WO2019105131A1 - 用于监控的图像识别方法及系统、计算设备以及可读存储介质 - Google Patents

用于监控的图像识别方法及系统、计算设备以及可读存储介质 Download PDF

Info

Publication number
WO2019105131A1
WO2019105131A1 PCT/CN2018/109771 CN2018109771W WO2019105131A1 WO 2019105131 A1 WO2019105131 A1 WO 2019105131A1 CN 2018109771 W CN2018109771 W CN 2018109771W WO 2019105131 A1 WO2019105131 A1 WO 2019105131A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
determined
recognition
convolutional neural
neural network
Prior art date
Application number
PCT/CN2018/109771
Other languages
English (en)
French (fr)
Inventor
陈年春
Original Assignee
深圳中兴网信科技有限公司
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 深圳中兴网信科技有限公司 filed Critical 深圳中兴网信科技有限公司
Publication of WO2019105131A1 publication Critical patent/WO2019105131A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • 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
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N7/00—Television systems
    • H04N7/18—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast

Definitions

  • the present application relates to the field of image recognition, for example, to an image recognition method and system for monitoring, a computer device, and a computer readable storage medium.
  • intelligent monitoring and identification in the field of environmental protection is the main direction of future development, such as real-time monitoring of the quality of the weather, monitoring the size of pollutants in the city, and monitoring the pollution sources of the city to complete the monitoring of air.
  • the coal monitoring in the monitoring city's pollution sources has been in the artificial timed inspection and monitoring. Every once in a while, the environmental protection department has appointed a special staff to go to the site for inspection, and the captured violations are kept in the camera. And uploaded to the environmental business system. In this process, the staff needs to go to the coal yard several times to obtain the violations on the site, and the utilization rate of the personnel is very low.
  • the present application is intended to address at least one of the technical problems existing in the related art.
  • the present application proposes an image recognition method for monitoring.
  • a computer device is presented in the present application.
  • the present application proposes a computer readable storage medium.
  • the present application provides an image recognition method for monitoring, comprising: acquiring an image to be determined; comparing the image to be determined with the recognition model, identifying whether the image to be determined includes an object feature, obtaining a recognition result; and transmitting the recognition result.
  • the image recognition method for monitoring obtained by the present application obtains an image to be judged and compares the image to be determined with the recognition model, thereby identifying whether the image of the object to be judged includes the feature of the object, and transmitting the recognition result to the environmental protection service system.
  • the environmental protection department can obtain the on-site image of the coal yard in real time, and obtain the violation status of the coal yard through the identification model. It is not necessary to assign a special person to monitor the coal yard, avoiding the repeated round-trip shooting of the staff, and eliminating the use of the shooting interval to drill the hole. The effective coverage of the dust-proof net of the coal yard is ensured, thereby reducing the pollution of the coal yard dust to the air.
  • the method before acquiring the image to be determined, the method further includes: establishing a recognition model, and the establishing the recognition model comprises: receiving a basic image, where the basic image includes: a first image and a second image, the first image includes the object feature, and the second image
  • the image contains non-object features
  • the first layer of convolutional neural network and the second layer of convolutional neural network are built using convolutional neural network technology of the tensorflow framework (a machine learning open source framework introduced by Google); and according to the first layer Convolutional neural network and second layer convolutional neural network, establish a fully connected network to obtain a model structure; feature extraction of the base image using a convolutional neural network algorithm of the tensorflow framework to extract object features of the first image And extracting the non-object features of the second image, and storing the object features and the non-object features into the model structure to obtain a recognition model.
  • the establishing the recognition model comprises: receiving a basic image, where the basic image includes: a first image and a second
  • a model structure is established using a convolutional neural network technique of the tensorflow framework, and a first image having an object feature and a second image having a non-object feature are added as a base image to the model structure, wherein the tensorfiow frame is Google
  • the company develops a tensor-based data flow graph calculation framework based on DistBelief, a machine learning open source framework from Google.
  • the convolutional neural network algorithm of the tensorflow framework is used to extract the object features and non-object features, and is added to the recognition model, so that the recognition model can learn how to distinguish whether the image has objects according to the added object features and non-object features, and the recognition speed is fast.
  • the accuracy is high, and the whole process does not require personnel participation; the more the number of basic images added in the model structure, the more accurate the recognition result is, that is, the higher the accuracy of the recognition model, and the use of the recognition model to get rid of the manual image recognition Reduce the workload of the environmental protection business for the image, improve the work efficiency, and ensure the effective monitoring of the coal yard, effectively prevent the coal yard from using the picture shooting cycle to drill the air to cause pollution.
  • the image to be determined is stored to the recognition model.
  • the recognition model completes the recognition of the image to be determined
  • the image to be determined is stored in the recognition model, so that the number of learning objects for the feature of the object in the recognition model is increased, and when the next image is judged
  • the recognition model recognizes the result more accurately, and the possibility of error is lower, and as the number of images to be judged increases, the recognition model is continuously improved, and the recognition result is more reliable.
  • storing the image to be determined to the recognition model comprises: performing feature extraction on the determined image using a convolutional neural network algorithm of the tensorflow framework, and storing the feature of the image to be determined into the recognition model.
  • the convolutional neural network algorithm using the tensorflow framework performs feature extraction on the image to be judged, that is, the same algorithm as the feature extraction of the base image is used to ensure the feature extracted from the image to be determined and the object extracted from the base image. Consistency of features and non-object features, avoiding the inconsistency of learning content in the recognition model due to different extraction algorithms, resulting in inaccurate recognition of the recognition model. Convolutional neural network algorithm using tensorflow framework for feature extraction And storing the feature of the image to be judged in the recognition model, so that the recognition model performs feature learning, thereby ensuring the accuracy of the recognition model for the next image determination, thereby improving reliability.
  • the method before the feature extraction is performed on the image to be judged by the convolutional neural network algorithm using the tensorflow framework, and the feature of the image to be determined is stored in the recognition model, the method further comprises: classifying the image to be determined according to the recognition result, Obtaining a classification category; transmitting an image to be judged and a classification category, and receiving an audit result of the image to be determined and the classification category, the audit result indicating whether the image to be judged is consistent with the classification category; when the audit result indicates the image to be judged and the classification When the categories are consistent, the convolutional neural network algorithm using the tensorflow framework performs feature extraction on the determined image, and stores the feature of the image to be determined into the recognition model; when the audit result indicates the image to be determined and the classification category Storage does not match when it does not match.
  • the image to be judged is further classified according to the recognition result, the classification category is obtained, and the image to be determined and the corresponding classification category are sent out, so that the auditor can
  • the classification category and the corresponding image to be judged are reviewed to determine whether the classification category is correct and the corresponding image to be judged meets the requirements.
  • the convolutional neural network algorithm using the tensorflow framework is used to characterize the image.
  • the content of the model learning is reversed, resulting in the logic of the recognition model after learning is unclear, and the occurrence of errors and the like is recognized.
  • the present application provides an image recognition system for monitoring.
  • the image recognition system for monitoring includes: an acquisition unit configured to acquire an image to be determined; and an identification unit configured to compare the image to be determined with the recognition model, and identify Determining whether the image contains the object feature, and obtaining the recognition result; and the sending unit is configured to send the recognition result.
  • the image recognition system for monitoring provides the image to be determined by the acquisition unit, and the recognition unit compares the image to be determined with the recognition model, thereby identifying whether the image of the object to be determined includes the feature of the object, and the sending unit sends the recognition result to the
  • the environmental protection department can obtain the on-site image of the coal yard in real time, and obtain the violation status of the coal yard in real time through the identification model. It is not necessary to assign a special person to monitor the coal yard, avoiding the repeated round-trip shooting of the staff and eliminating the coal yard. The use of the shooting interval to drill the voids ensures the effective coverage of the dust-proof net of the coal yard, thereby reducing the pollution of the coal yard dust to the air.
  • the image recognition system for monitoring further includes: an establishing unit configured to establish a recognition model; the establishing unit comprising: a first receiving unit, a modeling unit, and a storage unit.
  • the first receiving unit is configured to receive a base image, the base image comprising: a first image and a second image, the first image comprising an object feature, the second image comprising a non-object feature;
  • the modeling unit being configured as a convolutional neural network using a tensorflow frame
  • the technology establishes a first layer of convolutional neural network and a second layer of convolutional neural network, and establishes a fully connected network according to the first layer of convolutional neural network and the second layer of convolutional neural network to obtain a model structure;
  • the storage unit is set to use a convolutional neural network algorithm of the tensorflow framework performs feature extraction on the base image to extract an object feature of the first image and a non-object feature of the second image, and the object feature and the non-object feature Stored in
  • a model structure is established using a convolutional neural network technique of the tensorflow framework, and a first image having an object feature and a second image having a non-object feature are added as a base image to the model structure, wherein the tensorfiow frame is Google developed a framework based on data flow graph calculation based on DistBelief.
  • the convolutional neural network algorithm of the tensorflow framework is used to extract the object features and non-object features, and is added to the recognition model, so that the recognition model can learn how to distinguish whether the image has objects according to the added object features and non-object features, and the recognition speed is fast.
  • the accuracy is high, and the whole process does not require personnel participation; the more the number of basic images added in the model structure, the more accurate the recognition result is, that is, the higher the accuracy of the recognition model, and the use of the recognition model to get rid of the manual image recognition Reduce the workload of the environmental protection business department for the image, improve the work efficiency, and ensure the effective monitoring of the coal yard, effectively prevent the coal yard from using the picture shooting cycle to drill the air to cause pollution.
  • the image recognition system for monitoring further comprises: an update unit configured to store the image to be determined to the recognition model.
  • the recognition model completes the recognition of the image to be determined
  • the image to be determined is stored in the recognition model, so that the number of learning objects for the feature of the object in the recognition model is increased, and when the next image is judged
  • the recognition model recognizes the result more accurately, and the possibility of error is lower, and as the number of images to be judged increases, the recognition model is continuously improved, and the recognition result is more reliable.
  • the updating unit comprises: an updating subunit, configured to perform feature extraction on the determined image using a convolutional neural network algorithm of the tensorflow framework, and store the feature of the image to be determined into the recognition model.
  • the convolutional neural network algorithm using the tensorflow framework performs feature extraction on the image to be judged, that is, the same algorithm as the feature extraction of the base image is used to ensure the feature extracted from the image to be determined and the object extracted from the base image. Consistency of features and non-object features, avoiding the inconsistency of learning content in the recognition model due to different extraction algorithms, resulting in inaccurate recognition of the recognition model. Convolutional neural network algorithm using tensorflow framework for feature extraction And storing the feature of the image to be judged in the recognition model, so that the recognition model performs feature learning, thereby ensuring the accuracy of the recognition model for the next image determination, thereby improving reliability.
  • the updating unit further includes: a classifying unit configured to classify the image to be determined according to the recognition result to obtain a classification category; send the image to be determined and the classification category, and receive the Determining the judgment image and the audit result of the classification category, the audit result indicating whether the image to be judged is consistent with the classification category; the update subunit is further configured to: when the audit result indicates the image to be determined and the classification When the categories are consistent, the feature of the image to be determined is extracted by using a convolutional neural network algorithm of the tensorflow framework, and the feature of the image to be determined is stored into the recognition model; when the review result indicates that the image is to be determined When the image does not match the classification category, the storage fails.
  • a classifying unit configured to classify the image to be determined according to the recognition result to obtain a classification category
  • send the image to be determined and the classification category and receive the Determining the judgment image and the audit result of the classification category, the audit result indicating whether the image to be judged is consistent with the classification category
  • the classification category of the image to be determined is determined according to the recognition result, and the image to be determined and the corresponding classification category are sent out, so that the auditor can classify the classification. And corresponding to the image to be judged for review, determining whether the classification category is correct and the corresponding image to be judged meets the requirements, and when the image to be determined is consistent with the classification category, the convolution neural network algorithm using the tensorflow framework is used to extract the feature image. Otherwise, without storing, by classifying the image to be judged and reviewing whether the classification is correct, the classification correctness of the image to be judged added to the recognition model is ensured, and the recognition of the model learning content is reversed when the classification error occurs. The logic of the recognition model after learning is unclear, and the occurrence of an error or the like is recognized.
  • the present application provides a computer device including a memory, a processor, and a computer program stored on the memory and operable on the processor, and the image recognition method for monitoring according to any of the above is implemented when the processor executes the computer program .
  • the computer device when the processor executes the computer program, realizes by acquiring the image to be determined, and comparing the image to be determined with the recognition model, thereby identifying whether the image to be judged includes the object feature, and transmitting the recognition result to the environmental protection service.
  • the environmental protection department can obtain the on-site image of the coal yard in real time, and obtain the violation status of the coal yard through the identification model. It is not necessary to assign a special person to monitor the coal yard, which avoids the repeated round-trip shooting of the staff and eliminates the use of the shooting interval of the coal yard. Drilling the air to ensure the effective coverage of the dust-proof net of the coal yard, thereby reducing the pollution of the coal yard dust to the air.
  • the present application proposes a computer readable storage medium having stored thereon a computer program that, when executed by a processor, implements an image recognition method for monitoring as described above.
  • the computer-readable storage medium provided by the present application when executed by a processor, implements acquiring an image to be determined, and compares the image to be determined with the recognition model, thereby identifying whether the image to be determined includes an object feature, and transmitting the recognition result
  • the environmental protection department can obtain the on-site image of the coal yard in real time, and obtain the violation status of the coal yard through the identification model. It is not necessary to assign a special person to monitor the coal yard, avoiding the repeated round-trip shooting of the staff and eliminating the coal yard.
  • the use of the shooting interval to drill the voids ensures the effective coverage of the dust-proof net of the coal yard, thereby reducing the pollution of the coal yard dust to the air.
  • FIG. 1 is a flow chart showing an image recognition method for monitoring according to an embodiment of the present application
  • FIG. 2 is a flow chart showing an image recognition method for monitoring according to an embodiment of the present application
  • FIG. 3 is a flow chart showing an image recognition method for monitoring according to an embodiment of the present application.
  • FIG. 4 is a flow chart showing an image recognition method for monitoring according to an embodiment of the present application.
  • FIG. 5 is a flow chart showing an image recognition method for monitoring according to an embodiment of the present application.
  • Figure 6 shows a schematic block diagram of an image recognition system for monitoring of one embodiment of the present application
  • Figure 7 shows a schematic block diagram of a computer device of one embodiment of the present application.
  • FIG. 1 is a flow chart showing an image recognition method for monitoring according to an embodiment of the present application.
  • the method includes the following steps 102 to 106.
  • step 102 an image to be determined is acquired.
  • step 104 the image to be determined is compared with the recognition model, and it is identified whether the image to be determined contains the feature of the object, and the recognition result is obtained.
  • step 106 the recognition result is transmitted.
  • the environmental protection department can obtain the image to be determined and compare the image to be determined with the recognition model to identify whether the image is included in the image to be determined, and send the recognition result to the environmental protection service system, and the environmental protection department can obtain the real-time information.
  • the scene image of the coal yard, and the violation of the coal yard by the identification model without the need to assign a special person to monitor the coal yard, avoiding the staff to repeat the round-trip shooting image, to prevent the coal yard from using the shooting interval to drill holes, to ensure the prevention of the coal yard
  • the effective coverage of the dust net reduces the pollution of the coal yard dust to the air.
  • FIG. 2 is a schematic flowchart of an image recognition method for monitoring according to an embodiment of the present application. The method includes the following steps 202 to 212.
  • a base image is received, the base image comprising: a first image and a second image, the first image comprising an object feature and the second image comprising a non-object feature.
  • a first layer of convolutional neural network and a second layer of convolutional neural network are established using a convolutional neural network technique of the tensorflow framework; and a first layer of convolutional neural network and a second layer of convolutional neural network are established. Fully connected to the network to get the model structure.
  • step 206 feature extraction is performed on the base image using a convolutional neural network algorithm of the tensorflow framework to extract the object features of the first image and the non-object features of the second image, and store the object features and the non-object features to the model structure.
  • the recognition model is obtained.
  • step 208 an image to be determined is acquired.
  • step 210 the image to be determined is compared with the recognition model, and it is identified whether the image to be determined contains the feature of the object, and the recognition result is obtained.
  • step 212 the recognition result is transmitted.
  • a model structure is established using a convolutional neural network technique of the tensorflow framework, and a first image having object features and a second image having non-object features are added as a base image to the model structure, wherein the tensorflow frame is Google developed a framework based on data flow graph calculation based on DistBelief.
  • the convolutional neural network algorithm of the tensorflow framework is used to extract the object features and non-object features, and is added to the recognition model, so that the recognition model can learn how to distinguish whether the image has objects according to the added object features and non-object features, and the recognition speed is fast.
  • the accuracy is high, and the whole process does not require personnel participation; the more the number of basic images added in the model structure, the more accurate the recognition result is, that is, the higher the accuracy of the recognition model, for example, the number of basic images is 10000 and 500 respectively.
  • Zhang Shi the recognition model gives higher reliability to the number of basic pictures in the recognition result given by the image to be judged.
  • the recognition model By using the recognition model, the image recognition is eliminated by manual use, and the workload of the environmental protection business department for the image is reduced. Work efficiency, and ensure the effective monitoring of the coal yard, effectively prevent the coal yard from using the picture shooting cycle to drill holes to pollute the air.
  • FIG. 3 is a schematic flowchart of an image recognition method for monitoring according to an embodiment of the present application. The method includes the following steps 302 to 314.
  • a base image is received, the base image comprising: a first image and a second image, the first image comprising an object feature and the second image comprising a non-object feature.
  • a first layer convolutional neural network and a second layer convolutional neural network are established using a tensorflow framework convolutional neural network technique; and a first layer convolutional neural network and a second layer convolutional neural network are established. Fully connected to the network to get the model structure.
  • step 306 feature extraction is performed on the base image using a convolutional neural network algorithm of the tensorflow framework to extract the object features of the first image and the non-object features of the second image, and store the object features and the non-object features to the model structure.
  • the recognition model is obtained.
  • step 308 an image to be determined is acquired.
  • step 310 the image to be determined is compared with the recognition model, and it is identified whether the image to be determined contains the feature of the object, and the recognition result is obtained.
  • step 312 the recognition result is transmitted.
  • step 314 the image to be determined is stored to the recognition model.
  • the recognition model completes the recognition of the image to be determined
  • the image to be determined is stored in the recognition model, so that the number of learning objects for the feature of the object in the recognition model is increased, and when the next image is judged
  • the recognition model recognizes the result more accurately, and the possibility of error is lower, and as the number of images to be judged increases, the recognition model is continuously improved, and the recognition result is more reliable.
  • the identification model has a limited number of basic images at the beginning of the establishment.
  • the image to be judged is added as a base image to the recognition model, so that the number of learning samples of the recognition model is continuously increased, and the recognition model is The accuracy of image recognition to be judged is increased, making the recognition result more reliable.
  • FIG. 4 is a schematic flowchart of an image recognition method for monitoring according to an embodiment of the present application. The method includes the following steps 402 to 414.
  • a base image is received, the base image comprising: a first image and a second image, the first image comprising an object feature and the second image comprising a non-object feature.
  • a first layer convolutional neural network and a second layer convolutional neural network are established using a tensorflow framework convolutional neural network technique; and a first layer convolutional neural network and a second layer convolutional neural network are established. Fully connected to the network to get the model structure.
  • step 406 feature extraction is performed on the base image using a convolutional neural network algorithm of the tensorflow framework to extract the object features of the first image and the non-object features of the second image, and store the object features and non-object features to the model structure.
  • the recognition model is obtained.
  • step 408 an image to be determined is acquired.
  • step 410 the image to be determined is compared with the recognition model, and it is identified whether the image to be determined contains the feature of the object, and the recognition result is obtained.
  • step 412 the recognition result is transmitted.
  • step 414 the feature image is extracted using the convolutional neural network algorithm of the tensorflow framework, and the features of the image to be determined are stored into the recognition model.
  • the convolutional neural network algorithm using the tensorflow framework performs feature extraction on the image to be judged, that is, the same algorithm as the feature extraction of the base image is used to ensure the feature extracted from the image to be determined and the object extracted from the base image. Consistency of features and non-object features, avoiding the inconsistency of learning content in the recognition model due to different extraction algorithms, resulting in inaccurate recognition of the recognition model. Convolutional neural network algorithm using tensorflow framework for feature extraction And storing the feature of the image to be judged in the recognition model, so that the recognition model performs feature learning, thereby ensuring the accuracy of the recognition model for the next image determination, thereby improving reliability.
  • FIG. 5 is a schematic flowchart of an image recognition method for monitoring according to an embodiment of the present application. The method includes the following steps 502 to 516.
  • a base image is received, the base image comprising: a first image and a second image, the first image comprising an object feature and the second image comprising a non-object feature.
  • a first layer convolutional neural network and a second layer convolutional neural network are established using a tensorflow framework convolutional neural network technique; and a first layer convolutional neural network and a second layer convolutional neural network are established. Fully connected to the network to get the model structure.
  • step 506 feature extraction is performed on the base image using a convolutional neural network algorithm of the tensorflow framework to extract the object features of the first image and the non-object features of the second image, and store the object features and the non-object features to the model structure.
  • the recognition model is obtained.
  • step 508 an image to be determined is acquired.
  • step 510 the image to be determined is compared with the recognition model, and it is identified whether the image to be determined contains the feature of the object, and the recognition result is obtained.
  • step 512 the recognition result is transmitted.
  • step 514 according to the recognition result, determining a classification category of the image to be determined; transmitting the image to be determined and the classification category, and receiving an audit result of the image to be determined and the classification category, the audit result indicating the image to be judged and the classification category Is it consistent?
  • step 516 when the audit result indicates that the image to be determined is consistent with the classification category, feature extraction is performed on the image to be determined using a convolutional neural network algorithm of the tensorflow framework, and the image to be determined is The feature is stored in the recognition model; when the audit result indicates that the image to be determined is inconsistent with the classification category, the storage fails.
  • the image to be judged is further classified according to the recognition result, the classification category is obtained, and the image to be determined and the corresponding classification category are sent out, so that the auditor can
  • the classification category and the corresponding image to be judged are reviewed to determine whether the classification category is correct and whether the corresponding image to be judged meets the requirements.
  • the convolutional neural network algorithm using the tensorflow framework is used to characterize the image.
  • the learning model after learning is unclear and identifies the occurrence of an error.
  • the recognition model identifies the image to be judged as having a dust-proof net, and stores the image to be judged in the recognition model. Since the wrong classification causes the recognition model to treat the image in the next image recognition.
  • the recognition is performed, an erroneous recognition result is given, and the image to be judged is classified according to the recognition result, and sent to the auditor for review, and the image to be judged is stored in the recognition model only when the classification category is consistent with the image to be determined. To avoid identifying the wrong learning of the model.
  • FIG. 6 shows a schematic block diagram of an image recognition system for monitoring of one embodiment of the present application.
  • the image recognition system 600 for monitoring includes an acquisition unit 602, an identification unit 604, and a transmission unit 606.
  • the obtaining unit 602 is configured to acquire an image to be determined; the identifying unit 604 is configured to compare the image to be determined with the recognition model, identify whether the image to be determined includes the feature of the object, and obtain a recognition result; and the sending unit 606 is configured to send the recognition result.
  • the image to be determined is acquired by the obtaining unit 602, and the identifying unit 604 compares the image to be determined with the recognition model to identify whether the object feature is included in the image to be determined, and the sending unit 606 sends the recognition result to the environmental protection service system.
  • the environmental protection department can obtain the on-site image of the coal yard in real time, and obtain the violation status of the coal yard through the identification model. It is not necessary to assign a special person to monitor the coal yard, avoiding the repeated round-trip shooting of the staff, and eliminating the use of the shooting interval to drill the space. It ensures the effective coverage of the dust-proof net of the coal yard, thus reducing the pollution of the coal yard dust to the air.
  • the image recognition system for monitoring further comprises: an establishing unit 608 configured to establish a recognition model.
  • the establishing unit 608 includes a first receiving unit 610, a modeling unit 612, and a storage unit 614.
  • the first receiving unit 610 is configured to receive a base image, the base image comprising: a first image and a second image, the first image comprising an object feature and the second image comprising a non-object feature.
  • the modeling unit 612 is configured to establish a first layer convolutional neural network and a second layer convolutional neural network using a convolutional neural network technique of the tensorflow framework; and according to the first layer convolutional neural network and the second layer convolutional neural network, Establish a fully connected network and get the model structure.
  • the storage unit 614 is configured to perform feature extraction on the base image using a convolutional neural network algorithm of the tensorflow framework to extract the object features of the first image and the non-object features of the second image, and store the object features and the non-object features to the model structure. In the middle, the recognition model is obtained.
  • modeling unit 612 builds a model structure using convolutional neural network techniques of the tensorflow framework, and storage unit 614 adds a first image having object features and a second image having non-object features as a base image to the model structure.
  • the tensorfiow framework is a framework developed by Google based on DistBelief for tensor data flow graph calculation, and DisBelief is a manual deep learning system launched by Google in 2011.
  • Using the tensorflow neural network algorithm of the tensorflow framework to extract features from the base image to extract object features and non-object features, and add object features and non-object features to the recognition model, so that the recognition model can be based on the added object features and non-objects.
  • the recognition model gives a higher reliability for the number of basic pictures in the recognition result given by the image to be judged, and is freed by manual use by using the recognition model.
  • Image recognition reduce the workload of the environmental protection business for the image, improve the work efficiency, and ensure the effective monitoring of the coal yard, effectively prevent the coal yard from using the picture shooting cycle to drill the air to cause pollution.
  • the image recognition system 600 for monitoring further includes an update unit 616 configured to store the image to be determined to the recognition model.
  • the update unit 616 stores the image to be determined into the recognition model, so that the number of learning objects for the feature of the object in the recognition model is increased, and the recognition unit 604 is in the opposite direction.
  • the recognition result of the recognition model is more accurate, and the possibility of error is lower, and as the number of images to be judged increases, the recognition model is continuously improved, and the recognition result is more reliable.
  • the identification model has a limited number of basic images at the beginning of the establishment.
  • the image to be judged is added as a base image to the recognition model, so that the number of learning samples of the recognition model is continuously increased, and the recognition model is The accuracy of image recognition to be judged is increased, making the recognition result more reliable.
  • the updating unit 616 includes an update subunit 618 configured to perform feature extraction on the image to be judged using a convolutional neural network algorithm of the tensorflow framework, and store the features of the image to be determined into the recognition model.
  • the update sub-unit 618 uses the convolutional neural network algorithm of the tensorflow framework to perform feature extraction on the image to be judged, that is, the same algorithm as the feature extraction of the base image is used, and the feature extracted from the image to be determined is guaranteed.
  • the object features extracted from the image and the consistency of the non-object features are avoided, and the learning content in the recognition model is inconsistent due to the difference of the extraction algorithm, thereby causing the recognition model to be inaccurate in the image recognition.
  • the convolutional neural network algorithm using the tensorflow framework is used to judge The image is extracted, and the feature of the image to be judged is stored in the recognition model, so that the recognition model performs feature learning, thereby ensuring the accuracy of the recognition model for the next image, thereby improving reliability.
  • the updating unit 616 further includes: a classification unit 620, configured to determine a classification category of the image to be determined according to the recognition result; send the image to be determined and the classification category, and receive the An image to be judged and an audit result of the classification category, the audit result indicating whether the image to be determined is consistent with the classification category; the update subunit 618 is further configured to: when the audit result indicates the image to be determined When the classification categories are consistent, feature extraction is performed on the image to be determined using a convolutional neural network algorithm of the tensorflow framework, and features of the image to be determined are stored in the recognition model; when the audit result indicates When the image to be judged is inconsistent with the classification category, the storage fails.
  • a classification unit 620 configured to determine a classification category of the image to be determined according to the recognition result
  • send the image to be determined and the classification category and receive the An image to be judged and an audit result of the classification category, the audit result indicating whether the image to be determined is consistent with the classification category
  • the classification unit 620 before the update unit adds the image to be determined to the recognition model, the classification unit 620 further determines the classification category of the image to be determined according to the recognition result, and sends the image to be determined and the corresponding classification category.
  • the update subunit 618 uses the volume of the tensorflow framework. The product neural network algorithm performs feature extraction on the judged image. Otherwise, it does not store, classifies the image to be judged, and audits whether the classification is correct.
  • the recognition model identifies the image to be judged as having a dust-proof net, and stores the image to be judged in the recognition model. Since the wrong classification causes the recognition model to treat the image in the next image recognition.
  • the recognition is performed, an erroneous recognition result is given, and the image to be judged is classified according to the recognition result, and sent to the auditor for review, and the image to be judged is stored in the recognition model only when the classification category is consistent with the image to be determined. To avoid identifying the wrong learning of the model.
  • FIG. 7 shows a schematic block diagram of a computer device 700 according to an embodiment of the present application.
  • the computer device 700 includes a memory 702, a processor 704, and a computer program stored on the memory 702 and executable on the processor 704.
  • the processor 704 implements any one of the above when executing the computer program. Item is used for image recognition methods for monitoring.
  • the computer device 700 provided by the present application, when the processor 704 executes the computer program, realizes that the image to be determined is acquired, and the image to be determined is compared with the recognition model, thereby identifying whether the image to be determined includes the object feature, and transmitting the recognition result to the
  • the environmental protection department can obtain the on-site image of the coal yard in real time, and obtain the violation status of the coal yard through the identification model. It is not necessary to assign a special person to monitor the coal yard, which avoids the repeated round-trip shooting of the staff and eliminates the use of the coal yard. The shooting interval is used to ensure the effective coverage of the dust-proof net of the coal yard, thereby reducing the pollution of the coal yard dust to the air.
  • a computer readable storage medium on which a computer program is stored, and when the computer program is executed by the processor, an image recognition method for monitoring according to any of the above is implemented.
  • the computer-readable storage medium provided by the present application when executed by a processor, implements acquiring an image to be determined, and compares the image to be determined with the recognition model, thereby identifying whether the image to be determined includes an object feature, and transmitting the recognition result
  • the environmental protection department can obtain the on-site image of the coal yard in real time, and obtain the violation status of the coal yard through the identification model. It is not necessary to assign a special person to monitor the coal yard, avoiding the repeated round-trip shooting of the staff and eliminating the coal yard.
  • the use of the shooting interval to drill the voids ensures the effective coverage of the dust-proof net of the coal yard, thereby reducing the pollution of the coal yard dust to the air.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Biology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Signal Processing (AREA)
  • Image Analysis (AREA)

Abstract

公开了一种用于监控的图像识别方法及系统、一种计算机设备以及可读存储介质。其中用于监控的图像识别方法包括：获取待判断图像；将待判断图像与识别模型进行比较，识别待判断图像是否包含物体特征，得到识别结果；发送识别结果。

Description

用于监控的图像识别方法及系统、计算设备以及可读存储介质
本申请要求在2017年11月30日提交中国专利局、申请号为201711242658.3的中国专利申请的优先权，该申请的全部内容通过引用结合在本申请中。
技术领域
本申请涉及图像识别领域，例如涉及一种用于监控的图像识别方法及系统、一种计算机设备以及计算机可读存储介质。
背景技术
智能城市项目中对于环保领域中智能监测以及识别是未来发展的主要方向，例如实时监控天气的质量、监控城市的污染物大小、以及监控城市的污染来源三方面完成对空气的监测。现阶段，在监控城市的污染来源中煤炭监控一直还处于人工定时巡检监控，每隔一段时间，由环保部门委派一名专门的工作人员前往现场检查，将拍摄得到的违规事件保存在摄像机中，并上传到环保业务系统中。在此过程中，工作人员需要多次往返煤场获取现场的违规事件，人员的利用率很低。由于环保部门无法实时监控到煤场的情况，极易出现工作人员不能及时发现煤场违规不使用防尘网覆盖情况的发生，无法对煤场实现实时有效的监控，造成大气的污染。
发明内容
本申请旨在至少解决相关技术中存在的技术问题之一。
本申请提出了一种用于监控的图像识别方法。
本申请的提出了一种用于监控的图像识别系统。
本申请的提出了一种计算机设备。
本申请的提出了一种计算机可读存储介质。
本申请提出了一种用于监控的图像识别方法，包括：获取待判断图像；将待判断图像与识别模型进行比较，识别待判断图像是否包含物体特征，得到识别结果；发送识别结果。
本申请提供的用于监控的图像识别方法，通过获取待判断图像，并将待判断图像与识别模型进行比较，从而识别待判断图像中是否包含物体特征，并将 识别结果发送至环保业务系统中，环保部门可以实时获取煤场的现场图像，并通过识别模型得到煤场的违规情况，无需委派专人对煤场进行监控，避免了工作人员重复往返拍摄图像，杜绝了煤场利用拍摄间隔钻空子，保证了煤场的防尘网的有效覆盖，从而减少煤场粉尘对于空气的污染。
根据本申请的上述用于监控的图像识别方法，还可以具有以下技术特征：
在一实施例中，在获取待判断图像之前，还包括：建立识别模型，建立识别模型包括：接收基础图像，基础图像包括：第一图像和第二图像，第一图像包含物体特征，第二图像包含非物体特征；使用tensorflow框架(GOOGLE公司推出的一种机器学习的开源框架)的卷积神经网络技术建立第一层卷积神经网络以及第二层卷积神经网络；以及根据第一层卷积神经网络以及第二层卷积神经网络，建立全连接网络，得到模型结构；使用tensorflow框架的卷积神经网络算法对所述基础图像进行特征提取，以提取所述第一图像的物体特征和所述第二图像的非物体特征进行提取，并将所述物体特征和所述非物体特征存储至模型结构中，得到识别模型。
在本公开中，使用tensorflow框架的卷积神经网络技术建立模型结构，并将具有物体特征的第一图像以及具有非物体特征的第二图像作为基础图像添加到模型结构，其中，tensorfiow框架是谷歌公司基于DistBelief(GOOGLE公司推出的一种机器学习的开源框架)进行研发的用于张量基于数据流图计算的框架。使用tensorflow框架的卷积神经网络算法对物体特征以及非物体特征进行提取，并添加到识别模型，使得识别模型能够根据添加的物体特征以及非物体特征学习如何分辨出图像是否具有物体，识别速度快，准确度高，并且全过程不需要人员参与；对模型结构中添加的基础图像数量越多识别的结果越准确，即识别模型的准确率越高，通过使用识别模型摆脱了由人工进行图片识别，减少环保业务部分对于图像的工作量，提高工作效率，并且保证对于煤场的有效监控，有效杜绝煤场利用图片拍摄周期钻空子对空气造成污染。
在一实施例中，将待判断图像存储至识别模型。
在本实施例中，识别模型在完成对待判断图像进行识别后，将待判断图像存储到识别模型中，使得识别模型中对于物体特征的学习对象的数量增加，在对下一次的图像进行判断时，识别模型识别的结果更准确，出现错误的可能性更低，并且随着待判断图像数量的不断增加，识别模型不断完善，识别结果更可靠。
在一实施例中，将待判断图像存储至识别模型包括：使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，并将待判断图像的特征存储至识别模型中。
在本实施例中，使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，即使用与基础图像的特征提取相同的算法，保证了从待判断图像提取的特征与从基础图像提取的物体特征以及非物体特征的一致性，避免出现因为提取算法的不同造成识别模型中学习内容不一致，从而造成识别模型对待判断图像识别不准确，使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，并将待判断图像的特征存储在识别模型中，以使识别模型进行特征学习，从而保证了识别模型对下一次图像判断的准确性，从而提高可靠性。
在一实施例中，在使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，并将待判断图像的特征存储至识别模型中之前，还包括：根据识别结果，对待判断图像进行分类，得到分类类别；发送待判断图像以及分类类别，并接收待判断图像以及分类类别的审核结果，所述审核结果表明待判断图像与所述分类类别是否对应一致；当审核结果表明待判断图像与分类类别对应一致时，使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，并将待判断图像的特征存储至识别模型中；当所述审核结果表明所述待判断图像与所述分类类别不对应一致时，存储失败。
在本实施例中，在将待判断图像添加到识别模型之前，还要根据识别结果对待判断图像进行分类，得到分类类别，并且将待判断图像以及对应的分类类别发送出去，以使审核人员对分类类别以及对应的待判断图像进行审核，确定分类类别的正确以及对应的待判断图像是否符合要求，对于待判断图像与分类类别一致时，使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，否则，不进行存储，通过对待判断图像进行分类以及对待判断图像的分类正确性进行审核，保证了添加到识别模型中的待判断图像的分类正确性，避免由于分类出现错误时，使得识别模型学习内容出现相反，造成学习后的识别模型逻辑不清，识别出现错误等情况的发生。
本申请提出了一种用于监控的图像识别系统，用于监控的图像识别系统包括：获取单元，设置为获取待判断图像；识别单元，设置为将待判断图像与识别模型进行比较，识别待判断图像是否包含物体特征，得到识别结果；发送单元，设置为发送识别结果。
本申请提供的用于监控的图像识别系统，通过获取单元获取待判断图像，识别单元将待判断图像与识别模型进行比较，从而识别待判断图像中是否包含物体特征，发送单元将识别结果发送至环保业务系统中，环保部门可以实时获取煤场的现场图像，并通过识别模型实时得到煤场的违规情况，无需委派专人对煤场进行监控，避免了工作人员重复往返拍摄图像，杜绝了煤场利用拍摄间隔钻空子，保证了煤场的防尘网的有效覆盖，从而减少煤场粉尘对于空气的污染。
在一实施例中，用于监控的图像识别系统还包括：建立单元，设置为建立识别模型；建立单元包括：第一接收单元、建模单元和存储单元。第一接收单元设置为接收基础图像，基础图像包括：第一图像和第二图像，第一图像包含物体特征，第二图像包含非物体特征；建模单元设置为使用tensorflow框架的卷积神经网络技术建立第一层卷积神经网络以及第二层卷积神经网络，以及根据第一层卷积神经网络以及第二层卷积神经网络建立全连接网络，以得到模型结构；存储单元设置为使用tensorflow框架的卷积神经网络算法对所述基础图像进行特征提取，以提取所述第一图像的物体特征和所述第二图像的非物体特征，并将所述物体特征和所述非物体特征存储至模型结构中，以得到识别模型。
在本实施例中，使用tensorflow框架的卷积神经网络技术建立模型结构，并将具有物体特征的第一图像以及具有非物体特征的第二图像作为基础图像添加到模型结构，其中，tensorfiow框架是谷歌公司基于DistBelief进行研发的用于张量基于数据流图计算的框架。使用tensorflow框架的卷积神经网络算法对物体特征以及非物体特征进行提取，并添加到识别模型，使得识别模型能够根据添加的物体特征以及非物体特征学习如何分辨出图像是否具有物体，识别速度快，准确度高，并且全过程不需要人员参与；对模型结构中添加的基础图像数量越多识别的结果越准确，即识别模型的准确率越高，通过使用识别模型摆脱了由人工进行图片识别，减少环保业务部门对于图像的工作量，提高工作效率，并且保证对于煤场的有效监控，有效杜绝煤场利用图片拍摄周期钻空子对空气造成污染。
在一实施例中，用于监控的图像识别系统还包括：更新单元，设置为将待判断图像存储至识别模型。
在本实施例中，识别模型在完成对待判断图像进行识别后，将待判断图像存储到识别模型中，使得识别模型中对于物体特征的学习对象的数量增加，在 对下一次的图像进行判断时，识别模型识别的结果更准确，出现错误的可能性更低，并且随着待判断图像数量的不断增加，识别模型不断完善，识别结果更可靠。
在一实施例中，更新单元包括：更新子单元，设置为使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，并将待判断图像的特征存储至识别模型中。
在本实施例中，使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，即使用与基础图像的特征提取相同的算法，保证了从待判断图像提取的特征与从基础图像提取的物体特征以及非物体特征的一致性，避免出现因为提取算法的不同造成识别模型中学习内容不一致，从而造成识别模型对待判断图像识别不准确，使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，并将待判断图像的特征存储在识别模型中，以使识别模型进行特征学习，从而保证了识别模型对下一次图像判断的准确性，从而提高可靠性。
在一实施例中，更新单元还包括：分类单元，设置为根据所述识别结果，对所述待判断图像进行分类，得到分类类别；发送所述待判断图像以及所述分类类别，并接收所述待判断图像以及所述分类类别的审核结果，所述审核结果表明待判断图像与所述分类类别是否一致；更新子单元还设置为当所述审核结果表明所述待判断图像与所述分类类别一致时，使用tensorflow框架的卷积神经网络算法对所述待判断图像进行特征提取，并将所述待判断图像的特征存储至所述识别模型中；当所述审核结果表明所述待判断图像与所述分类类别不一致时，存储失败。
在本实施例中，在将待判断图像添加到识别模型之前，还要根据识别结果确定待判断图像的分类类别，并且将待判断图像以及对应的分类类别发送出去，以使审核人员对分类类别以及对应的待判断图像进行审核，确定分类类别的正确以及对应的待判断图像是否符合要求，对于待判断图像与分类类别一致时，使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，否则，不进行存储，通过对待判断图像进行分类以及对分类是否正确进行审核，保证了添加到识别模型中的待判断图像的分类正确性，避免由于分类出现错误时，使得识别模型学习内容出现相反，造成学习后的识别模型逻辑不清，识别出现错误等情况的发生。
本申请提出了一种计算机装置，包括存储器、处理器及存储在存储器上并 可在处理器上运行的计算机程序，处理器执行计算机程序时实现如上述任一项的用于监控的图像识别方法。
本申请提供的计算机装置，处理器执行计算机程序时实现通过获取待判断图像，并将待判断图像与识别模型进行比较，从而识别待判断图像中是否包含物体特征，并将识别结果发送至环保业务系统中，环保部门可以实时获取煤场的现场图像，并通过识别模型得到煤场的违规情况，无需委派专人对煤场进行监控，避免了工作人员重复往返拍摄图像，杜绝了煤场利用拍摄间隔钻空子，保证了煤场的防尘网的有效覆盖，从而减少煤场粉尘对于空气的污染。
本申请提出了一种计算机可读存储介质，其上存储有计算机程序，计算机程序被处理器执行时实现如上述任一项的用于监控的图像识别方法。
本申请提供的计算机可读存储介质，计算机程序被处理器执行时实现获取待判断图像，并将待判断图像与识别模型进行比较，从而识别待判断图像中是否包含物体特征，并将识别结果发送至环保业务系统中，环保部门可以实时获取煤场的现场图像，并通过识别模型得到煤场的违规情况，无需委派专人对煤场进行监控，避免了工作人员重复往返拍摄图像，杜绝了煤场利用拍摄间隔钻空子，保证了煤场的防尘网的有效覆盖，从而减少煤场粉尘对于空气的污染。
本申请的附加方面和优点将在下面的描述部分中变得明显，或通过本申请的实践了解到。
附图概述
图1示出了本申请的一个实施例的用于监控的图像识别方法的流程示意图；
图2示出了本申请的一个实施例的用于监控的图像识别方法的流程示意图；
图3示出了本申请的一个实施例的用于监控的图像识别方法的流程示意图；
图4示出了本申请的一个实施例的用于监控的图像识别方法的流程示意图；
图5示出了本申请的一个实施例的用于监控的图像识别方法的流程示意图；
图6示出了本申请的一个实施例的用于监控的图像识别系统的示意框图；
图7示出了本申请的一个实施例的计算机装置的示意框图。
具体实施方式
为了能够更清楚地理解本申请的上述目的、特征和优点，下面结合附图和具体实施方式对本申请进行进一步的详细描述。
在下面的描述中阐述了很多具体细节以便于充分理解本申请，但是，本 申请还可以采用其他不同于在此描述的其他方式来实施，因此，本申请的保护范围并不限于下面公开的具体实施例的限制。
本申请实施例，提出一种用于监控的图像识别方法，图1示出了本申请的一个实施例的用于监控的图像识别方法的流程示意图。该方法包括如下步骤102至步骤106。
在步骤102中，获取待判断图像。
在步骤104中，将待判断图像与识别模型进行比较，识别待判断图像是否包含物体特征，得到识别结果。
在步骤106中，发送识别结果。
在该实施例中，通过获取待判断图像，并将待判断图像与识别模型进行比较，从而识别待判断图像中是否包含物体特征，并将识别结果发送至环保业务系统中，环保部门可以实时获取煤场的现场图像，并通过识别模型得到煤场的违规情况，无需委派专人对煤场进行监控，避免了工作人员重复往返拍摄图像，杜绝了煤场利用拍摄间隔钻空子，保证了煤场的防尘网的有效覆盖，从而减少煤场粉尘对于空气的污染。
图2为本申请一实施例提供的一种用于监控的图像识别方法的流程示意图。其中，该方法包括以下步骤202至步骤212。
在步骤202中，接收基础图像，基础图像包括：第一图像和第二图像，第一图像包含物体特征，第二图像包含非物体特征。
在步骤204中，使用tensorflow框架的卷积神经网络技术建立第一层卷积神经网络以及第二层卷积神经网络；以及根据第一层卷积神经网络以及第二层卷积神经网络，建立全连接网络，得到模型结构。
在步骤206中，使用tensorflow框架的卷积神经网络算法对基础图像进行特征提取，以提取第一图像的物体特征和第二图像的非物体特征，并将物体特征和非物体特征存储至模型结构中，得到识别模型。
在步骤208中，获取待判断图像。
在步骤210中，将待判断图像与识别模型进行比较，识别待判断图像是否包含物体特征，得到识别结果。
在步骤212中，发送识别结果。
在该实施例中，使用tensorflow框架的卷积神经网络技术建立模型结构，并将具有物体特征的第一图像以及具有非物体特征的第二图像作为基础图像 添加到模型结构，其中，tensorflow框架是谷歌公司基于DistBelief进行研发的用于张量基于数据流图计算的框架。使用tensorflow框架的卷积神经网络算法对物体特征以及非物体特征进行提取，并添加到识别模型，使得识别模型能够根据添加的物体特征以及非物体特征学习如何分辨出图像是否具有物体，识别速度快，准确度高，并且全过程不需要人员参与；对模型结构中添加的基础图像数量越多识别的结果越准确，即识别模型的准确率越高，例如：基础图像数量分别在10000张与500张时，识别模型对于待判断图像给出的识别结果中基础图片数量在10000张的可靠性更高，通过使用识别模型摆脱了由人工进行图片识别，减少环保业务部门对于图像的工作量，提高工作效率，并且保证对于煤场的有效监控，有效杜绝煤场利用图片拍摄周期钻空子对空气造成污染。
图3为本申请一实施例提供的一种用于监控的图像识别方法的流程示意图。其中，该方法包括如下步骤302至步骤314。
在步骤302中，接收基础图像，基础图像包括：第一图像和第二图像，第一图像包含物体特征，第二图像包含非物体特征。
在步骤304中，使用tensorflow框架的卷积神经网络技术建立第一层卷积神经网络以及第二层卷积神经网络；以及根据第一层卷积神经网络以及第二层卷积神经网络，建立全连接网络，得到模型结构。
在步骤306中，使用tensorflow框架的卷积神经网络算法对基础图像进行特征提取，以提取第一图像的物体特征和第二图像的非物体特征，并将物体特征和非物体特征存储至模型结构中，得到识别模型。
在步骤308中，获取待判断图像。
在步骤310中，将待判断图像与识别模型进行比较，识别待判断图像是否包含物体特征，得到识别结果。
在步骤312中，发送识别结果。
在步骤314中，将待判断图像存储至识别模型。
在该实施例中，识别模型在完成对待判断图像进行识别后，将待判断图像存储到识别模型中，使得识别模型中对于物体特征的学习对象的数量增加，在对下一次的图像进行判断时，识别模型识别的结果更准确，出现错误的可能性更低，并且随着待判断图像数量的不断增加，识别模型不断完善，识别结果更可靠。
如：识别模型在建立初期基础图像数量有限，通过不断的将待判断图像添加到识别模型，即将待判断图像作为基础图像添加至识别模型，使得识别模型的学习样本数量不断增加，识别模型对下一个待判断图像识别准确率升高，使得识别结果更可靠。
图4为本申请一实施例提供的一种用于监控的图像识别方法的流程示意图。其中，该方法包括如下步骤402至步骤414。
在步骤402中，接收基础图像，基础图像包括：第一图像和第二图像，第一图像包含物体特征，第二图像包含非物体特征。
在步骤404中，使用tensorflow框架的卷积神经网络技术建立第一层卷积神经网络以及第二层卷积神经网络；以及根据第一层卷积神经网络以及第二层卷积神经网络，建立全连接网络，得到模型结构。
在步骤406中，使用tensorflow框架的卷积神经网络算法对基础图像进行特征提取，以提取第一图像的物体特征和第二图像的非物体特征，并将物体特征和非物体特征存储至模型结构中，得到识别模型。
在步骤408中，获取待判断图像。
在步骤410中，将待判断图像与识别模型进行比较，识别待判断图像是否包含物体特征，得到识别结果。
在步骤412中，发送识别结果。
在步骤414中，使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，并将待判断图像的特征存储至识别模型中。
在该实施例中，使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，即使用与基础图像的特征提取相同的算法，保证了从待判断图像提取的特征与从基础图像提取的物体特征以及非物体特征的一致性，避免出现因为提取算法的不同造成识别模型中学习内容不一致，从而造成识别模型对待判断图像识别不准确，使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，并将待判断图像的特征存储在识别模型中，以使识别模型进行特征学习，从而保证了识别模型对下一次图像判断的准确性，从而提高可靠性。
图5为本申请一实施例提供的一种用于监控的图像识别方法的流程示意图。其中，该方法包括如下步骤502至步骤516。
在步骤502中，接收基础图像，基础图像包括：第一图像和第二图像， 第一图像包含物体特征，第二图像包含非物体特征。
在步骤504中，使用tensorflow框架的卷积神经网络技术建立第一层卷积神经网络以及第二层卷积神经网络；以及根据第一层卷积神经网络以及第二层卷积神经网络，建立全连接网络，得到模型结构。
在步骤506中，使用tensorflow框架的卷积神经网络算法对基础图像进行特征提取，以提取第一图像的物体特征和第二图像的非物体特征，并将物体特征和非物体特征存储至模型结构中，得到识别模型。
在步骤508中，获取待判断图像。
在步骤510中，将待判断图像与识别模型进行比较，识别待判断图像是否包含物体特征，得到识别结果。
在步骤512中，发送识别结果。
在步骤514中，根据识别结果，确定待判断图像的分类类别；发送待判断图像以及分类类别，并接收待判断图像以及分类类别的审核结果，所述审核结果表明待判断图像与所述分类类别是否一致。
在步骤516中，当所述审核结果表明所述待判断图像与所述分类类别一致时，使用tensorflow框架的卷积神经网络算法对所述待判断图像进行特征提取，并将所述待判断图像的特征存储至所述识别模型中；当所述审核结果表明所述待判断图像与所述分类类别不一致时，存储失败。
在该实施例中，在将待判断图像添加到识别模型之前，还要根据识别结果对待判断图像进行分类，得到分类类别，并且将待判断图像以及对应的分类类别发送出去，以使审核人员对分类类别以及对应的待判断图像进行审核，确定分类类别是否正确以及对应的待判断图像是否符合要求，对于待判断图像与分类类别一致时，使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，否则，不进行存储，通过对待判断图像进行分类以及进行审核，保证了添加到识别模型中的待判断图像的分类正确性，避免由于分类出现错误时，使得识别模型学习内容出现相反，造成学习后的识别模型逻辑不清，识别出现错误等情况的发生。
如：识别模型将待判断图像错误的识别为存在防尘网，并将此待判断图像存储在识别模型中，由于错误的分类会使在对下一次的图像识别中，识别模型会对待判断图像识别时给出错误的识别结果，通过根据识别结果，对待判断图像进行分类，并且发送给审核人员进行审核，只有在分类类别与待判 断图像一致时，才会将待判断图像存储在识别模型中，从而避免识别模型的错误学习。
本申请实施例，提出一种用于监控的图像识别系统。图6示出了本申请的一个实施例的用于监控的图像识别系统的示意框图。其中用于监控的图像识别系统600包括：获取单元602、识别单元604和发送单元606。
获取单元602，设置为获取待判断图像；识别单元604，设置为将待判断图像与识别模型进行比较，识别待判断图像是否包含物体特征，得到识别结果；发送单元606，设置为发送识别结果。
在该实施例中，通过获取单元602获取待判断图像，识别单元604将待判断图像与识别模型进行比较，从而识别待判断图像中是否包含物体特征，发送单元606将识别结果发送至环保业务系统中，环保部门可以实时获取煤场的现场图像，并通过识别模型得到煤场的违规情况，无需委派专人对煤场进行监控，避免了工作人员重复往返拍摄图像，杜绝了煤场利用拍摄间隔钻空子，保证了煤场的防尘网的有效覆盖，从而减少煤场粉尘对于空气的污染。
在一实施例中，用于监控的图像识别系统还包括：建立单元608，设置为建立识别模型。建立单元608包括：第一接收单元610、建模单元612和存储单元614。第一接收单元610设置为接收基础图像，基础图像包括：第一图像和第二图像，第一图像包含物体特征，第二图像包含非物体特征。建模单元612设置为使用tensorflow框架的卷积神经网络技术建立第一层卷积神经网络以及第二层卷积神经网络；以及根据第一层卷积神经网络以及第二层卷积神经网络，建立全连接网络，得到模型结构。存储单元614设置为使用tensorflow框架的卷积神经网络算法对基础图像进行特征提取，以提取第一图像的物体特征和第二图像的非物体特征，并将物体特征和非物体特征存储至模型结构中，得到识别模型。
在该实施例中，建模单元612使用tensorflow框架的卷积神经网络技术建立模型结构，存储单元614将具有物体特征的第一图像以及具有非物体特征的第二图像作为基础图像添加到模型结构，其中，tensorfiow框架是谷歌公司基于DistBelief进行研发的用于张量基于数据流图计算的框架，而DisBelief是谷歌(Google)于2011年推出的人工深度学习系统。使用tensorflow框架的卷积神经网络算法对基础图像进行特征提取，以提取物体特征以及非物体特征，并将物体特征和非物体特征添加到识别模型，使得识别模型能够 根据添加的物体特征以及非物体特征学习如何分辨出图像的是否具有物体，识别速度快，准确度高，并且全过程不需要人员参与；对模型结构中添加的基础图像数量越多识别的结果越准确，即识别模型的准确率越高，例如：基础图像数量分别在10000张与500张时，识别模型对于待判断图像给出的识别结果中基础图片数量在10000张的可靠性更高，通过使用识别模型摆脱了由人工进行图片识别，减少环保业务部分对于图像的工作量，提高工作效率，并且保证对于煤场的有效监控，有效杜绝煤场利用图片拍摄周期钻空子对空气造成污染。
在一实施例中，用于监控的图像识别系统600还包括：更新单元616，设置将待判断图像存储至识别模型。
在该实施例中，在识别模型完成对待判断图像进行识别后，更新单元616将待判断图像存储到识别模型中，使得识别模型中对于物体特征的学习对象的数量增加，识别单元604在对下一次的图像进行判断时，识别模型识别的结果更准确，出现错误的可能性更低，并且随着待判断图像数量的不断增加，识别模型不断完善，识别结果更可靠。
如：识别模型在建立初期基础图像数量有限，通过不断的将待判断图像添加到识别模型，即将待判断图像作为基础图像添加至识别模型，使得识别模型的学习样本数量不断增加，识别模型对下一个待判断图像识别准确率升高，使得识别结果更可靠。
在一实施例中，更新单元616包括：更新子单元618，设置为使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，并将待判断图像的特征存储至识别模型中。
在该实施例中，更新子单元618使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，即使用与基础图像的特征提取相同的算法，保证了从待判断图像提取的特征与从基础图像提取的物体特征以及非物体特征的一致性，避免出现因为提取算法的不同造成识别模型中学习内容不一致，从而造成识别模型对待判断图像识别不准确，使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，并将待判断图像的特征存储在识别模型中，以使识别模型进行特征学习，从而保证了识别模型对下一次图像判断的准确性，从而提高可靠性。
在一实施例中，更新单元616还包括：分类单元620，设置为根据所述 识别结果，确定所述待判断图像的分类类别；发送所述待判断图像以及所述分类类别，并接收所述待判断图像以及所述分类类别的审核结果，所述审核结果表明所述待判断图像与所述分类类别是否一致；更新子单元618还设置为当所述审核结果表明所述待判断图像与所述分类类别一致时，使用tensorflow框架的卷积神经网络算法对所述待判断图像进行特征提取，并将所述待判断图像的特征存储至所述识别模型中；当所述审核结果表明所述待判断图像与所述分类类别不一致时，存储失败。
在该实施例中，在所述更新单元将待判断图像添加到识别模型之前，还要分类单元620根据识别结果确定待判断图像的分类类别，并且将待判断图像以及对应的分类类别发送出去，以使审核人员对分类类别以及对应的待判断图像进行审核，确定分类类别是否正确以及对应的待判断图像是否符合要求，对于待判断图像与分类类别一致时，更新子单元618使用tensorflow框架的卷积神经网络算法对待判断图像进行特征提取，否则，不进行存储，通过对待判断图像进行分类以及对分类是否正确进行审核，保证了添加到识别模型中的待判断图像的分类正确性，避免由于分类出现错误时，使得识别模型学习内容出现相反，造成学习后的识别模型逻辑不清，识别出现错误等情况的发生。
如：识别模型将待判断图像错误的识别为存在防尘网，并将此待判断图像存储在识别模型中，由于错误的分类会使在对下一次的图像识别中，识别模型会对待判断图像识别时给出错误的识别结果，通过根据识别结果，对待判断图像进行分类，并且发送给审核人员进行审核，只有在分类类别与待判断图像一致时，才会将待判断图像存储在识别模型中，从而避免识别模型的错误学习。
本申请实施例，提出一种计算机设备，图7示出了本申请的一个实施例的计算机装置700的示意框图。其中，该计算机装置700包括：存储器702、处理器704及存储在所述存储器702上并可在所述处理器704上运行的计算机程序，所述处理器704执行计算机程序时实现如上述任一项用于监控的图像识别方法。
本申请提供的计算机装置700，处理器704执行计算机程序时实现通过获取待判断图像，并将待判断图像与识别模型进行比较，从而识别待判断图像中是否包含物体特征，并将识别结果发送至环保业务系统中，环保部门可以 实时获取煤场的现场图像，并通过识别模型得到煤场的违规情况，无需委派专人对煤场进行监控，避免了工作人员重复往返拍摄图像，杜绝了煤场利用拍摄间隔钻空子，保证了煤场的防尘网的有效覆盖，从而减少煤场粉尘对于空气的污染。
本申请实施例，提出了一种计算机可读存储介质，其上存储有计算机程序，计算机程序被处理器执行时实现如上述任一项用于监控的图像识别方法。
本申请提供的计算机可读存储介质，计算机程序被处理器执行时实现获取待判断图像，并将待判断图像与识别模型进行比较，从而识别待判断图像中是否包含物体特征，并将识别结果发送至环保业务系统中，环保部门可以实时获取煤场的现场图像，并通过识别模型得到煤场的违规情况，无需委派专人对煤场进行监控，避免了工作人员重复往返拍摄图像，杜绝了煤场利用拍摄间隔钻空子，保证了煤场的防尘网的有效覆盖，从而减少煤场粉尘对于空气的污染。
在本说明书的描述中，术语“一个实施例”、“一些实施例”、“具体实施例”等的描述意指结合该实施例或示例描述的具体特征、结构、材料或特点包含于本申请的至少一个实施例或示例中。在本说明书中，对上述术语的示意性表述不一定指的是相同的实施例或实例。而且，描述的具体特征、结构、材料或特点可以在任何的一个或多个实施例或示例中以合适的方式结合。

Claims (12)

  1. 一种用于监控的图像识别方法，包括：
    获取待判断图像；
    将所述待判断图像与识别模型进行比较，识别所述待判断图像是否包含物体特征，得到识别结果；
    发送所述识别结果。
  2. 根据权利要求1所述的方法，在获取待判断图像之前，还包括：建立识别模型，所述建立识别模型包括：
    接收基础图像，所述基础图像包括：第一图像和第二图像，所述第一图像包含所述物体特征，所述第二图像包含非物体特征；
    使用tensorflow框架的卷积神经网络技术建立第一层卷积神经网络以及第二层卷积神经网络；以及
    根据所述第一层卷积神经网络以及第二层卷积神经网络，建立全连接网络，得到模型结构；
    使用tensorflow框架的卷积神经网络算法对所述基础图像进行特征提取，以提取所述第一图像的所述物体特征和所述第二图像的所述非物体特征，并将所述物体特征和所述非物体特征存储至所述模型结构中，得到所述识别模型。
  3. 根据权利要求1或2所述的方法，在发送所述识别结果之后，还包括：将所述待判断图像存储至所述识别模型。
  4. 根据权利要求3所述的方法，其中，所述将所述待判断图像存储至所述识别模型包括：
    使用所述tensorflow框架的卷积神经网络算法对所述待判断图像进行特征提取，并将所述待判断图像的特征存储至所述识别模型中。
  5. 根据权利要求4所述的方法，在使用所述tensorflow框架的卷积神经网络算法对所述待判断图像进行特征提取，并将所述待判断图像的特征存储至所述识别模型中之前，还包括：
    根据所述识别结果，确定所述待判断图像的分类类别；
    发送所述待判断图像以及所述分类类别，并接收所述待判断图像以及所述分类类别的审核结果，所述审核结果表明待判断图像与所述分类类别是否一致；
    当所述审核结果表明所述待判断图像与所述分类类别一致时，使用所述 tensorflow框架的卷积神经网络算法对所述待判断图像进行特征提取，并将所述待判断图像的特征存储至所述识别模型中；当所述审核结果表明所述待判断图像与所述分类类别不一致时，存储失败。
  6. 一种用于监控的图像识别系统，其中，
    获取单元，设置为获取待判断图像；
    识别单元，设置为将所述待判断图像与识别模型进行比较，识别所述待判断图像是否包含物体特征，得到识别结果；
    发送单元，设置为发送所述识别结果。
  7. 根据权利要求6所述的系统，还包括：
    建立单元，设置为建立识别模型；所述建立单元包括：第一接收单元、建模单元和存储单元；
    所述第一接收单元设置为接收基础图像，所述基础图像包括：第一图像和第二图像，所述第一图像包含所述物体特征，所述第二图像包含非物体特征；
    所述建模单元设置为使用tensorflow框架的卷积神经网络技术建立第一层卷积神经网络以及第二层卷积神经网络；以及根据所述第一层卷积神经网络以及第二层卷积神经网络，建立全连接网络，得到模型结构；
    所述存储单元设置为使用tensorflow框架的卷积神经网络算法对所述基础图像进行特征提取，以提取所述第一图像的所述物体特征和所述第二图像的所述非物体特征，并将所述物体特征和所述非物体特征存储至所述模型结构中，得到所述识别模型。
  8. 根据权利要求6或7所述的系统，还包括：
    更新单元，设置为将所述待判断图像存储至所述识别模型。
  9. 根据权利要求8所述的系统，其中，所述更新单元包括：
    更新子单元，设置为使用所述tensorflow框架的卷积神经网络算法对所述待判断图像进行特征提取，并将所述待判断图像的特征存储至所述识别模型中。
  10. 根据权利要求9所述的系统，其中，所述更新单元还包括：
    分类单元，设置为根据所述识别结果，对所述待判断图像进行分类，得到分类类别；
    发送所述待判断图像以及所述分类类别，并接收所述待判断图像以及所 述分类类别的审核结果，其中所述审核结果表明所述待判断图像与所述分类类别是否一致；
    所述更新子单元，还设置为当所述审核结果表明所述待判断图像与所述分类类别一致时，使用所述tensorflow框架的卷积神经网络算法对所述待判断图像进行特征提取，并将所述待判断图像的特征存储至所述识别模型中；当所述审核结果中所述待判断图像与所述分类类别不一致时，存储失败。
  11. 一种计算机设备，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，其中，所述处理器执行所述计算机程序时实现如权利要求1至5中任一项所述的用于监控的图像识别方法。
  12. 一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现如权利要求1至5中任一项所述的用于监控的图像识别方法。
PCT/CN2018/109771 2017-11-30 2018-10-11 用于监控的图像识别方法及系统、计算设备以及可读存储介质 WO2019105131A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201711242658.3A CN108009497A (zh) 2017-11-30 2017-11-30 图像识别监控方法、系统、计算设备以及可读存储介质
CN201711242658.3 2017-11-30

Publications (1)

Publication Number Publication Date
WO2019105131A1 true WO2019105131A1 (zh) 2019-06-06

Family

ID=62055750

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/109771 WO2019105131A1 (zh) 2017-11-30 2018-10-11 用于监控的图像识别方法及系统、计算设备以及可读存储介质

Country Status (2)

Country Link
CN (1) CN108009497A (zh)
WO (1) WO2019105131A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113255409A (zh) * 2020-02-13 2021-08-13 中国移动通信集团广东有限公司 一种图像识别方法、系统及装置

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108009497A (zh) * 2017-11-30 2018-05-08 深圳中兴网信科技有限公司 图像识别监控方法、系统、计算设备以及可读存储介质
US10990364B2 (en) * 2018-06-06 2021-04-27 Ab Initio Technology Llc Updating executable graphs
CN109118779B (zh) * 2018-10-12 2021-05-11 东软集团股份有限公司 交通违章信息识别方法、设备及可读存储介质
CN111374608B (zh) * 2018-12-29 2021-08-03 尚科宁家(中国)科技有限公司 一种针对扫地机器人镜头的污物检测方法及装置、设备、介质
CN109982051B (zh) * 2019-04-19 2021-06-29 东莞市南星电子有限公司 具有动物识别功能的监控摄像方法、监控相机和存储介质
CN110287770B (zh) * 2019-05-10 2021-08-17 中国船舶重工集团公司第七一五研究所 一种基于卷积神经网络的水中个体目标匹配识别方法
CN111439645A (zh) * 2020-04-03 2020-07-24 宜家智能科技(广州)有限公司 一种电梯智能监控方法
CN112861751B (zh) * 2021-02-22 2024-01-12 中国中元国际工程有限公司 一种机场行李房人员管理方法及装置
CN116189000B (zh) * 2022-11-11 2023-11-14 北京市测绘设计研究院 地块分析方法及装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7705878B2 (en) * 1998-08-17 2010-04-27 Halliburton Energy Services, Inc. Method and apparatus to create a down-hole video log to transmit down-hole video data
CN106779064A (zh) * 2016-11-25 2017-05-31 电子科技大学 基于数据特征的深度神经网络自训练方法
CN106951889A (zh) * 2017-05-23 2017-07-14 煤炭科学技术研究院有限公司 井下高危区域动目标监测和管理系统
CN108009497A (zh) * 2017-11-30 2018-05-08 深圳中兴网信科技有限公司 图像识别监控方法、系统、计算设备以及可读存储介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7705878B2 (en) * 1998-08-17 2010-04-27 Halliburton Energy Services, Inc. Method and apparatus to create a down-hole video log to transmit down-hole video data
CN106779064A (zh) * 2016-11-25 2017-05-31 电子科技大学 基于数据特征的深度神经网络自训练方法
CN106951889A (zh) * 2017-05-23 2017-07-14 煤炭科学技术研究院有限公司 井下高危区域动目标监测和管理系统
CN108009497A (zh) * 2017-11-30 2018-05-08 深圳中兴网信科技有限公司 图像识别监控方法、系统、计算设备以及可读存储介质

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113255409A (zh) * 2020-02-13 2021-08-13 中国移动通信集团广东有限公司 一种图像识别方法、系统及装置

Also Published As

Publication number Publication date
CN108009497A (zh) 2018-05-08

Similar Documents

Publication Publication Date Title
WO2019105131A1 (zh) 用于监控的图像识别方法及系统、计算设备以及可读存储介质
CN109639481B (zh) 一种基于深度学习的网络流量分类方法、系统及电子设备
EP3937128A2 (en) Image defect detection method and apparatus, electronic device, storage medium and product
CN108764456B (zh) 机载目标识别模型构建平台、机载目标识别方法及设备
CN110232379A (zh) 一种车辆姿态检测方法及系统
CN113037783B (zh) 一种异常行为检测方法及系统
CN112613569B (zh) 图像识别方法、图像分类模型的训练方法及装置
CN108809948A (zh) 一种基于深度学习的异常网络连接检测方法
CN113989627B (zh) 一种基于异步联邦学习的城市防控图像检测方法和系统
CN111723656B (zh) 一种基于YOLO v3与自优化的烟雾检测方法及装置
CN110019519A (zh) 数据处理方法、装置、存储介质和电子装置
CN110956123B (zh) 一种富媒体内容的审核方法、装置、服务器及存储介质
CN108197634A (zh) 自动抄表方法、系统、计算机设备和可读存储介质
CN109460365A (zh) 一种系统性能测试方法、装置、设备及存储介质
CN117749836B (zh) 一种基于人工智能的物联网终端监控方法及系统
CN110163081A (zh) 基于ssd的实时区域入侵检测方法、系统及存储介质
CN117333808A (zh) 一种面向轻量化部署的建筑物火灾识别方法及系统
CN113591885A (zh) 目标检测模型训练方法、设备及计算机存储介质
CN105469116A (zh) 一种基于人机交互的幼儿识图及数据扩充方法
CN111498630B (zh) 一种远程的电梯检验视频辅助的诊断方法
CN114581819A (zh) 一种视频行为识别方法及系统
CN107305540A (zh) 地址切分识别方法
US11354935B2 (en) Object recognizer emulation
CN116563770B (zh) 车辆颜色的检测方法、装置及设备和介质
CN113971814A (zh) 数据处理方法、装置、电子设备及计算机存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18883118

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 05/10/2020)

122 Ep: pct application non-entry in european phase

Ref document number: 18883118

Country of ref document: EP

Kind code of ref document: A1