognition of the image, the output node is set to 1. Usually, the number of nodes of the current hidden layer is not less than that of the next layer, and a large number of experimental analyses in the previous period are performed to obtain that the hidden layer is 4 layers, and the number of nodes of the first hidden layer is 100. Because no mature theory is used for deducing the number of the neural nodes of each layer of the DBN at present, the number of the nodes of the back hidden layer depends on the experimental effect in the experiment, and the optimal number of the nodes is selected after a plurality of attempts, namely 100, 50 and 10 respectively.
As a result of evaluation shown in fig. 7, three types of image samples of green, red, and yellow lights are clearly classified, the DBN evaluation value of the green light sample is concentrated around 0.8, the red light evaluation value is concentrated around 0.4, and the yellow light evaluation value is concentrated around 0.5. According to the DBN evaluation value, signal lamp color classification and identification are carried out on the image sample, and a judgment standard is given out through analysis
Figure BDA0001851472150000081
Therefore, as shown in fig. 7, with the small red circle, in 300 test samples, only 6 samples including two green lamps, one red lamp and three yellow lamps have a classification and identification error, and the other samples all obtain correct results.
The invention provides a good effect of signal lamp image identification by using RGBMR characteristics and a DBN network model, and comparison research experiments with HSV characteristics and a BPNN network are respectively carried out in order to further prove the superiority of the method.
1. Analysis by comparison with HSV characteristics
HSV (hue, saturation, and brightness) color space is suitable for the visual characteristics of humans and is widely used in the field of image classification and recognition. However, for different signal light images, the V values in their HSV space do not differ much, so only the mean and variance of the H, S features are extracted for the identification study. After a series of experimental comparisons, a DBN model suitable for the characteristics is constructed as
DBN2[4；100,100,50,50,10；1]
The recognition result is shown in fig. 8, the method can effectively recognize the green light image with the accuracy rate of about 93%, but the evaluation values of the yellow light and the red light are in the same interval of [0.9,1.2], which indicates that the method cannot effectively recognize the red light and the yellow light.
2. Comparison analysis with BPNN model
The BPNN is a neural network model with the characteristics of large-scale parallel operation, self-organization, self-learning and the like, theoretically, the BPNN can be used for fitting a nonlinear model with any precision, and has very successful application in the relevant field of pattern recognition. Unlike unsupervised training of the DBN model, the BPNN network requires supervised training, so more samples are required.
In the experiment, 600 signal lamp image samples are selected, wherein 200 signal lamp images are selected from green, red and yellow lamps respectively, and in order to obtain more convincing experiment results, the RGBMR characteristics provided by the invention are extracted, and the HSV characteristics are also extracted for comparison research. The first 100 sets of feature values of each state image sample are used for supervised training of the BPNN model, and the last 100 sets of features are used for testing. The model construction of the BPNN was analyzed and selected as [10,3], and the recognition results are shown in Table 1. Therefore, the method combining the RGBMR characteristics with the BPNN model has a slightly poor identification effect on yellow lamps, and has identification accuracy on green lamps and red lamps of more than 90%; the recognition rate of HSV characteristics combined with the BPNN model to green light is good, but the recognition rate to red light and yellow light is low, and the comprehensive recognition rate is only 81.3%; the RGBMR characteristic combined DBN method provided by the invention has the highest recognition rate, and the comprehensive recognition rate of signal lamp images reaches 98%.
TABLE 1 comparison of recognition rate results
Figure BDA0001851472150000091
The invention provides a new signal lamp state image feature, namely RGBMR, which is used for evaluating and identifying a signal lamp image by using a DBN network model after the feature data is extracted. Through a large amount of experimental analysis and comparative study with the common image HSV space characteristics and the BPNN network identification effect, the method provided by the invention is proved to be effectively applied to identifying the running state of the server of the machine room.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (3)

1. An intelligent identification method for equipment signal lamps is characterized by comprising the following steps:
s1: preprocessing the acquired signal lamp image;
s2: extracting the RGB maximum ratio characteristic of the signal lamp image after pretreatment;
the specific algorithm flow of step S2 includes:
s21: extracting and segmenting each region of interest containing a signal lamp from the preprocessed signal lamp image by using a Kalman tracking algorithm, wherein each region of interest only contains one signal lamp;
s22: calculating the number of pixels in each region of interest, and acquiring RGB (red, green and blue) tricolor brightness values of each pixel;
s23: respectively sorting the RGB three-primary-color brightness values of each pixel from large to small, reading the first 10% of the RGB three-primary-color brightness values as RGB maximum value sequences, and respectively recording the RGB three-primary-color brightness values as a sequence RM10、GM10And BM10；
S24: respectively calculate RM10、GM10And BM10The maximum mean values obtained are respectively denoted as RMa、GMaAnd BMa；
S25: calculating the maximum average values R respectivelyMa、GMaAnd BMaThe relative ratio of (A) and (B) as the RGB maximum ratio features are respectively recorded as MRr、MRgAnd MRbWherein, in the step (A),
Figure FDA0002369485680000011
Figure FDA0002369485680000012
Figure FDA0002369485680000013
s3: training the DBN network, and finely adjusting the DBN network;
s4: and inputting the extracted RGB maximum ratio features into the trimmed DBN network for recognition to obtain a recognition result.
2. The intelligent identification method for the equipment signal lamp according to claim 1, wherein the step S1 specifically includes: and denoising the acquired signal lamp image.
3. The intelligent identification method for the equipment signal lamp according to claim 1, wherein the method for fine-tuning the DBN network comprises the following steps: and utilizing a BP algorithm to finely adjust the trained DBN network.
CN201811297199.3A 2018-11-01 2018-11-01 Intelligent identification method for equipment signal lamp Active CN109492651B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811297199.3A CN109492651B (en) 2018-11-01 2018-11-01 Intelligent identification method for equipment signal lamp

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811297199.3A CN109492651B (en) 2018-11-01 2018-11-01 Intelligent identification method for equipment signal lamp

Publications (2)

Publication Number Publication Date
CN109492651A CN109492651A (en) 2019-03-19
CN109492651B true CN109492651B (en) 2020-05-15

Family

ID=65693506

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811297199.3A Active CN109492651B (en) 2018-11-01 2018-11-01 Intelligent identification method for equipment signal lamp

Country Status (1)

Country Link
CN (1) CN109492651B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112287853A (en) * 2020-11-02 2021-01-29 无锡超维智能科技有限公司 Dynamic intelligent image recognition algorithm based on IT equipment indicator light and equipment model
CN112818853B (en) * 2021-02-01 2022-07-19 中国第一汽车股份有限公司 Traffic element identification method, device, equipment and storage medium

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103235938A (en) * 2013-05-03 2013-08-07 北京国铁华晨通信信息技术有限公司 Method and system for detecting and identifying license plate

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103366190B (en) * 2013-07-26 2017-03-29 中国科学院自动化研究所 A kind of method of identification traffic signss
US11087228B2 (en) * 2015-08-12 2021-08-10 Bae Systems Information And Electronic Systems Integration Inc. Generic probabilistic approximate computational inference model for streaming data processing
CN108427958B (en) * 2018-02-02 2021-06-01 哈尔滨工程大学 Adaptive weight convolution neural network underwater sonar image classification method based on deep learning

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103235938A (en) * 2013-05-03 2013-08-07 北京国铁华晨通信信息技术有限公司 Method and system for detecting and identifying license plate

Also Published As

Publication number Publication date
CN109492651A (en) 2019-03-19

Similar Documents

Publication Publication Date Title
CN113486865B (en) Power transmission line suspended foreign object target detection method based on deep learning
CN113139521B (en) Pedestrian boundary crossing monitoring method for electric power monitoring
CN104992223A (en) Intensive population estimation method based on deep learning
CN111339883A (en) Method for identifying and detecting abnormal behaviors in transformer substation based on artificial intelligence in complex scene
CN104809463A (en) High-precision fire flame detection method based on dense-scale invariant feature transform dictionary learning
CN104598924A (en) Target matching detection method
CN105095870A (en) Pedestrian re-recognition method based on transfer learning
CN105279485B (en) The detection method of monitoring objective abnormal behaviour under laser night vision
CN108090472A (en) Pedestrian based on multichannel uniformity feature recognition methods and its system again
CN104616026B (en) A kind of monitoring scene type discrimination method towards intelligent video monitoring
CN110728216A (en) Unsupervised pedestrian re-identification method based on pedestrian attribute adaptive learning
CN109492651B (en) Intelligent identification method for equipment signal lamp
CN109344842A (en) A kind of pedestrian&#39;s recognition methods again based on semantic region expression
CN108345900B (en) Pedestrian re-identification method and system based on color texture distribution characteristics
CN116629465B (en) Smart power grids video monitoring and risk prediction response system
CN112464983A (en) Small sample learning method for apple tree leaf disease image classification
CN110618129A (en) Automatic power grid wire clamp detection and defect identification method and device
CN116824641B (en) Gesture classification method, device, equipment and computer storage medium
CN105426926A (en) Method and device for performing detection and classification on active-matrix organic light emitting diode (AMOLED)
CN115984546B (en) Sample base generation method for anomaly detection of fixed scene
Yeu et al. Investigation on different color spaces on faster RCNN for night-time human occupancy modelling
Rahimpour et al. Attention-based few-shot person re-identification using meta learning
CN115393802A (en) Railway scene unusual invasion target identification method based on small sample learning
CN106780471B (en) Transformer substation equipment infrared image change detection method based on Markov random field
CN114581769A (en) Method for identifying houses under construction based on unsupervised clustering

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
GR01 Patent grant
GR01 Patent grant