A convolutional neural network teaching aid and system

Abstract

The application discloses a convolutional neural network teaching aid and system, and relates to the technical field of artificial intelligence convolutional neural networks, which comprises a box body, an XY stepping motor module is movably arranged in the box body, a transparent display frame is movably arranged in the middle of the box body, a display is movably arranged at the bottom of the transparent display frame, a switching power supply is movably arranged at the lower right side of the box body, a four-channel signal generator is movably arranged at the lower left side of the box body, and a control panel is movably arranged at the lower side of the box body. Signals generated by the four channels of the four-channel signal generator are sent to a signal processing unit, the processed signals are input into the INPUT6 port of a controller, one end of a driver 1 is connected with the OUTPUT1 of the controller, the other end is connected with the XY stepping motor module, and the XY stepping motor module is used for position control and driving of an X stepping motor; one end of the driver is connected with the OUTPUT2 of the controller, one end of the driver is connected with the OUTPUT3 of the controller, and the driver and the driver are respectively used for position control and driving of double Y stepping motors.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence convolutional neural network, in particular to a convolutional neural network teaching aid and system. BACKGROUND

[0002] At present, the convolutional neural network is a kind of feedforward neural network containing convolution calculation and having a deep structure, is one of the representative algorithms of deep learning, the convolutional neural network is constructed by imitating the visual perception mechanism of biology, the weight parameters in the network are adjusted layer by layer in reverse through the gradient descent method to minimize the loss function, the precision of the network is improved through frequent iteration training, the lower layer is alternately composed of convolution layer and pooling layer, the higher layer is the hidden layer corresponding to the traditional multilayer perceptron and the regression classifier, the core process of the convolutional neural network is to perform convolution on the input image, extract the local features, and map the learned feature representation to the label space of the sample to form a fully connected layer.

[0003] However, the principle and calculation process of the convolutional neural network are still too abstract, the mathematical formula is obscure and harsh, and there is no effective teaching aid to intuitively present the dynamic process of the convolutional neural network, so that scholars are difficult to understand and cannot master the whole process, the convolution operation mentioned in the convolutional neural network and the convolution of the physical signal system have the same name but different concepts, which causes confusion and confusion in the use process of scholars, and is not conducive to the skilled use of the convolutional neural network tool. SUMMARY

[0004] The present application aims to provide a convolutional neural network teaching aid and system to solve the problems in the background art.

[0005] To achieve the above object, the present application provides the following technical scheme: a convolutional neural network teaching aid and system, comprising a box body, an XY stepping motor module is movably installed inside the box body, a transparent display frame is movably installed in the middle of the inside of the box body, a display is movably installed at the bottom end of the transparent display frame, a switching power supply is movably installed at the lower right side of the box body, a four-channel signal generator is movably installed at the lower left side of the box body, and a control panel is movably installed at the lower side of the inside of the box body.

[0006] Preferably, the XY stepping motor module comprises an X stepping motor and a Y stepping motor, the X stepping motor is movably installed inside the box body between the two Y stepping motors, and the two Y stepping motors are movably installed on both sides of the inside of the box body.

[0007] Preferably, the control panel comprises a control switch and a recording switch, the control switch is movably installed at the front left side of the control panel, and the recording switch is located at the right side of the control switch.

[0008] Preferably, the control panel further comprises a reset switch and a CNN / signal system convolution selection switch, the reset switch is movably installed at the middle of the right side of the front end of the control panel, and the CNN / signal system convolution selection switch is located at the right side of the reset switch.

[0009] A convolutional neural network system comprises a controller, the controller is connected to a 220V power supply, a switch K1 corresponding to a control switch is connected to the INPUT1 end of the controller, a switch K2 corresponding to a recording switch is connected to the INPUT2 end of the controller, a switch K3 corresponding to a reset switch is connected to the INPUT3 end of the controller, a selection switch S1 corresponding to a CNN / signal system convolution selection switch is connected to the INPUT4 and INPUT5 ends of the controller, and a display 4 is interconnected with the controller.

[0010] Preferably, the 6-channel generated signal of the four-channel signal generator 6 is sent to a signal processing unit, the processed signal is input to the INPUT6 and INPUT7 ports of the controller, one end of a driver 1 is connected to the OUTPUT1 of the controller, the other end is connected with an XY stepper motor module, and is used for position control and driving of the X stepper motor, one end of a driver 2 is connected to the OUTPUT2 of the controller, one end of a driver 3 is connected to the OUTPUT3 of the controller, and the driver 2 and the driver 3 control the position control and driving of the double-side Y stepper motor respectively.

[0011] Preferably, the on-off switch power supply, the CH1 channel and the CH2 channel of the four-channel signal generator 6 send step signals to a signal processing unit 1, the signal processing unit 1 generates a signal e(τ), and the CH3 channel and the CH4 channel of the four-channel signal generator 6 send step signals to a signal processing unit 2.

[0012] Preferably, the signal processing unit 2 generates a signal h(t-τ), and inputs the e(τ) and h(t-τ) signals to the INPUT6 and INPUT7 ends of the controller respectively, the signals are synthesized, then the selection switch S1 is connected to the INPUT5 gear, and the controller receives the signal to control the display to display the signal system convolution calculation process on the lower side.

[0013] The convolutional neural network algorithm comprises a convolutional neural network forward propagation algorithm, and the convolutional neural network forward propagation algorithm specifically comprises:

[0014]

[0015] wherein the total number of input nodes i is M, w ij is the total number of convolution kernels, a i , after being processed by the lth network layer, a jth output node is generated, and after being subjected to an activation function, a feature node of a jth input channel of a next network layer is formed

[0016] The convolution neural network algorithm comprises a convolution algorithm, and the convolution input algorithm specifically comprises:

[0017] a i = sigma (z) = sigma (a i-1 * W i + b i )

[0018] wherein a i represents the output of convolution, sigma represents a convolution function, a i-1 represents the output of the previous layer, W i represents a connection weight, and b i represents a bias value.

[0019] Compared with the prior art, the present application has the beneficial effects that:

[0020] The present application makes the principle and calculation process of the convolution neural network tend to be visualized, and the mathematical formula becomes easy to understand, and the dynamic process of the convolution neural network is directly presented through effective teaching aids, so that the scholars can quickly understand and master the convolution neural network, the convolution operation and the signal system convolution concept in the convolution neural network are clearly distinguished, the scholars will not be confused in the use process, and the convolution neural network tool is used skillfully, so that the effective teaching aids can directly present the dynamic process of the convolution neural network and clearly distinguish the meaning of convolution in different fields, which is beneficial to the teaching of the convolution neural network and the signal system convolution, and has important significance for using the convolution neural network tool. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The teaching aid overall schematic diagram provided by the embodiment of the present application is shown in the figure;

[0022] Figure 2 The electrical schematic diagram provided by the embodiment of the present application is shown in the figure;

[0023] Figure 3 The initial signal synthesis schematic diagram provided by the embodiment of the present application is shown in the figure;

[0024] Figure 4 The CNN example process schematic diagram of the present application is shown in the figure;

[0025] Figure 5 The signal system convolution operation process schematic diagram provided by the embodiment of the present application is shown in the figure.

[0026] In the figure: 1, box; 2, XY stepper motor module; 201, X stepper motor; 202, Y stepper motor; 3, transparent display frame; 4, display; 5, switching power supply; 6, four-channel signal generator; 7, control switch; 8, recording switch; 9, reset switch; 10, CNN / signal system convolution selection switch; 11, control panel. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0028] Please refer to Figures 1-5 The present application provides a technical solution: a convolutional neural network teaching aid and system, comprising a box 1, an XY stepper motor module 2 is movably installed inside the box 1, a transparent display frame 3 is movably installed in the middle of the inside of the box 1, a display 4 is movably installed at the bottom end of the transparent display frame 3, a switching power supply 5 is movably installed at the lower right side of the box 1, a four-channel signal generator 6 is movably installed at the lower left side of the box 1, and a control panel 11 is movably installed at the lower side of the inside of the box 1.

[0029] The XY stepper motor module 2 comprises an X stepper motor 201 and two Y stepper motors 202, the X stepper motor 201 is movably installed inside the box 1 between the two Y stepper motors 202, and the two Y stepper motors 202 are movably installed on both sides of the inside of the box 1;

[0030] The control panel 11 comprises a control switch 7 and a recording switch 8, the control switch 7 is movably installed at the front left side of the control panel 11, and the recording switch 8 is located to the right of the control switch 7;

[0031] The control panel 11 further comprises a reset switch 9 and a CNN / signal system convolution selection switch 10, the reset switch 9 is movably installed at the middle of the front right side of the control panel 11, and the CNN / signal system convolution selection switch 10 is located to the right of the reset switch 9;

[0032] The controller is connected to a 220V power supply, the switch K1 corresponds to the control switch 7 and is connected to the INPUT1 end of the controller, the switch K2 corresponds to the recording switch 8 and is connected to the INPUT2 end of the controller, the switch K3 corresponds to the reset switch 9 and is connected to the INPUT3 end of the controller, the selection switch S1 corresponds to the CNN / signal system convolution selection switch 10 and is connected to the INPUT4 and INPUT5 ends of the controller with double contacts, and the display 4 is interconnected with the controller;

[0033] One end of the driver 1 is connected with the OUTPUT1 of the controller, and the other end is connected with the XY step motor module 2, which is used for the position control and driving of the X step motor 201; one end of the driver 2 is connected with the OUTPUT2 of the controller, and one end of the driver 3 is connected with the OUTPUT3 of the controller; the driver 2 and the driver 3 control the position control and driving of the double-side Y step motor 202 respectively.

[0034] The four channels of the four-channel signal generator 6 generate signals and send the signals to the signal processing unit; the processed signals are input to the INPUT6 and INPUT7 ports of the controller.

[0035] The switch power supply 5 and the four-channel signal generator 6 are turned on; the CH1 channel and the CH2 channel of the four-channel signal generator 6 send step signals to the signal processing unit 1; the signal processing unit 1 generates a signal e(τ); the CH3 channel and the CH4 channel of the four-channel signal generator 6 send step signals to the signal processing unit 2.

[0036] The signal processing unit 2 generates a signal h(t-τ) and inputs the signals e(τ) and h(t-τ) to the INPUT6 and INPUT7 ports of the controller respectively; the signals are synthesized; then the selection switch S1 is turned to the INPUT5 gear; the controller receives the signals and controls the display 4 to display the signal system convolution calculation process on the lower side.

[0037] The convolution neural network algorithm includes a convolution neural network forward propagation algorithm, and the convolution neural network forward propagation algorithm specifically includes:

[0038]

[0039] wherein the total number of input nodes i is M, w ij is the total number of convolution kernels, a i represents a node a

[0040] The convolution neural network algorithm includes a convolution algorithm, and the convolution input algorithm specifically includes:

[0041] a i = σ(z) = σ(a i-1 * W i + b i )

[0042] wherein a i represents the output of convolution, σ represents a convolution function, a i-1 represents the output of the previous layer, W i represents a connection weight, and b iRepresentative bias values.

[0043] Working principle: the teaching aid of the application is composed of a control panel 11, an XY stepping motor module 2, a filter-acting transparent display frame 3, a display 4, a four-channel signal generator 6, a switching power supply 5 and a box 1, wherein the control panel 11 includes a control switch 7, a recording switch 8, a reset switch 9 and a CNN / signal convolution selection switch 10, through the CNN / signal system convolution selection switch 10, the comparison of the convolution process in the fields of image processing and signal system can be realized, wherein the XY stepping motor module 2 realizes the XY direction movement of the filter-acting transparent display frame 3. The single platform in the X direction fixes the filter-acting transparent display frame 3, the platform moves left and right, the two platforms on the Y direction are synchronous and in the same position, the platform drives the single platform in the X direction to move up and down, wherein the transparent filter-acting transparent display frame 3 is fixed on the single platform in the X direction by a clip and moves up and down, the purpose is to make the content in the transparent filter-acting transparent display frame 3 correspond to the original image in the display 4, the XY stepping motor module moves according to the track from left to right and from top to bottom, and the step distance of each movement is the length of one grid of the original image. The four-channel signal generator 6 generates time domain waveforms, and the signal system convolution calculation waveforms are synthesized through a signal processing unit, the synthesized signal is displayed through the display 4, and the signal convolution calculation process is simulated, wherein the display 4 is divided into three parts, the left upper side displays the image CNN convolution process, the right upper side displays the convolution result of each step, and the lower side displays the convolution process and the convolution calculation result of the signal system convolution, wherein the switching power supply 5 is used for power supply, the switches in the control panel 11 include the functions of switching the display 4, recording data and resetting the stepping motor module, the stepping motor module is placed outside the display 4, and the filter-acting transparent display frame 3 is placed in front of the display 4, the teaching aid can select the calculation process and result of the image CNN convolution and the signal system convolution, each step is executed through button confirmation, and the intuitive presentation of the process and the result are highlighted;

[0044] When the case carries out CNN convolution dynamic process demonstration, switch on the switch power supply 5, switch on the selection switch S1, and select the CNN / signal system convolution selection switch 10 to the INPUT4 gear, the controller receives the signal, and displays the initial image input to the controller on the upper left side of the display 4, switch on the switch K1 corresponding to the control switch 7, the controller transmits the driving signal to the driver, the driver receives the signal and drives the stepping motor to drive the transparent filter action transparent display frame 3 to move on the display 4 from left to right and from top to bottom in steps, display the CNN convolution three-dimensional process, and the controller carries out CNN convolution calculation once every step, when the convolution process is completely calculated, switch off the switch K1 corresponding to the control switch 7, the controller displays the final calculation result on the upper right side of the display 4, switch on the switch K2 corresponding to the recording switch 8, the controller stores the calculation result, switch on the switch K3 corresponding to the reset switch 9, the controller transmits the driving signal to the driver, the driver receives the signal and drives the stepping motor to drive the filter action transparent display frame 3 to return to the initial position;

[0045] When the case carries out CNN pooling dynamic process demonstration, the display 4 displays the stored convolution calculation result on the upper left side, switch on the switch K1 corresponding to the control switch 7, the controller transmits the driving signal to the driver, the driver receives the signal and drives the stepping motor to drive the transparent filter action transparent display frame 3 to move on the display 4 in steps from left to right and from top to bottom, the controller carries out CNN pooling calculation once every step, when the pooling process is completely calculated, switch off the switch K1 corresponding to the control switch 7, the controller displays the final calculation result on the upper right side of the display 4, switch on the switch K2 corresponding to the recording switch 8, the controller stores the pooling calculation result;

[0046] When the case carries out CNN full connection dynamic process demonstration, the controller displays the stored pooling result on the upper left side of the display 4, and arranges it into a row in the order from top to bottom and from left to right once, and displays the full connection result on the upper right side of the display 4;

[0047] When the chassis carries out the signal system convolution dynamic process demonstration, turn on the switching power supply 5, the four-channel signal generator 6, the CH1, CH2 channel sends the step signal to the signal processing unit 1, the signal processing unit 1 generates the signal e(τ), the four-channel signal generator 6, the CH3 channel and the CH4 channel send the step signal to the signal processing unit 2, the signal processing unit 2 generates the signal h(t-τ), and e(τ), h(t-τ) signal is input to the controller INPUT6, INPUT7 end respectively, signal synthesis, then turn on the selection switch S1 corresponding to the CNN / signal system convolution selection switch 10 to the INPUT5 gear, the controller receives the signal, controls the display 4 lower side to display the convolution calculation of the signal system first folding process, press the switch K1 corresponding to the control switch 7, at this time the control display 4 lower side displays the shift process of the signal system convolution calculation, press the switch K1 corresponding to the control switch 7 again, at this time the control display 4 lower side displays the multiplication result of the signal system convolution calculation, press the switch K1 corresponding to the control switch 7 again, at this time the control display 4 lower side displays the translation process of the signal system convolution calculation, press the switch K1 corresponding to the control switch 7 again, at this time the control display 4 lower side displays the final result of the signal system convolution calculation.

[0048] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that are not specifically enumerated herein. Numerous such embodiments have been provided for a thorough description of the embodiments and the scope of the present application. Since embodiments of the application can be modified many times and still fall within the scope of the application, like numerals can refer to like elements throughout the description. Moreover, applicant does not intend for any patentable subject matter to be disclaimed.

[0049] While the embodiments of the application have been shown and described herein, it is understood that the application should cover all modifications and alternatives of the embodiments thereof falling within the scope of the application as defined by the appended claims and their equivalents.

Claims

1. A convolutional neural network tutorial comprising a box (1), characterized in that, The box (1) is movably installed with an XY stepping motor module (2), a transparent display frame (3) is movably installed in the middle of the box (1), a display (4) is movably installed at the bottom of the transparent display frame (3), a switching power supply (5) is movably installed at the lower right side of the box (1), and a four-channel signal generator (6) is movably installed at the lower left side of the box (1). A control panel (11) is movably installed at the lower side of the box (1). The XY stepping motor module (2) comprises an X stepping motor (201) and a Y stepping motor (202), the X stepping motor (201) is movably installed in the box (1) between the two Y stepping motors (202), and the two Y stepping motors (202) are movably installed on both sides of the box (1). The control panel (11) comprises a control switch (7) and a recording switch (8), the control switch (7) is movably installed at the front left side of the control panel (11), and the recording switch (8) is located at the right side of the control switch (7). The control panel (11) further comprises a reset switch (9) and a CNN / signal system convolution selection switch (10), the reset switch (9) is movably installed at the middle of the front right side of the control panel (11), and the CNN / signal system convolution selection switch (10) is located at the right side of the reset switch (9). The XY stepping motor module (2) is placed outside the display (4), the transparent display frame (3) for realizing filter effect is placed in front of the display (4), the teaching aid is used to show the calculation process and result of image CNN convolution or signal system convolution, the XY stepping motor module (2) realizes the XY direction movement of the transparent display frame (3) for realizing filter effect, so that the content in the transparent display frame (3) for realizing filter effect forms a corresponding relationship with the original image in the display (4), the XY stepping motor module (2) moves according to the track from left to right and from top to bottom, and the step distance of each movement is the length of one grid of the original image.

2. A convolutional neural network system suitable for use with the convolutional neural network tutorial of claim 1, wherein, The controller is connected to a 220V power supply, the switch K1 corresponds to the control switch (7) and is connected to the INPUT1 end of the controller, the switch K2 corresponds to the recording switch (8) and is connected to the INPUT2 end of the controller, the switch K3 corresponds to the reset switch (9) and is connected to the INPUT3 end of the controller, the selection switch S1 corresponds to the CNN / signal system convolution selection switch (10), the double contacts are respectively connected to the INPUT4 and INPUT5 ends of the controller, and the display (4) is interconnected with the controller. The four channels of the four-channel signal generator (6) generate signals and send them to a signal processing unit, the processed signals are input to the INPUT6 port of the controller, one end of the driver one is connected to the OUTPUT1 of the controller, the other end is connected to the XY stepping motor module (2), and is used for position control and driving of the X stepping motor (201), one end of the driver two is connected to the OUTPUT2 of the controller, one end of the driver three is connected to the OUTPUT3 of the controller, and the driver two and the driver three control the position control and driving of the double-side Y stepping motor (202) respectively. Turn on the switching power supply (5), four-channel signal generator (6) CH1 channel and CH2 channel sends step signal to signal processing unit one, signal processing unit one generates signal e(τ), four-channel signal generator (6) CH3 channel and CH4 channel sends step signal to signal processing unit two, further analysis is carried out through convolution neural network algorithm; The signal processing unit two generates signal h(t-τ), and e(τ), h(t-τ) signals are respectively input to the controller INPUT6, INPUT7 end, signal synthesis, then turn on the selection switch S1 to the INPUT5 gear, the controller receives the signal, controls the lower side of the display (4) to display the signal system convolution calculation process.

Images