CN111144269A - Signal-related behavior identification method and system based on deep learning - Google Patents
Signal-related behavior identification method and system based on deep learning Download PDFInfo
- Publication number
- CN111144269A CN111144269A CN201911339906.5A CN201911339906A CN111144269A CN 111144269 A CN111144269 A CN 111144269A CN 201911339906 A CN201911339906 A CN 201911339906A CN 111144269 A CN111144269 A CN 111144269A
- Authority
- CN
- China
- Prior art keywords
- data
- module
- layer
- feature extraction
- neural 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.)
- Granted
Links
- 230000006399 behavior Effects 0.000 title claims abstract description 57
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000013135 deep learning Methods 0.000 title claims abstract description 20
- 238000000605 extraction Methods 0.000 claims abstract description 40
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 32
- 238000010801 machine learning Methods 0.000 claims abstract description 21
- 238000007781 pre-processing Methods 0.000 claims abstract description 17
- 238000012549 training Methods 0.000 claims description 21
- 108010076504 Protein Sorting Signals Proteins 0.000 claims description 18
- 238000011176 pooling Methods 0.000 claims description 14
- 238000012545 processing Methods 0.000 claims description 12
- 230000001133 acceleration Effects 0.000 claims description 8
- 238000005457 optimization Methods 0.000 claims description 7
- 230000000694 effects Effects 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 4
- 238000002790 cross-validation Methods 0.000 claims description 3
- 230000004069 differentiation Effects 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 description 11
- 230000006870 function Effects 0.000 description 10
- 230000004913 activation Effects 0.000 description 9
- 238000010586 diagram Methods 0.000 description 6
- 230000007547 defect Effects 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 230000006872 improvement Effects 0.000 description 2
- 238000007637 random forest analysis Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Abstract
The invention relates to the technical field of behavior recognition, in particular to a signal correlation behavior recognition method and a signal correlation behavior recognition system based on deep learning.A difference in data preprocessing is utilized for signals acquired by a sensor to promote data volume, a multi-input convolutional neural network and artificial specified statistics are utilized to carry out feature extraction together, and a machine learning classifier with optimized parameters is utilized to carry out classification recognition; the method has the advantages of simple realization, practical value in the aspect of intelligent product application in the action recognition field, easy popularization and promotion and the like.
Description
The technical field is as follows:
the invention relates to the technical field of behavior recognition, in particular to a signal correlation behavior recognition method and system based on deep learning.
Background art:
the behavior recognition is an emerging research field, has wide application range, can be used for recognizing the behavior state of a human body or an object, and has huge development potential. The behavior recognition has a plurality of realization methods including image recognition and signal recognition; among various methods for behavior recognition, a more common and mature method is signal-related behavior recognition in which an inertial sensor is used to collect signals and then recognize the signals.
With the rapid development of the technology in recent years, sensors are more and more miniaturized and low in cost, and meanwhile, functions of electronic products such as mobile phones are continuously expanded, at present, signal sequences such as acceleration angular velocity and the like are collected through sensors built in the electronic products such as the mobile phones, behaviors of objects such as human bodies or vehicles are judged through a behavior recognition algorithm, finally suggestions or decisions are given according to recorded behavior statistical data, and the method has important significance in multiple aspects such as healthy life, safe travel and the like. Various intelligent devices are popularized continuously at present, and a behavior recognition algorithm has wide prospect and huge practical application value.
In the field of behavior recognition, accuracy of recognition is the most basic requirement and the most critical issue. At present, a behavior recognition method for acquiring signals by utilizing a sensor has certain defects in a recognition algorithm (such as a convolutional neural network or a machine learning algorithm), and is particularly characterized by low recognition accuracy. The existing behavior identification method has defects mainly caused by various factors such as small signal data quantity, insufficient signal feature extraction degree, unsuitability of a classifier and the like. Therefore, a technical scheme capable of effectively improving the accuracy of behavior recognition under the conditions of not adding new equipment and not changing a sensor data acquisition mode is urgently needed.
The invention content is as follows:
aiming at the defects and shortcomings in the prior art, the invention provides a signal-related behavior recognition method and system based on deep learning, which can effectively improve the accuracy of behavior recognition under the conditions of not adding new equipment and not changing a sensor data acquisition mode.
The invention is achieved by the following measures:
a signal correlation behavior recognition method based on deep learning comprises a model training stage and an actual application stage, and is characterized in that the model training stage comprises the following steps:
step 1: collecting a signal sequence and recording the behavior type;
step 2: sequentially filtering, segmenting and differentiating the signal sequence acquired in the step 1 through a preprocessing module, and combining data before and after differentiation to obtain a plurality of groups of data;
and step 3: sending the multiple groups of data obtained in the step 2 and the corresponding behavior types thereof into a multi-input convolutional neural network for training;
and 4, step 4: splitting the multi-input convolutional neural network trained in the step 3, only reserving the structure before the full connection layer, and then using the structure and the manually-specified signal sequence statistic together as a feature extraction module;
and 5: carrying out data feature extraction on the multiple groups of data obtained in the step 2 by using the feature extraction module obtained in the step 4 to obtain data features;
step 6: and (5) sending the features extracted in the step (5) and the corresponding behavior types into a machine learning classifier for training and parameter optimization, and selecting the optimal classifier to obtain a feature classification module.
The practical application stage of the invention comprises the following contents:
and processing the acceleration and angular speed signals acquired in real time sequentially through a preprocessing module, a feature extraction module and a feature classification module which are generated in a model training stage, and outputting a behavior recognition result of the behavior signals acquired in real time.
After signal preprocessing in step 2, the data format is a multi-channel one-dimensional data sequence, meanwhile, since signals collected by the inertial sensor are triaxial acceleration or angular velocity signal sequences, the difference of the signal sequences has physical significance, the data signals are subjected to difference processing, the signal sequences before and after the difference processing jointly form a plurality of groups of signal sequence data, and the data information corresponding to each behavior is expanded.
In step 3 of the invention, the structure of the multi-input convolutional neural network comprises a convolutional layer, a pooling layer, a Flatten layer, a merging layer and a full-connection layer; in addition, Dropout layer boosting effect to avoid over-fitting can be added, and Dropout layer can be located behind the convolution layer or between the full connection layers.
In step 4 of the invention, the trained convolutional neural network is required to be split, the full-link layer is removed, and the convolutional layer, the pooling layer, the Flatten layer and the merging layer are reserved. The method comprises the step of using the split (without full connection layer) convolutional neural network and an artificially specified sequence statistic together as a feature extraction module, wherein the artificially specified sequence statistic can be the mean value, standard deviation, maximum value, minimum value, median of a single sequence, the ratio of the mean values, the ratio of the standard deviations and the covariance of different sequences, and the energy value or the energy ratio of each frequency band after Fourier change of the sequence.
In step 5 of the invention, the characteristics of the data are obtained by the multiple groups of data in step 2 through the characteristic extraction module generated in step 4.
In step 6, the feature data is used for training various machine learning classifiers, cross validation is carried out while the machine learning classifiers are trained to realize parameter optimization, and the machine learning classifier with the highest accuracy is selected as a feature classification module after training is finished; the method is equivalent to that the convolutional neural network trained in the step 3 is supplemented with manual appointed statistics in the merging layer, then the fully-connected layer is replaced by using a machine learning classifier, and deep learning and machine learning jointly form a new algorithm.
The invention also provides a system for identifying the signal correlation behaviors based on deep learning, which is provided with a data acquisition module, a data preprocessing module, a data feature extraction module and a data feature classification module, and is characterized in that the data preprocessing module is provided with a differential processing module for carrying out differential processing on an acquired signal sequence so as to obtain a plurality of groups of data; the data feature extraction module is provided with a deep learning feature extraction module and an artificial feature extraction module, the deep learning feature extraction module is formed by removing a full connection layer from a trained multi-input convolutional neural network, and the artificial feature extraction module is formed by calculating statistics of data of each channel.
The data feature classification module is respectively trained by a plurality of machine learning classifiers (such as a support vector machine, a random forest and K neighbor), optimized in parameters and selected and determined according to accuracy.
For signals acquired by a sensor, the data volume is improved by utilizing the difference in data preprocessing, the multi-input convolutional neural network and the artificial designated statistic are utilized to carry out feature extraction together, and a machine learning classifier with optimized parameters is utilized to carry out classification and identification, so that compared with the existing algorithm, the method has the following advantages: the algorithm of the invention can improve the accuracy of human behavior recognition and carry out more accurate recognition; for the convolutional neural network algorithm, the data volume is improved, the manual appointed statistic is added as the characteristic, parameter optimization is carried out, and the most suitable machine learning classifier is used for classification. For a machine learning algorithm, the method adopts the multi-input convolutional neural network to automatically extract the features on the basis of improving the data volume, and can avoid the problems of too subjective features and insufficient feature importance caused by manual designation of all the features. Meanwhile, the algorithm of the invention only needs to be partially changed without additionally adding equipment, has good confidentiality, can be mixed with other improvement methods for use, and has wide prospect and development space. In conclusion, the algorithm in the invention has good promotion effect and simple realization, has practical value in the aspect of intelligent product application in the field of behavior recognition, and is easy to popularize.
Description of the drawings:
FIG. 1 is a signal flow diagram of a model training phase in the present invention.
Fig. 2 is a signal flow diagram in the practical application stage of the present invention.
FIG. 3 is a schematic signal flow diagram of the pre-processing stage of the present invention.
FIG. 4 is a schematic diagram of the multiple input convolutional neural network in step 3 of the present invention.
Fig. 5 is a schematic diagram of the feature extraction module described in step 4 of the present invention.
FIG. 6 is a schematic diagram of the optimized selection of the feature classification module in step 6 of the present invention.
FIG. 7 is a flow chart of one embodiment of the present invention.
The specific implementation mode is as follows:
the invention is further described below with reference to the figures and examples.
The invention provides a signal correlation behavior recognition method based on deep learning, which comprises a model training stage and an actual application stage as shown in the attached figures 1 and 2, wherein the model training stage comprises the following contents:
step 1: acquiring signals, acquiring acceleration and angular velocity signal sequences through an acceleration sensor and a gyroscope, and recording behavior types;
step 2: filtering, segmenting and differentiating the signal sequence acquired in the step 1 through a preprocessing module, and combining data before and after differentiation to obtain a plurality of groups of preprocessed data (as shown in figure 3);
and step 3: sending the multiple groups of data obtained in the step 2 and the corresponding behavior types thereof into a multi-input convolutional neural network for training (as shown in figure 4);
and 4, step 4: splitting the multi-input convolutional neural network trained in the step 3, only reserving the structure before the full-connection layer, and using the structure and the manually-specified signal sequence statistic as a feature extraction module (as shown in the attached figure 5);
and 5: carrying out data feature extraction on the multiple groups of data obtained in the step 2 by using the feature extraction module obtained in the step 4 to obtain data features;
step 6: and (3) sending the features extracted in the step (5) and the corresponding behavior types thereof into a machine learning classifier for training and parameter optimization, and selecting the optimal classifier to obtain a feature classification module (as shown in figure 6).
The invention also comprises a step 7 of practical application stage, which specifically comprises the following contents: during practical application, acceleration and angular speed signals acquired in real time sequentially pass through the preprocessing module and the feature extraction module generated in the model training stage, the feature classification module processes the signals, and behavior recognition results of the behavior signals acquired in real time are output.
Example 1:
in this example, according to the signal-related behavior recognition method based on deep learning, a use example is provided in which an accelerometer and a gyroscope are used to continuously record human behavior signal data for a period of time, a specific flow is shown in fig. 7, and the specific implementation steps are as follows:
step 1: the data acquisition mode is that the testee continues some kind of behaviors, in this example, an accelerometer and a gyroscope are adopted, the human behavior signal data of a period of time is continuously recorded, and the type of the human behavior signal data is marked;
step 2: the preprocessing module carries out noise filtering preprocessing on the acquired sensor signals; dividing the filtered sampling data into sliding window data with a preset total data length as a sample (the overlapping part of adjacent sliding windows is the preset data length); in this example, the sample data length of each sliding window is 128, and the overlapping part length of adjacent sliding windows is 64; the number of channels is 6 channels of the three-axis data of the acceleration and the angular velocity; carrying out difference on the signal sequences of 128 data of each channel of the 6 channels to obtain sequences of 127 data of each channel of the 6 channels; at this time, the total data is changed into data sequences of two groups of 6 channels before and after difference, and the lengths of the two groups of data are 128 and 127 respectively;
and step 3: in principle, there is no specific requirement on the structure of the convolutional neural network to be constructed, and the structure of the convolutional neural network adopted in this example is as follows: the data sequence (length 128) without difference sequentially passes through a one-dimensional convolutional layer (filter number 32, size 5, activation function ReLU), a Dropout layer (ratio 0.5), a one-dimensional maximum pooling layer (pooling size 12), and a Flatten layer. The data sequence (length is 127) differentiated in the step 2 sequentially passes through a one-dimensional convolutional layer (the number of filters is 32, the size is 5, the activation function is ReLU), a one-dimensional convolutional layer (the number of filters is 32, the size is 4, the activation function is ReLU), a Dropout layer (the ratio is 0.5), a one-dimensional maximum pooling layer (the pooling size is 12) and a Flatten layer; a merging layer for merging the two groups of data after processing through respective layers, a full connection layer (the number of neurons is 100, the activation function is ReLU), a Dropout layer (the ratio is 0.5), a full connection layer (the number of neurons is 6, which is the number of human behavior categories to be judged in the example, and the activation function is softmax); and 4, step 4: the split convolutional neural network is trained in the step 3, and only the convolutional layer and the pooling layer of the trained convolutional neural network have the better capability of extracting the signal characteristics; the split convolutional neural network only reserves the structure before the full connection layer, and the split neural network does not reserve the full connection layer, so the structure of the full connection layer of the convolutional neural network trained in the step 3, including the number of layers, the number of neurons, the activation function and other parameter information, can be kept secret;
in this example, the structure of the split convolutional neural network includes: one-dimensional convolutional layers (the number of filters is 32, the size is 5, and the activation function is ReLU), one-dimensional maximum pooling layers (the pooling size is 12), and a Flatten layer, through which an undifferentiated data sequence (length is 128) is to pass; the differentiated data sequence (length is 127) passes through one-dimensional convolutional layers (the number of filters is 32, the size is 5, and the activation function is ReLU), one-dimensional convolutional layers (the number of filters is 32, the size is 4, and the activation function is ReLU), one-dimensional maximum pooling layers (the pooling size is 12), and a Flatten layer; and a merging layer for merging the two groups of data after the two groups of data are processed by respective layers. The part is used as a part for deeply learning and extracting the characteristics in the characteristic extraction module;
the 5 statistics of the 6 channel data sequences without difference, in this example, the signal sequences of 128 data per channel, such as the mean, standard deviation, maximum, minimum, and median, are used as the part for manually extracting the features in the feature extraction module.
And 5: the data passes through the feature extraction module generated in the step 4, and the data is converted into the features of the data; the characteristics not only have artificially appointed signal time domain (or frequency domain) statistic characteristics, but also have characteristics automatically obtained according to signal characteristics after the first half (convolution layer, pooling layer and the like) of the convolutional neural network is processed, and the defect that the characteristics extracted by the traditional machine learning method or the traditional deep learning method are not comprehensive enough is avoided by a method combining artificial extraction and deep learning extraction characteristics;
step 6: training a machine learning classifier by using the data characteristics extracted in the step 5, and performing parameter optimization by using a cross validation and grid search method; in the embodiment, the support vector machine with optimized parameters, K adjacent neighbors and the random forest which have the highest accuracy are used as the feature classification module.
The following results are compared with the accuracy of the previous example output result in the prior art 1(CN110113116A human behavior recognition method based on WIFI channel information), the prior art 2(CN110222730A user identity recognition method based on inertial sensor and recognition model construction method), and the prior art 3(CN108345846A human behavior recognition method and recognition system based on convolutional neural network), respectively, and the results are as follows:
the adopted technical scheme | Average rate of accuracy |
Prior document 1 | 89.1985% |
Prior document 2 | 90.1432% |
Prior document 3 | 87.9376% |
Inventive example 1 | 92.0163% |
Compared with the prior art, the technical scheme provided by the invention obviously improves the identification accuracy.
Compared with the prior art, the invention has the following advantages: through the improvement of the preprocessing module, the noise-filtered data is subjected to difference processing to form a plurality of groups of data before and after the difference processing, the total data volume is increased, the data volume corresponding to each behavior is increased, and the characteristics of various behaviors can be better extracted by utilizing the data increased after the difference processing; by improving the feature extraction module, the structure of the trained multi-input convolutional neural network before the full connection layer and the manually specified statistic are used as the feature extraction module: the multi-input convolutional neural network and the manual specified statistic amount jointly extract the signal data characteristics, so that the characteristic extraction capability can be improved; by improving the feature classification module, when classifying the features, the full connection layer is not utilized for classification, but the machine learning classifier with optimized parameters is used for classification, so that the classification effect can be improved.
Claims (9)
1. A signal correlation behavior recognition method based on deep learning comprises a model training stage and an actual application stage, and is characterized in that the model training stage comprises the following steps:
step 1: collecting a signal sequence and recording the behavior type;
step 2: sequentially filtering, segmenting and differentiating the signal sequence acquired in the step 1 through a preprocessing module, and combining data before and after differentiation to obtain a plurality of groups of data;
and step 3: sending the multiple groups of data obtained in the step 2 and the corresponding behavior types thereof into a multi-input convolutional neural network for training;
and 4, step 4: splitting the multi-input convolutional neural network trained in the step 3, only reserving the structure before the full connection layer, and using the structure and the manually-specified signal sequence statistic as a feature extraction module together;
and 5: carrying out data feature extraction on the multiple groups of data obtained in the step 2 by using the feature extraction module obtained in the step 4 to obtain data features;
step 6: and (5) sending the features extracted in the step (5) and the corresponding behavior types into a machine learning classifier for training and parameter optimization, and selecting the optimal classifier to obtain a feature classification module.
2. The method according to claim 1, wherein the practical application stage comprises the following steps: and processing the acceleration and angular speed signals acquired in real time sequentially through a preprocessing module, a feature extraction module and a feature classification module which are generated in a model training stage, and outputting a behavior recognition result of the behavior signals acquired in real time.
3. The method for identifying the signal correlation behaviors based on the deep learning as claimed in claim 1, wherein the structure of the multi-input convolutional neural network in the step 3 comprises a convolutional layer, a pooling layer, a Flatten layer, a merging layer and a full-link layer.
4. The method as claimed in claim 3, wherein Dropout layer boosting effect is added to avoid over-fitting, and the Dropout layer is located behind the convolutional layer or between the fully connected layers.
5. The method for identifying the signal correlation behaviors based on the deep learning of claim 1, wherein in the step 4, the convolutional neural network which has been trained in the step 3 is split, the fully-connected layer is removed, the convolutional layer, the pooling layer, the Flatten layer and the merging layer are reserved, and the convolutional neural network which is obtained by splitting and does not contain the fully-connected layer and the manually-specified sequence statistics are jointly used as a feature extraction module.
6. The method according to claim 1, wherein in step 6, the feature data is used for training a plurality of machine learning classifiers, the cross validation is performed while the machine learning classifiers are trained to optimize parameters, and the machine learning classifier with the highest accuracy is selected as the feature classification module after the training is completed.
7. A system for identifying signal-related behaviors based on deep learning is provided with a data acquisition module, a data preprocessing module, a data feature extraction module and a data feature classification module, and is characterized in that the data preprocessing module is provided with a differential processing module for carrying out differential processing on an acquired signal sequence so as to obtain a plurality of groups of data; the data feature extraction module is provided with a deep learning feature extraction module and an artificial feature extraction module, the deep learning feature extraction module is formed by removing a full connection layer from a trained multi-input convolutional neural network, and the artificial feature extraction module is formed by calculating statistics of data of each channel.
8. The system according to claim 7, wherein the artificial feature extraction module comprises one or more of a mean module, a standard deviation module, a maximum module, a minimum module, or a median module.
9. The system according to claim 7, wherein the data feature classification module is trained by a plurality of machine learning classifiers, and performs parameter optimization and then determines the data feature classification according to the accuracy selection.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911339906.5A CN111144269B (en) | 2019-12-23 | 2019-12-23 | Signal correlation behavior recognition method and system based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911339906.5A CN111144269B (en) | 2019-12-23 | 2019-12-23 | Signal correlation behavior recognition method and system based on deep learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111144269A true CN111144269A (en) | 2020-05-12 |
CN111144269B CN111144269B (en) | 2023-11-24 |
Family
ID=70519317
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911339906.5A Active CN111144269B (en) | 2019-12-23 | 2019-12-23 | Signal correlation behavior recognition method and system based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111144269B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112487939A (en) * | 2020-11-26 | 2021-03-12 | 深圳市热丽泰和生命科技有限公司 | Pure vision light weight sign language recognition system based on deep learning |
CN113011262A (en) * | 2021-02-18 | 2021-06-22 | 广州大学华软软件学院 | Multi-size cell nucleus recognition device and method based on convolutional neural network |
CN114615118A (en) * | 2022-03-14 | 2022-06-10 | 中国人民解放军国防科技大学 | Modulation identification method based on multi-terminal convolution neural network |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107239803A (en) * | 2017-07-21 | 2017-10-10 | 国家海洋局第海洋研究所 | Utilize the sediment automatic classification method of deep learning neutral net |
CN108305243A (en) * | 2017-12-08 | 2018-07-20 | 五邑大学 | A kind of magnetic tile surface defect detection method based on deep learning |
CN108388348A (en) * | 2018-03-19 | 2018-08-10 | 浙江大学 | A kind of electromyography signal gesture identification method based on deep learning and attention mechanism |
CN108717569A (en) * | 2018-05-16 | 2018-10-30 | 中国人民解放军陆军工程大学 | It is a kind of to expand full convolutional neural networks and its construction method |
CN108875674A (en) * | 2018-06-29 | 2018-11-23 | 东南大学 | A kind of driving behavior recognition methods based on multiple row fusion convolutional neural networks |
CN109214250A (en) * | 2017-07-05 | 2019-01-15 | 中南大学 | A kind of static gesture identification method based on multiple dimensioned convolutional neural networks |
US20190114511A1 (en) * | 2017-10-16 | 2019-04-18 | Illumina, Inc. | Deep Learning-Based Techniques for Training Deep Convolutional Neural Networks |
CN110427965A (en) * | 2019-06-25 | 2019-11-08 | 重庆邮电大学 | Convolutional neural networks structural reduction and image classification method based on evolution strategy |
CN110502991A (en) * | 2019-07-18 | 2019-11-26 | 武汉理工大学 | Internal combustion engine health monitor method and system based on random convolutional neural networks structure |
-
2019
- 2019-12-23 CN CN201911339906.5A patent/CN111144269B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109214250A (en) * | 2017-07-05 | 2019-01-15 | 中南大学 | A kind of static gesture identification method based on multiple dimensioned convolutional neural networks |
CN107239803A (en) * | 2017-07-21 | 2017-10-10 | 国家海洋局第海洋研究所 | Utilize the sediment automatic classification method of deep learning neutral net |
US20190114511A1 (en) * | 2017-10-16 | 2019-04-18 | Illumina, Inc. | Deep Learning-Based Techniques for Training Deep Convolutional Neural Networks |
CN108305243A (en) * | 2017-12-08 | 2018-07-20 | 五邑大学 | A kind of magnetic tile surface defect detection method based on deep learning |
CN108388348A (en) * | 2018-03-19 | 2018-08-10 | 浙江大学 | A kind of electromyography signal gesture identification method based on deep learning and attention mechanism |
CN108717569A (en) * | 2018-05-16 | 2018-10-30 | 中国人民解放军陆军工程大学 | It is a kind of to expand full convolutional neural networks and its construction method |
CN108875674A (en) * | 2018-06-29 | 2018-11-23 | 东南大学 | A kind of driving behavior recognition methods based on multiple row fusion convolutional neural networks |
CN110427965A (en) * | 2019-06-25 | 2019-11-08 | 重庆邮电大学 | Convolutional neural networks structural reduction and image classification method based on evolution strategy |
CN110502991A (en) * | 2019-07-18 | 2019-11-26 | 武汉理工大学 | Internal combustion engine health monitor method and system based on random convolutional neural networks structure |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112487939A (en) * | 2020-11-26 | 2021-03-12 | 深圳市热丽泰和生命科技有限公司 | Pure vision light weight sign language recognition system based on deep learning |
CN113011262A (en) * | 2021-02-18 | 2021-06-22 | 广州大学华软软件学院 | Multi-size cell nucleus recognition device and method based on convolutional neural network |
CN113011262B (en) * | 2021-02-18 | 2023-10-13 | 广州大学华软软件学院 | Multi-size cell nucleus identification device and method based on convolutional neural network |
CN114615118A (en) * | 2022-03-14 | 2022-06-10 | 中国人民解放军国防科技大学 | Modulation identification method based on multi-terminal convolution neural network |
CN114615118B (en) * | 2022-03-14 | 2023-09-22 | 中国人民解放军国防科技大学 | Modulation identification method based on multi-terminal convolution neural network |
Also Published As
Publication number | Publication date |
---|---|
CN111144269B (en) | 2023-11-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111144269B (en) | Signal correlation behavior recognition method and system based on deep learning | |
CN108491077B (en) | Surface electromyographic signal gesture recognition method based on multi-stream divide-and-conquer convolutional neural network | |
CN105976809B (en) | Identification method and system based on speech and facial expression bimodal emotion fusion | |
CN111062250B (en) | Multi-subject motor imagery electroencephalogram signal identification method based on deep feature learning | |
CN110353702A (en) | A kind of emotion identification method and system based on shallow-layer convolutional neural networks | |
CN110600054B (en) | Sound scene classification method based on network model fusion | |
CN106909784A (en) | Epileptic electroencephalogram (eeg) recognition methods based on two-dimentional time-frequency image depth convolutional neural networks | |
CN103366180A (en) | Cell image segmentation method based on automatic feature learning | |
CN101930285B (en) | Handwriting recognition method based on surface electromyographic signal | |
CN105841961A (en) | Bearing fault diagnosis method based on Morlet wavelet transformation and convolutional neural network | |
CN110399846A (en) | A kind of gesture identification method based on multichannel electromyography signal correlation | |
CN111150372B (en) | Sleep stage staging system combining rapid representation learning and semantic learning | |
CN108510011B (en) | User travel mode analysis method based on mobile phone multi-sensor | |
CN107582077A (en) | A kind of human body state of mind analysis method that behavior is touched based on mobile phone | |
CN108209870A (en) | Long-term EEG monitoring automatic seizure detection method based on convolutional neural networks | |
CN114897025A (en) | Human body posture recognition model establishing method and human body posture recognition method | |
CN112465069A (en) | Electroencephalogram emotion classification method based on multi-scale convolution kernel CNN | |
CN110074779A (en) | A kind of EEG signal identification method and device | |
CN109254654B (en) | Driving fatigue feature extraction method combining PCA and PCANet | |
CN113180659B (en) | Electroencephalogram emotion recognition method based on three-dimensional feature and cavity full convolution network | |
CN112698720B (en) | Movement imagery identification method based on mutual information feature extraction and multi-person fusion | |
CN112006696A (en) | Emotion recognition method based on skin electric signal | |
CN112370058A (en) | Method for identifying and monitoring emotion of user based on mobile terminal | |
CN109271889A (en) | A kind of action identification method based on the double-deck LSTM neural network | |
CN114842280A (en) | Automatic micro-seismic signal identification algorithm based on convolutional neural network |
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 |