CN114898463A - Sitting posture identification method based on improved depth residual error network - Google Patents
Sitting posture identification method based on improved depth residual error network Download PDFInfo
- Publication number
- CN114898463A CN114898463A CN202210498389.1A CN202210498389A CN114898463A CN 114898463 A CN114898463 A CN 114898463A CN 202210498389 A CN202210498389 A CN 202210498389A CN 114898463 A CN114898463 A CN 114898463A
- Authority
- CN
- China
- Prior art keywords
- data
- sitting posture
- pressure
- residual error
- error 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
- 230000001976 improved effect Effects 0.000 title claims abstract description 41
- 238000000034 method Methods 0.000 title claims abstract description 30
- 239000011159 matrix material Substances 0.000 claims abstract description 38
- 238000012545 processing Methods 0.000 claims abstract description 36
- 238000001914 filtration Methods 0.000 claims abstract description 9
- 239000010409 thin film Substances 0.000 claims abstract description 7
- 230000002159 abnormal effect Effects 0.000 claims abstract description 5
- 238000011176 pooling Methods 0.000 claims description 11
- 230000009467 reduction Effects 0.000 claims description 10
- 238000012512 characterization method Methods 0.000 claims description 5
- 230000000694 effects Effects 0.000 claims description 5
- 238000005728 strengthening Methods 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 4
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 230000035945 sensitivity Effects 0.000 claims description 3
- 238000012549 training Methods 0.000 claims description 3
- 230000036544 posture Effects 0.000 description 60
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000000354 decomposition reaction Methods 0.000 description 3
- 230000007613 environmental effect Effects 0.000 description 3
- 239000010408 film Substances 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 230000000276 sedentary effect Effects 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000001939 inductive effect Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000003340 mental effect Effects 0.000 description 1
- 238000003825 pressing Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 230000008093 supporting effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01L—MEASURING FORCE, STRESS, TORQUE, WORK, MECHANICAL POWER, MECHANICAL EFFICIENCY, OR FLUID PRESSURE
- G01L5/00—Apparatus for, or methods of, measuring force, work, mechanical power, or torque, specially adapted for specific purposes
- G01L5/0028—Force sensors associated with force applying means
-
- 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
-
- 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
- G06N3/08—Learning methods
Abstract
The invention discloses a sitting posture identification method based on an improved depth residual error network, which comprises the following steps: collecting pressure values of each point of the array type thin film pressure sensor in a period of time, and carrying out weighted average to obtain a pressure matrix corresponding to the array type thin film pressure sensor; inputting the pressure matrix into an upper computer, and carrying out standardized processing; performing cross weighted filtering processing on the standardized pressure matrix to obtain a pressure matrix with a smooth graph surface; correcting the abnormal extreme value of the pressure matrix with the smooth graph surface by adopting an extreme value processing function; and inputting the corrected pressure matrix into a pre-trained CEGN-ResNet model, and outputting the confidence coefficient of each sitting posture.
Description
Technical Field
The invention relates to a sitting posture identification method based on an improved depth residual error network, and belongs to the technical field of bad sitting posture and sedentary monitoring.
Background
The sitting posture identification algorithm is one of the core technologies of the intelligent seat device. The intelligent seat needs to realize real-time monitoring of the sitting posture and provide sitting posture suggestions for users. Therefore, an easy and accurate sitting posture identification algorithm is a prerequisite. In recent years, more and more people change from physical labor to mental labor, which also leads to the rising of people suffering from cervical and lumbar diseases due to sedentary sitting and wrong sitting postures. Meanwhile, the intelligent seat equipment is lost in the market for a long time and cannot meet the requirements of the sedentary crowd.
Many researches show that the sensors depended on by the current sitting posture recognition algorithm are not easy to integrate, have low usability and have great limitation. How to utilize the sensor which is beneficial to integration, and the key for seat intellectualization is to efficiently, accurately and real-timely identify the sitting posture of the user.
Chinese patent CN108898805A discloses a sitting posture reminding device and system, wherein the sitting posture reminding device includes a sitting posture collecting module, a microprocessor, an alarm module, a supporting component and a clamping part. Chinese patent CN113384266A discloses an intelligent sitting posture monitoring and analyzing system and method, which provides a set of method for simply analyzing the left-leaning and right-leaning of sitting posture by using the regional pressure ratio and uploading data to a server platform to realize reminding. However, the invention patent does not relate to a technology for adjusting reminding time according to user behaviors, does not relate to the technical content of how to distinguish the sitting person entity, does not make innovation on the piezoelectric conversion acquisition module, and does not solve the problem of needing a negative power supply.
Therefore, aiming at the defects of the prior art, the requirements of the system module on the hardware structure can be effectively reduced while the hardware end is easy to deploy, the operation is fast, and the precision is high. Meanwhile, the system aims to achieve the purposes of reducing unnecessary operation and non-inductive operation of a user, and improves the accuracy and the efficiency of real-time detection of the system while reducing system hardware modules installed in the seat.
Furthermore, on the one hand, due to the differences in understanding to the person skilled in the art; on the other hand, since the inventor has studied a lot of documents and patents when making the present invention, but the space is not limited to the details and contents listed in the above, however, the present invention is by no means free of the features of the prior art, but the present invention has been provided with all the features of the prior art, and the applicant reserves the right to increase the related prior art in the background.
Disclosure of Invention
The invention aims to provide a sitting posture identification method based on an improved depth residual error network, which solves the defects of the prior art.
A sitting posture identification method based on an improved depth residual error network comprises the following steps:
collecting pressure values of each point of the array type thin film pressure sensor in a period of time, and carrying out weighted average to obtain a pressure matrix corresponding to the array type thin film pressure sensor;
inputting the pressure matrix into an upper computer, and carrying out standardized processing;
performing cross weighted filtering processing on the standardized pressure matrix to obtain a pressure matrix with a smooth graph surface;
correcting the abnormal extreme value of the pressure matrix with the smooth graph surface by adopting an extreme value processing function;
and inputting the corrected pressure matrix into a pre-trained CEGN-ResNet model, and outputting a confidence matrix for predicting each sitting posture.
Further, the cross weighted filtering method comprises the following steps:
selecting 2 adjacent pixels on the upper, lower, left and right sides of a target pixel to construct a 2-cross field (x is 2), and setting v d And the point position pressure value with the distance d (d is more than or equal to 0 and less than or equal to x) from the P pixel is represented by the following formula:
wherein v is d A value representing the position from the target pixel d, x representing the maximum distance from the pixel point location to be calculated, x being 2, i.e. d being 1, 2, due to the 2-cross field selected. α, β, and γ are weighting parameters, and in order to achieve a good effect by approximately satisfying a gaussian distribution in the horizontal and vertical directions, α is 4, β is 1, and γ is 7, and random noise is filtered.
Further, the extremum processing function method includes:
processing the pressure greater than a given threshold value by a logarithmic function, and reserving the data characteristics of the small pressure point, wherein the threshold value is determined by the overall distribution of the pressure data, the base number of the logarithmic function is determined by the characteristics of the pressure data, and the extreme value processing function is as follows:
parameters a and b control the scaling amplitude of the processing function, and directly influence the sensitivity for distinguishing the sitting posture from the unbalanced sitting posture; the parameter k is used to maintain the coherence of the function image, and l is used to determine which values belong to the extremum range, determined by analyzing the distribution of the sitting posture data pressure.
Further, the CEGN-ResNet model training method comprises the following steps:
establishing an attention module for the corrected pressure matrix, so that the corrected pressure matrix is input and characteristic strengthening data is output through the attention module;
processing the feature enhancement data to obtain final attention characterization output data;
an improved residual error network is established by the attention representation output data;
and after the attention characterization data are input into the improved residual error network, a prediction label and an error direct to an actual result are given, a gradient descent method is used for minimizing a loss function, and a CEGN-ResNet model is obtained through continuous iteration.
Further, the attention module is used for extracting spatial features of the data, and the extraction method comprises the following steps:
performing down-sampling on the data with the dimension H, W, C through maximum pooling and average pooling to obtain dimension reduction data with the dimension H, W, 2;
performing 7 × 7 convolution operation on the data subjected to dimensionality reduction to obtain feature enhanced data with the dimensionality of H × W × 1;
and processing the feature enhancement data through a softmax function to obtain final attention feature output data.
Further, the improved residual error network module comprises the following steps:
decomposing the large-size convolution operation, and replacing the large-size 7 × 7 convolution cascade with two groups of 3 × 3 convolutions and one group of average pooling;
and replacing the batch regularization with group regularization, and performing regularization by adopting the group regularization according to the integral distribution and the characteristics of the batch regularization.
Compared with the prior art, the invention has the following beneficial effects:
the sitting posture is predicted by using data collected by the sensor, the sitting posture can be deployed in a seat, the mode is simple, the cost is low, and the feasibility of commercialization and marketization is high;
the cross weighting algorithm is used for noise reduction, the computational power consumption is low, the noise generated by the characteristics of the sensor is well removed, and the accuracy rate of the overall sitting posture identification is improved;
the extreme value is processed by using the extreme value processing function, the computational power consumption is low, and the probability that the left and right inclined sitting postures are mistakenly judged as the legs of the Erlang is reduced;
the attention module is fused, so that the problem that the network cannot capture the characteristics well due to different sitting habits of different people is solved;
an improved residual error network is used, so that the data characteristic and the prediction characteristic of the sitting posture are improved, and the accuracy of sitting posture identification is improved;
the algorithm can efficiently identify the sitting posture, can be deployed to a hardware end, and has a wide application prospect.
Drawings
FIG. 1 is a schematic view of a 2-cross domain processed by the cross-mean method of the present invention;
FIG. 2 is an image of an extremum processing function under default parameters of the present invention;
FIG. 3 is an illustration of the structure of the attention module of the present invention;
FIG. 4 is a general block diagram of the improved residual attention network of the present invention;
FIG. 5 is a comparison graph of the improvement of the residual block in the improved residual attention network of the present invention.
Detailed Description
In order to make the technical means, the creation characteristics, the achievement purposes and the effects of the invention easy to understand, the invention is further described with the specific embodiments.
As shown in fig. 1-3, a sitting posture identification method based on an improved depth residual error network is disclosed, the method comprising:
step 2, after the measurement is finished, inputting the pressure matrix into an upper computer, and standardizing the pressure matrix to a numerical range of 0-255;
step 3, performing cross weighted filtering processing on the standardized pressure matrix to obtain a pressure matrix with a smooth graph surface; the purpose of this step is to filter the noise mixed in the image digitization, and at this time, the abnormal value of the sitting posture pressure matrix is processed, and the pressure graph surface is smooth;
step 4, in order to prevent the sitting posture from being judged mistakenly as the bad sitting posture such as the leg-Erlang due to physiological left or right inclination when the person sits normally, correcting the abnormal extreme value of the pressure matrix with the smooth graph surface by adopting an extreme value processing function, and removing extreme difference values on the left side and the right side;
step 5, inputting the corrected pressure matrix into a pre-trained CEGN-ResNet model, and outputting confidence coefficients of all sitting postures;
in the step, in order to strengthen the characteristics of data and enable an algorithm to accurately identify sitting regions, a space attention module is used for strengthening the characteristics of the data in a CEGN-ResNet model by a corrected pressure matrix, sitting posture data with strengthened characteristics are finally obtained, the preprocessed data are input into an improved depth residual error network module, numerical results output by a residual error network are processed by SoftMax, and the numerical results are finally output as confidence degrees of all sitting postures in an interval of [0,1 ].
In this embodiment, the cross weighted filtering method may be used to filter noise generated by individual points of the array type thin film pressure sensor, so that each point of the pressure matrix is connected into a smooth curved surface to eliminate noise generated by environmental factors, wherein the cross weighted filtering method includes:
selecting 2 adjacent pixels on the upper, lower, left and right sides of a target pixel to construct a 2-cross field (x is 2), and setting upsilon d And the point position pressure value with the distance d (d is more than or equal to 0 and less than or equal to x) from the P pixel is represented, and the structural formula is as follows:
wherein v is d A value representing the position from the target pixel d, x representing the maximum distance from the pixel point location to be calculated, x being 2, i.e. d being 1, 2, due to the 2-cross field selected. α, β, and γ are weighting parameters, and in order to achieve a good effect by approximately satisfying a gaussian distribution in the horizontal and vertical directions, α is 4, β is 1, and γ is 7, and random noise is filtered.
In this embodiment, the extremum processing function method includes:
processing the pressure greater than a given threshold value by a logarithmic function, and reserving the data characteristics of the small pressure point, wherein the threshold value is determined by the overall distribution of the pressure data, and the base number of the logarithmic function is determined by the characteristics of the pressure data; the purpose of this step is that physiological inclination during normal sitting can cause pressure difference between the left and right sides of the pressing surface, and the pressure difference is likely to be misjudged as bad sitting postures such as leg and leg of the Erlang; the extreme value processing function is used for reducing the left-right pressure difference of the pressure surface to a certain extent and reducing the probability of misjudgment as bad sitting posture; the extremum processing function is as follows:
parameters a and b control the scaling amplitude of the processing function, and directly influence the sensitivity for distinguishing the sitting posture from the unbalanced sitting posture; the parameter k is used to maintain the coherence of the function image, and l is used to determine which values belong to the extremum range, determined by analyzing the distribution of the sitting posture data pressure.
The CEGN-ResNet model training method comprises the following steps:
establishing an attention module for the corrected pressure matrix, so that the corrected pressure matrix is input, and feature strengthening data is output through the attention module;
processing the feature enhancement data to obtain final attention characterization output data;
an improved residual error network is established by the attention representation output data;
and after the attention characteristic data are input into the improved residual error network, a prediction label and an error direct to an actual result are given, the error is quantitatively expressed by a loss function, the loss function is minimized by using a gradient descent method, and a CEGN-ResNet model is obtained through continuous iteration.
In this embodiment, an improved residual error network, the improved method includes:
1. large-size convolution operation is decomposed, and due to the fact that the size of the sitting posture data is small, sitting posture features cannot be well collected by using a large convolution kernel, large-size 7 × 7 convolution cascade is replaced by two groups of 3 × 3 convolution and one group of average pooling;
2. group regularization is replaced by batch regularization, because the sitting posture prediction needs to predict frame by frame, the batch regularization which highly depends on integral distribution can lead to low prediction accuracy, and the group regularization simultaneously carries out regularization according to the integral distribution and self characteristics, so the sitting posture prediction can have good effect.
In step 5, the attention module is used for extracting spatial features of the data, and the extraction method comprises the following steps:
performing down-sampling on the data with the dimension H, W, C through maximum pooling and average pooling to obtain dimension reduction data with the dimension H, W, 2;
performing 7 × 7 convolution operation on the data subjected to dimensionality reduction to obtain feature enhanced data with dimensionality H × W × 1;
processing the feature enhancement data through a softmax function to obtain final attention feature data;
the spatial attention module is used, so that the problem that the recognition accuracy is influenced due to different sitting positions and sizes of different people caused by different body types and habits in actual use is solved, and the robustness of the algorithm is enhanced;
the improved residual network consists of an improved residual block, wherein the improved residual block consists of an improved small convolution layer and group regularization for sitting posture identification; the convolution layer uses decomposition convolution operation through decomposition convolution and group regularization processing, and because the size of sitting posture data is small, a large convolution kernel is not beneficial to feature extraction, so that the sitting posture identification is facilitated by decomposing the large convolution kernel;
in the invention, as shown in fig. 1, a cross noise reduction algorithm and an extreme value processing method are used in a data processing stage, a spatial attention mechanism is used in a feature searching and feature enhancing stage, and an improved residual error network is used in a sitting posture predicting stage.
The noise reduction algorithm uses a cross weighted filtering algorithm, as shown in the figure, each target point selects a 2-cross field thereof, and a weight is set according to the distance between the target point and each target point, wherein the weight is smaller when the distance is longer. The weight setting satisfies the gaussian distribution. The algorithm is used for processing noise generated by environmental factors. The algorithm can smooth the edge of the sitting posture image while removing noise generated by environmental factors, and improves the sitting posture identification accuracy to a certain extent.
As shown in fig. 2, the core of the extremum method is to perform a logarithmic function on data exceeding a threshold. The threshold is selected by the distribution of the overall pressure data, and the base of the logarithmic function is selected by the characteristics of the pressure data. The algorithm is used for reducing the probability that the left-right inclined sitting posture is misjudged as the leg-Erlang.
As shown in fig. 3, the core of the spatial attention module is to process the data through maximum pooling and average pooling, and then generate a feature matrix by convolution and activation functions, the feature matrix reflecting the feature distribution of the data. The algorithm is used for strengthening the characteristics of the sitting posture data and enhancing the robustness of the algorithm.
As shown in fig. 4, the core of the improved residual network is a 50-layer residual network, a decomposition convolution and a group regularization mechanism. The 50-layer residual error network meets the requirements of being deployed at a hardware end on the basis of meeting the speed and performance of an algorithm. The decomposed convolution is suitable for prediction of small-size sitting posture data. Group regularization satisfies the requirements of data small-batch prediction.
In summary, the invention has the following advantages:
the sitting posture is predicted by using data collected by the sensor, the sitting posture can be deployed in a seat, the mode is simple, the cost is low, and the feasibility of commercialization and marketization is high;
the cross weighting algorithm is used for noise reduction, the computational power consumption is low, the noise generated by the characteristics of the sensor is well removed, and the accuracy rate of the overall sitting posture identification is improved;
the extreme value is processed by using the extreme value processing function, the computational power consumption is low, and the probability that the left and right inclined sitting postures are mistakenly judged as the legs of the Erlang is reduced;
a space attention mechanism is fused, so that the problem that the network cannot capture characteristics well due to different sitting habits of different people is solved;
an improved residual error network is used, so that the data characteristic and the prediction characteristic of the sitting posture are improved, and the accuracy of sitting posture identification is improved;
the algorithm can efficiently identify the sitting posture, can be deployed to a hardware end, and has a wide application prospect.
The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several modifications and variations can be made without departing from the technical principle of the present invention, and these modifications and variations should also be regarded as the protection scope of the present invention.
Claims (6)
1. A sitting posture identification method based on an improved depth residual error network is characterized by comprising the following steps:
collecting pressure values of each point of the array type thin film pressure sensor in a period of time, and carrying out weighted average to obtain a pressure matrix corresponding to the array type thin film pressure sensor;
inputting the pressure matrix into an upper computer, and carrying out standardized processing;
performing cross weighted filtering processing on the standardized pressure matrix to obtain a pressure matrix with a smooth graph surface;
correcting the abnormal extreme value of the pressure matrix with the smooth graph surface by adopting an extreme value processing function;
and inputting the corrected pressure matrix into a pre-trained CEGN-ResNet model, and outputting a confidence matrix for predicting each sitting posture.
2. The sitting posture identification method based on the improved depth residual error network as claimed in claim 1, wherein the cross weighted filtering method comprises:
selecting 2 adjacent pixels on the upper, lower, left and right sides of a target pixel to construct a 2-cross field (x is 2), and setting v d Represents a distance d (0 ≦ P) from the P pixeld is less than or equal to x), the formula is as follows:
wherein v is d A value representing the position from the target pixel d, x representing the maximum distance from the pixel point location to be calculated, x being 2, i.e. d being 1, 2, due to the 2-cross field selected. α, β, and γ are weighting parameters, and in order to achieve a good effect by approximately satisfying a gaussian distribution in the horizontal and vertical directions, α is 4, β is 1, and γ is 7, and random noise is filtered.
3. The sitting posture identification method based on the improved depth residual error network as claimed in claim 1, wherein the extreme value processing function method comprises:
processing the pressure greater than a given threshold value by a logarithmic function, and reserving the data characteristics of the small pressure point, wherein the threshold value is determined by the overall distribution of the pressure data, the base number of the logarithmic function is determined by the characteristics of the pressure data, and the extreme value processing function is as follows:
parameters a and b control the scaling amplitude of the processing function, and directly influence the sensitivity for distinguishing the sitting posture from the unbalanced sitting posture; the parameter k is used to maintain the coherence of the function image, and l is used to determine which values belong to the extremum range, determined by analyzing the distribution of the sitting posture data pressure.
4. The sitting posture identification method based on the improved deep residual error network as claimed in claim 1, wherein the CEGN-ResNet model training method comprises:
establishing an attention module for the corrected pressure matrix, so that the corrected pressure matrix is input and characteristic strengthening data is output through the attention module;
processing the feature enhancement data to obtain final attention characterization output data;
an improved residual error network is established by the attention representation output data;
and after the attention characterization data are input into the improved residual error network, a prediction label and an error direct to an actual result are given, a gradient descent method is used for minimizing a loss function, and a CEGN-ResNet model is obtained through continuous iteration.
5. The sitting posture identification method based on the improved depth residual error network as claimed in claim 4, wherein the attention module is used for extracting the spatial features of the data, and the extraction method comprises the following steps:
performing down-sampling on the data with the dimension H, W, C through maximum pooling and average pooling to obtain dimension reduction data with the dimension H, W, 2;
performing 7 × 7 convolution operation on the data subjected to dimensionality reduction to obtain feature enhanced data with dimensionality H × W × 1;
and processing the feature enhancement data through a softmax function to obtain final attention feature output data.
6. The sitting posture identifying method based on the improved depth residual error network as claimed in claim 4, wherein the improved residual error network module is improved by the following steps:
decomposing the large-size convolution operation, and replacing the large-size 7 × 7 convolution cascade with two groups of 3 × 3 convolutions and one group of average pooling;
and replacing the batch regularization with group regularization, and performing regularization by adopting the group regularization according to the integral distribution and the characteristics of the batch regularization.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210498389.1A CN114898463B (en) | 2022-05-09 | Sitting posture identification method based on improved depth residual error network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210498389.1A CN114898463B (en) | 2022-05-09 | Sitting posture identification method based on improved depth residual error network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114898463A true CN114898463A (en) | 2022-08-12 |
CN114898463B CN114898463B (en) | 2024-05-14 |
Family
ID=
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111325166A (en) * | 2020-02-26 | 2020-06-23 | 南京工业大学 | Sitting posture identification method based on projection reconstruction and multi-input multi-output neural network |
CN112906720A (en) * | 2021-03-19 | 2021-06-04 | 河北工业大学 | Multi-label image identification method based on graph attention network |
CN114067153A (en) * | 2021-11-02 | 2022-02-18 | 暨南大学 | Image classification method and system based on parallel double-attention light-weight residual error network |
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111325166A (en) * | 2020-02-26 | 2020-06-23 | 南京工业大学 | Sitting posture identification method based on projection reconstruction and multi-input multi-output neural network |
CN112906720A (en) * | 2021-03-19 | 2021-06-04 | 河北工业大学 | Multi-label image identification method based on graph attention network |
CN114067153A (en) * | 2021-11-02 | 2022-02-18 | 暨南大学 | Image classification method and system based on parallel double-attention light-weight residual error network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110363182B (en) | Deep learning-based lane line detection method | |
CN110503630B (en) | Cerebral hemorrhage classifying, positioning and predicting method based on three-dimensional deep learning model | |
CN108122236B (en) | Iterative fundus image blood vessel segmentation method based on distance modulation loss | |
CN107330437B (en) | Feature extraction method based on convolutional neural network target real-time detection model | |
CN106295124B (en) | The method of a variety of image detecting technique comprehensive analysis gene subgraph likelihood probability amounts | |
CN109829437A (en) | Image processing method, text recognition method, device and electronic system | |
CN110503063B (en) | Falling detection method based on hourglass convolution automatic coding neural network | |
CN106874929B (en) | Pearl classification method based on deep learning | |
CN110532850B (en) | Fall detection method based on video joint points and hybrid classifier | |
CN112926541B (en) | Sleeping post detection method and device and related equipment | |
EP1964028A1 (en) | Method for automatic detection and classification of objects and patterns in low resolution environments | |
CN111954250B (en) | Lightweight Wi-Fi behavior sensing method and system | |
CN111401219B (en) | Palm key point detection method and device | |
CN111488850A (en) | Neural network-based old people falling detection method | |
CN108596087A (en) | A kind of driving fatigue degree detecting regression model based on dual network result | |
CN109101943A (en) | It is a kind of for detecting the machine vision method of Falls Among Old People | |
CN109886170A (en) | A kind of identification of oncomelania intelligent measurement and statistical system | |
CN116740728B (en) | Dynamic acquisition method and system for wafer code reader | |
CN111325166A (en) | Sitting posture identification method based on projection reconstruction and multi-input multi-output neural network | |
CN112465773A (en) | Facial nerve paralysis disease detection method based on human face muscle movement characteristics | |
KR20080079798A (en) | Method of face detection and recognition | |
Raghu et al. | Unsupervised texture classification using vector quantization and deterministic relaxation neural network | |
CN109002758B (en) | Face feature point positioning method, device, equipment and storage medium | |
CN114155952A (en) | Senile dementia illness auxiliary analysis system for elderly people | |
CN114898463A (en) | Sitting posture identification method based on improved depth residual error 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 |