阅读说明：本技术 具有坐姿提醒、自关闭和亮度调节功能的嵌入式ai台灯 (Embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions ) 是由 李卫东 刘平涛 张招 罗博文 于 2021-07-14 设计创作，主要内容包括：具有坐姿提醒、自关闭和亮度调节功能的嵌入式AI台灯,其摄像单元实时采集用户图像；语音输出单元将坐姿检测结果实时向用户提醒；FPGA处理单元的像素值计算模块检测台灯的光照亮度；用户检测模块检测台灯周围是否存在用户；坐姿检测模块对用户的坐姿进行检测；当坐姿检测模块检测到用户坐姿不正确时,将检测结果通过语音输出单元反馈给用户以进行坐姿提醒；当像素值计算模块检测台灯的光照亮度超出预设范围时进行台灯的光源亮度调节；当台灯周围用户不存在的状态下,若用户检测模块连续检测用户不存在的次数超过预设值,则关闭台灯。本发明能根据房间黑暗程度,自动调节光照亮度,用户忘记情况下自动关闭,坐姿错误的情况下提醒。(The embedded AI desk lamp has the functions of sitting posture reminding, self-closing and brightness adjustment, and the camera unit of the embedded AI desk lamp collects images of a user in real time; the voice output unit reminds the user of the sitting posture detection result in real time; a pixel value calculating module of the FPGA processing unit detects the illumination brightness of the desk lamp; the user detection module detects whether a user exists around the desk lamp; the sitting posture detection module detects the sitting posture of the user; when the sitting posture detection module detects that the sitting posture of the user is incorrect, the detection result is fed back to the user through the voice output unit to remind the user of the sitting posture; when the pixel value calculation module detects that the illumination brightness of the desk lamp exceeds a preset range, adjusting the light source brightness of the desk lamp; when the user around the desk lamp does not exist, if the number of times of continuous detection of the user that the user does not exist by the user detection module exceeds a preset value, the desk lamp is turned off. The invention can automatically adjust the illumination brightness according to the darkness degree of a room, automatically close the room when a user forgets, and remind the user of a wrong sitting posture.)

1. The embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions comprises a lamp holder (1), a rotating shaft (2), a lamp post (3) and a base (4), wherein one end of the rotating shaft (2) is rotatably connected with the lamp holder (1), the other end of the rotating shaft (2) is rotatably connected with the lamp post (3), and the bottom of the lamp post (3) is connected with the base (4); the device is characterized by also comprising a camera unit (5), a voice output unit (6) and an FPGA processing unit (7); the camera unit (5) is electrically connected with the FPGA processing unit (7), and the camera unit (5) is used for acquiring a user image in real time; the voice output unit (6) is electrically connected with the FPGA processing unit (7), and the voice output unit (6) is used for reminding a user of a sitting posture detection result in real time;

the FPGA processing unit (7) is provided with a pixel value calculating module (8), a user detecting module (9) and a sitting posture detecting module (10); the pixel value calculating module (8) is used for detecting the illumination brightness of the desk lamp; the user detection module (9) is used for detecting whether a user exists around the desk lamp; the sitting posture detection module (10) is used for detecting the sitting posture of a user;

When the sitting posture detection module (10) detects that the sitting posture of the user is incorrect, the detection result is fed back to the user through the voice output unit (6) to remind the user of the sitting posture; when the pixel value calculating module (8) detects that the illumination brightness of the desk lamp exceeds a preset range, the light source brightness of the desk lamp is adjusted; when the user around the desk lamp does not exist, if the number of times of continuously detecting that the user does not exist by the user detection module (9) exceeds a preset value, the desk lamp is turned off.

2. The embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions as claimed in claim 1, wherein the base (4) is provided with an on-off switch (11), a sitting posture monitoring key (12) and a brightness adjusting key (13), the on-off switch (11) is electrically connected with the FPGA processing unit (7), and the on-off switch (11) is used for controlling the power supply of the desk lamp to be turned on and off; the sitting posture reminding key is electrically connected with the FPGA processing unit (7) and is used for controlling a sitting posture detection function switch of the sitting posture detection module (10); the brightness adjusting key (13) is electrically connected with the FPGA processing unit (7), and the brightness adjusting key (13) is used for adjusting brightness of the manual desk lamp.

3. The embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions as claimed in claim 1, wherein the input of the FPGA processing unit (7) comprises preset illumination brightness, detection time interval, the number of times of continuous absence of user, the number of times of continuous incorrect sitting posture and image data input by the camera unit (5).

4. The embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions as claimed in claim 2, characterized in that the camera unit (5) comprises a camera connected to the side of the lamp post (3), and the image obtained by the camera is sent to the FPGA processing unit (7) in the base (4) for processing.

5. The embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions as claimed in claim 2, further comprising a resistor network with relay control, wherein the FPGA processing unit (7) controls the resistance by controlling the switch of the relay to provide preset power to the light source.

6. The embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions as claimed in claim 2, characterized in that the voice output unit (6) is equipped with a speaker, which is mounted on the lamp post (3), base (4) or lamp head (1).

7. The embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions as claimed in claim 1, characterized in that the pixel value calculating module (8) calculates the gray level average of all pixels by using the FPGA processing unit (7).

8. The embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions as claimed in claim 1 wherein the user detection module (9) utilizes the FPGA processing unit (7) to complete a classified deep neural network model A, the result of which comprises user presence and user absence;

The sitting posture detection module (10) completes a classified deep neural network model B by utilizing the FPGA processing unit (7), and the result of the deep neural network model B comprises sitting posture correction and sitting posture error.

9. The embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions as claimed in claim 8, wherein the user sitting posture detection method of the user detection module (9) comprises:

initializing to obtain the maximum allowable sitting error times set by a user; image data acquired from the camera unit (5) is used as input of the deep neural network model A and the deep neural network model B; judging whether a user exists or not by using a classification result of the deep neural network model A, and detecting the sitting posture state of the user by using the deep neural network model B under the condition that the user is in the monitoring area;

the counting is carried out once every preset time interval, and counting is carried out when the sitting posture of the user is judged to be incorrect;

and when the sitting times of the user continuously accumulate to reach a preset value by combining the maximum allowable sitting error times set by the user, sending alarm information to the user through a voice output unit (6).

10. The embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions as claimed in claim 8 wherein the deep neural network model A adopts VGG-like network model structure with RGB three-channel picture as input; after the picture is subjected to convolution, pooling, activation and full-connection combined operation, a vector representing two categories of users and no user is obtained, and the position of the maximum value of the vector is a judgment result;

The deep neural network model B adopts a VGG-type network model structure, and takes RGB three-channel pictures as input; after the picture is subjected to convolution, pooling, activation and full-connection combined operation, two classification vectors of correct and wrong sitting postures of the user are obtained, and the position of the maximum value of the vector is a judgment result;

the deep neural network model A and the deep neural network model B share a preset weight value of the neural network;

and the convolution operation of the deep neural network model A and the deep neural network model B adopts multiplication convolution or addition convolution.

Technical Field

The invention belongs to the technical field of desk lamps, and particularly relates to an embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions.

Background

The long time of using eyes and the incorrect sitting posture during writing are the important reasons for the myopia of the teenagers. Meanwhile, the incorrect writing posture can also cause the problems of spinal curvature and the like, and influence the healthy development of teenagers.

The desk lamp is a necessary product in daily families, is usually placed on a writing desk for illumination, and is convenient for reading, learning and working. When the user uses the desk lamp, different sitting postures can be adopted according to the living habits of the user, once the sitting postures are poor and become the personal habits of the user, serious influences can be caused on the eyesight and the cervical vertebra of the user, and the myopia can be directly caused. Meanwhile, in the long-time use process, a user often concentrates on learning and neglects the change of light rays in a room along with time, which is easy to happen in the case of long learning time of teenagers, for example, the writing operation time span of teenagers is from evening to night, the light rays are continuously reduced along with the time lapse, and the illumination brightness of the desk lamp is continuously increased to meet the requirement of proper illumination intensity. In addition, after the user uses the desk lamp, the user often forgets to turn off the power supply, which causes waste of electric energy.

To the position of sitting and remind the mode, the prior art appears "intelligence myopia prevention desk lamp", its basic principle is at the fixed ultrasonic sensor of desk edge, surveys the distance of head and sensor, if the distance is too close to, the desk lamp takes place to warn and corrects the position of sitting. However, the device has the inherent disadvantage that the distance threshold judgment needs personalized adjustment due to the differentiation of the height and the like, and meanwhile, the ultrasonic wave radiates the human body, so that the sensor and the like are easily shielded; whether the sitting posture mode is correct or not is judged in the mode that a user wears the sensor equipment through partial products, the using mode is complicated, and the user experience is not good.

To desk lamp self-closing and adjust luminance on, prior art often adjusts the human body through installing corresponding sensor and responding to and the illumination intensity mode of responding to, if carry out human response through infrared sensing, adjust through the photo resistance mode.

The sitting posture detection can be divided into classification judgment of correct sitting postures and incorrect sitting postures, and the artificial intelligence has good application in the fields of image classification, image recognition, image segmentation and the like. Meanwhile, compared with the traditional algorithm, the algorithm of the artificial intelligence on the image processing is less influenced by factors such as illumination intensity, contrast and the like. And the intelligent desk lamp part sends the taken image into the server for sitting posture detection by installing a camera. However, uploading user data to a server has disadvantages such as poor real-time performance and poor privacy, and is particularly problematic in privacy and thus not acceptable in the market.

In summary, a new intelligent desk lamp with functions of sitting posture reminding, automatic turn-off, automatic brightness adjustment and privacy protection is needed.

Disclosure of Invention

Therefore, the embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions provided by the invention can automatically adjust the illumination brightness according to the darkness of a room, automatically close the desk lamp when a user forgets to close the desk lamp, save electric energy, remind the user when the user has a wrong sitting posture, and pay attention to the privacy protection of the user.

In order to achieve the above purpose, the invention provides the following technical scheme: an embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions comprises a lamp holder, a rotating shaft, a lamp post and a base, wherein one end of the rotating shaft is rotatably connected with the lamp holder, the other end of the rotating shaft is rotatably connected with the lamp post, and the bottom of the lamp post is connected with the base; the device also comprises a camera shooting unit, a voice output unit and an FPGA processing unit; the camera shooting unit is electrically connected with the FPGA processing unit and is used for collecting user images in real time; the voice output unit is electrically connected with the FPGA processing unit and is used for reminding a user of a sitting posture detection result in real time;

The FPGA processing unit is provided with a pixel value calculating module, a user detecting module and a sitting posture detecting module; the pixel value calculation module is used for detecting the illumination brightness of the desk lamp; the user detection module is used for detecting whether users exist around the desk lamp; the sitting posture detection module is used for detecting the sitting posture of the user;

when the sitting posture detection module detects that the sitting posture of the user is incorrect, the detection result is fed back to the user through the voice output unit to remind the user of the sitting posture; when the pixel value calculation module detects that the illumination brightness of the desk lamp exceeds a preset range, adjusting the light source brightness of the desk lamp; when the user around the desk lamp does not exist, if the number of times that the user detection module continuously detects that the user does not exist exceeds a preset value, the desk lamp is turned off.

As a preferred scheme of the embedded AI table lamp with sitting posture reminding, self-closing and brightness adjusting functions, an on-off key, a sitting posture monitoring key and a brightness adjusting key are arranged on the base, the on-off key is electrically connected with the FPGA processing unit, and the on-off key is used for controlling the power supply of the table lamp to be turned on and off; the sitting posture reminding key is electrically connected with the FPGA processing unit and used for controlling a sitting posture detection function switch of the sitting posture detection module; the brightness adjusting key is electrically connected with the FPGA processing unit and used for adjusting brightness of the manual desk lamp.

As a preferred scheme of the embedded AI table lamp with sitting posture reminding, self-closing and brightness adjusting functions, the input of the FPGA processing unit includes preset illumination brightness, detection time interval, continuous absence times of users, continuous incorrect times of sitting postures and image data input by the camera unit.

As a preferred scheme of the embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions, the camera unit comprises a camera connected to the side part of the lamp post, and an image acquired by the camera is sent to the FPGA processing unit in the base for processing.

The embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions further comprises a resistor network with a relay control function, and the FPGA processing unit controls the resistor size by controlling the switch of the relay so as to provide preset power for the light source.

As a preferable scheme of the embedded AI table lamp with sitting posture reminding, self-closing and brightness adjusting functions, the voice output unit is provided with a speaker, and the speaker is mounted on a lamp post, a base or a lamp cap.

As a preferred scheme of the embedded AI table lamp with sitting posture reminding, self-closing and brightness adjusting functions, the pixel value calculating module adopts an FPGA processing unit to calculate the gray average value of all pixels.

As a preferred scheme of the embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions, the user detection module utilizes the FPGA processing unit to complete a classified deep neural network model A, and the result of the deep neural network model A comprises the existence of a user and the nonexistence of the user;

the sitting posture detection module utilizes the FPGA processing unit to complete a classified deep neural network model B, and the result of the deep neural network model B comprises sitting posture correction and sitting posture error.

As a preferred scheme of the embedded AI table lamp with sitting posture reminding, self-closing and brightness adjusting functions, the user sitting posture detection method of the user detection module comprises the following steps:

initializing to obtain the maximum allowable sitting error times set by a user; the image data obtained from the camera unit is used as the input of a deep neural network model A and a deep neural network model B; judging whether a user exists or not by using a classification result of the deep neural network model A, and detecting the sitting posture state of the user by using the deep neural network model B under the condition that the user is in the monitoring area;

the counting is carried out once every preset time interval, and counting is carried out when the sitting posture of the user is judged to be incorrect;

And when the sitting times of the user continuously accumulate to reach a preset value by combining the maximum allowable sitting error times set by the user, sending alarm information to the user through the voice output unit.

As a preferred scheme of the embedded AI desk lamp with sitting posture reminding, self-closing and brightness adjusting functions, the deep neural network model A adopts a VGG-like network model structure, and takes an RGB three-channel picture as input; after the picture is subjected to convolution, pooling, activation and full-connection combined operation, a vector representing two categories of users and no user is obtained, and the position of the maximum value of the vector is a judgment result;

the deep neural network model B adopts a VGG-type network model structure, and takes RGB three-channel pictures as input; after the picture is subjected to convolution, pooling, activation and full-connection combined operation, two classification vectors of correct and wrong sitting postures of the user are obtained, and the position of the maximum value of the vector is a judgment result;

the deep neural network model A and the deep neural network model B share a preset weight value of the neural network;

the convolution operation of the deep neural network model A and the deep neural network model B adopts multiplication convolution or addition convolution.

The invention has the following advantages: the device is provided with a camera shooting unit, a voice output unit and an FPGA processing unit; the camera unit is electrically connected with the FPGA processing unit and is used for acquiring a user image in real time; the voice output unit is electrically connected with the FPGA processing unit and is used for reminding a user of a sitting posture detection result in real time; the FPGA processing unit is provided with a pixel value calculating module, a user detecting module and a sitting posture detecting module; the pixel value calculation module is used for detecting the illumination brightness of the desk lamp; the user detection module is used for detecting whether a user exists around the desk lamp; the sitting posture detection module is used for detecting the sitting posture of the user; when the sitting posture detection module detects that the sitting posture of the user is incorrect, the detection result is fed back to the user through the voice output unit to remind the user of the sitting posture; when the pixel value calculation module detects that the illumination brightness of the desk lamp exceeds a preset range, adjusting the light source brightness of the desk lamp; when the user around the desk lamp does not exist, if the number of times of continuously detecting that the user does not exist by the user detection module exceeds a preset value, the desk lamp is turned off. The invention has three functions of sitting posture reminding, automatic closing and automatic brightness adjustment, has the advantages of excellent performance, complete functions and the like, only needs one camera shooting unit and one FPGA processing unit for the three functions, does not need to install different sensors for each function, and has simple and efficient scheme; the sitting posture image of the user is directly acquired through the camera unit and processed in the local FPGA processing unit without being uploaded to a cloud server, so that the privacy of the user is protected, meanwhile, a good supervision effect on the sitting posture of the user can be realized, the user is promoted to keep a good sitting posture, and the body health is protected; the automatic closing function is provided, and the automatic closing can be realized under the condition that a user forgets to close the automatic closing function, so that the energy and the electricity are saved; the automatic brightness adjusting device has an automatic brightness adjusting function, and under the condition that external light changes along with time, illumination brightness is automatically changed, so that extremely good experience is brought to a user.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below. It should be apparent that the drawings in the following description are merely exemplary, and that other implementation drawings can be derived from the drawings provided by those of ordinary skill in the art without any creative effort.

The structures, ratios, sizes, and the like shown in the present specification are only used for matching with the contents disclosed in the specification, so that those skilled in the art can understand and read the present invention, and do not limit the conditions under which the present invention can be implemented, so that the present invention has no technical significance.

Fig. 1 is a schematic structural diagram of an embedded AI table lamp with sitting posture reminding, self-closing and brightness adjusting functions according to an embodiment of the present invention;

fig. 2 is a schematic diagram of an FPGA processing unit of an embedded AI table lamp with sitting posture reminding, self-closing and brightness adjusting functions according to an embodiment of the present invention;

Fig. 3 is a diagram illustrating the steps of detecting the presence and the sitting posture of a user with an embedded AI table lamp having functions of sitting posture reminding, self-closing and brightness adjustment according to an embodiment of the present invention;

FIG. 4 is a schematic structural diagram of a deep neural network model A for detecting the presence of a user according to an embodiment of the present invention;

fig. 5 is a schematic structural diagram of a deep neural network model B for detecting a sitting posture of a user according to an embodiment of the present invention;

fig. 6 is a schematic diagram of the multiplication and addition calculations of convolution provided in the embodiment of the present invention.

In the figure, 1, a lamp cap; 2. a rotating shaft; 3. a lamp post; 4. a base; 5. an image pickup unit; 6. a voice output unit; 7. an FPGA processing unit; 8. a pixel value calculation module; 9. a user detection module; 10. a sitting posture detection module; 11. an on-off key; 12. a sitting posture monitoring key; 13. and a brightness adjusting key.

Detailed Description

The present invention is described in terms of particular embodiments, other advantages and features of the invention will become apparent to those skilled in the art from the following disclosure, and it is to be understood that the described embodiments are merely exemplary of the invention and that it is not intended to limit the invention to the particular embodiments disclosed. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Referring to fig. 1 and 2, an embedded AI table lamp with sitting posture reminding, self-closing and brightness adjusting functions is provided, and comprises a lamp holder 1, a rotating shaft 2, a lamp post 3 and a base 4, wherein one end of the rotating shaft 2 is rotatably connected with the lamp holder 1, the other end of the rotating shaft 2 is rotatably connected with the lamp post 3, and the bottom of the lamp post 3 is connected with the base 4; the device also comprises a camera unit 5, a voice output unit 6 and an FPGA processing unit 7; the camera unit 5 is electrically connected with the FPGA processing unit 7, and the camera unit 5 is used for acquiring a user image in real time; the voice output unit 6 is electrically connected with the FPGA processing unit 7, and the voice output unit 6 is used for reminding a user of a sitting posture detection result in real time;

the FPGA processing unit 7 is provided with a pixel value calculating module 8, a user detecting module 9 and a sitting posture detecting module 10; the pixel value calculating module 8 is used for detecting the illumination brightness of the desk lamp; the user detection module 9 is used for detecting whether a user exists around the desk lamp; the sitting posture detection module 10 is used for detecting the sitting posture of the user;

when the sitting posture detection module 10 detects that the sitting posture of the user is incorrect, the detection result is fed back to the user through the voice output unit 6 to remind the user of the sitting posture; when the pixel value calculation module 8 detects that the illumination brightness of the desk lamp exceeds a preset range, adjusting the light source brightness of the desk lamp; when the user around the desk lamp does not exist, if the number of times of continuously detecting that the user does not exist by the user detection module 9 exceeds a preset value, the desk lamp is turned off.

In this embodiment, the base 4 is provided with an on-off key 11, a sitting posture monitoring key 12 and a brightness adjusting key 13, the on-off key 11 is electrically connected with the FPGA processing unit 7, and the on-off key 11 is used for controlling the power supply of the desk lamp to be turned on and off; the sitting posture reminding key is electrically connected with the FPGA processing unit 7 and used for controlling a sitting posture detection function switch of the sitting posture detection module 10; the brightness adjusting key 13 is electrically connected with the FPGA processing unit 7, and the brightness adjusting key 13 is used for adjusting brightness of the manual desk lamp. The input of the FPGA processing unit 7 comprises preset illumination brightness, detection time interval, the continuous absence times of the user, the continuous incorrect sitting posture times and image data input by the camera unit 5.

In this embodiment, the camera unit 5 includes a camera connected to a side portion of the lamp post 3, and an image acquired by the camera is sent to the FPGA processing unit 7 in the base 4 for processing. The voice output unit 6 is provided with a speaker, which is mounted on the lamp post 3, the base 4 or the lamp cap 1. The LED lamp also comprises a resistor network with a relay control, and the FPGA processing unit 7 controls the resistance through controlling the switch of the relay so as to provide preset power for the light source. The calculation method of the pixel value calculation module 8 is to calculate the average value of the gray levels of all the pixels by using the FPGA processing unit 7.

Referring to fig. 3, in this embodiment, the method for detecting the user sitting posture of the user detection module 9 includes:

initializing to obtain the maximum allowable sitting error times set by a user; the image data acquired from the imaging unit 5 is used as the input of the deep neural network model a and the deep neural network model B; judging whether a user exists or not by using a classification result of the deep neural network model A, and detecting the sitting posture state of the user by using the deep neural network model B under the condition that the user is in the monitoring area;

the counting is carried out once every preset time interval, and counting is carried out when the sitting posture of the user is judged to be incorrect;

and when the sitting posture times of the user continuously accumulate to reach a preset value by combining the maximum allowable sitting posture times set by the user, alarm information is sent to the user through the voice output unit 6 so as to avoid that the user experience is influenced by too frequent reminding.

Specifically, the detection method for judging whether the user exists and the sitting posture is correct specifically comprises the following steps:

step S1, a sitting posture reminding function is started through a desk lamp key;

step S2, setting detection time interval, allowing the user to continuously have no times, and allowing the sitting posture to continuously have incorrect times;

Step S3, at intervals, transmitting image data acquired from the camera of the camera unit 5 to the FPGA processing unit 7, detecting whether a user exists by using the user detection module 9 in the FPGA processing unit 7, counting the times of continuously detecting the absence of the user, and disconnecting the power supply of the desk lamp when the accumulated times reach a preset value;

and step S4, when the user is judged to exist, the sitting posture detection module 10 is used for detecting sitting postures, the times of continuous sitting posture errors are counted, and when the times of the sitting posture errors of the user reach a preset value, a result is output to the voice output module for prompting the sitting posture errors.

To further improve the detection accuracy, step S3 requires:

step S31, the desk lamp needs to be placed in front of the user, and the included angle between the connection line of the desk lamp and the user and the front direction of the user does not exceed 45 degrees, so that the face of the user can be completely shot and recorded by a camera on the desk lamp;

in step S32, the camera takes one image at a time interval as a target of the arithmetic processing.

In this embodiment, the user detection module 9 completes a classified deep neural network model a by using the FPGA processing unit 7, and a result of the deep neural network model a includes the presence and absence of a user; the deep neural network model A adopts a VGG-type network model structure, and takes RGB three-channel pictures as input; after the picture is subjected to convolution, pooling, activation and full-connection combined operation, a vector representing two categories of users and no user is obtained, and the position of the maximum value of the vector is a judgment result.

Referring to fig. 4, the deep neural network model a inputs image data of 112 × 112 size, the neural network includes 7 convolutions, 4 pooling, 4 activations, and one full join operation, and outputs a 2-dimensional vector.

In this embodiment, the sitting posture detection module 10 completes a classified deep neural network model B by using the FPGA processing unit 7, and the result of the deep neural network model B includes a sitting posture correction and a sitting posture error; the deep neural network model B adopts a VGG-type network model structure, and takes an RGB three-channel picture as input; after the picture is subjected to convolution, pooling, activation and full-connection combined operation, two classification vectors of correct and wrong sitting postures of the user are obtained, and the position of the maximum value of the vector is a judgment result.

Referring to fig. 5, the deep neural network model B inputs image data of size 112 × 112, including 7 convolutions, 4 pooling, 4 activations, and one full join operation, and outputs a 2-dimensional vector.

In this embodiment, the deep neural network model a and the deep neural network model B share a preset weight value of the neural network; the number of parameters in the convolution kernel is greatly reduced (since only one identical convolution kernel is required to move around), reducing the complexity of the network.

Referring to fig. 6, in the present embodiment, the convolution operation of the deep neural network model a and the deep neural network model B adopts a multiplication convolution or an addition convolution. The main purpose of convolution is to extract features from the input image, convolution preserves the spatial relationship between pixels by learning image features using small squares of input data.

In summary, the present invention is provided with a camera unit 5, a voice output unit 6 and an FPGA processing unit 7; the camera unit 5 is electrically connected with the FPGA processing unit 7, and the camera unit 5 is used for acquiring a user image in real time; the voice output unit 6 is electrically connected with the FPGA processing unit 7, and the voice output unit 6 is used for reminding a user of a sitting posture detection result in real time; the FPGA processing unit 7 is provided with a pixel value calculating module 8, a user detecting module 9 and a sitting posture detecting module 10; the pixel value calculating module 8 is used for detecting the illumination brightness of the desk lamp; the user detection module 9 is used for detecting whether a user exists around the desk lamp; the sitting posture detection module 10 is used for detecting the sitting posture of the user; when the sitting posture detection module 10 detects that the sitting posture of the user is incorrect, the detection result is fed back to the user through the voice output unit 6 to remind the user of the sitting posture; when the pixel value calculation module 8 detects that the illumination brightness of the desk lamp exceeds a preset range, adjusting the light source brightness of the desk lamp; when the user around the desk lamp does not exist, if the number of times of continuously detecting that the user does not exist by the user detection module 9 exceeds the preset value, the desk lamp is turned off. The invention has three functions of sitting posture reminding, automatic closing and automatic brightness adjustment, has the advantages of excellent performance, complete functions and the like, only needs one camera unit 5 and one FPGA processing unit 7 for the three functions, does not need to install different sensors for each function, and has simple and efficient scheme; the sitting posture image of the user is directly acquired through the camera unit 5 and processed in the local FPGA processing unit 7 without being uploaded to a cloud server, so that the privacy of the user is protected, meanwhile, a good supervision effect on the sitting posture of the user can be realized, the user is promoted to keep a good sitting posture, and the body health is protected; the automatic closing function is provided, and the automatic closing can be realized under the condition that a user forgets to close the automatic closing function, so that the energy and the electricity are saved; the automatic brightness adjusting device has an automatic brightness adjusting function, and under the condition that external light rays change along with time, the illumination brightness is automatically changed, so that the user is provided with extremely good experience.

Although the invention has been described in detail above with reference to a general description and specific examples, it will be apparent to one skilled in the art that modifications or improvements may be made thereto based on the invention. Accordingly, such modifications and improvements are intended to be within the scope of the invention as claimed.