Children hyperactivity attention detection training instrument based on non-invasive brain-computer interface

Abstract

The invention discloses a children hyperactivity attention detection training instrument based on a non-invasive brain-computer interface. The children hyperactivity attention detection training instrument comprises a brain wave module, a Bluetooth communication module, an STM32 single-chip microcomputer and a 7-inch RGB touch screen. Wherein the brain wave module, the Bluetooth communication module, the STM32 single chip microcomputer and the 7-inch RGB touch screen are connected in sequence; the brain wave module is used for extracting brain wave original data and inputting the brain wave original data to the STM32 single-chip microcomputer through the Bluetooth module. The STM32 single chip microcomputer is used for processing the brain wave original data and then classifying the brain wave original data, wherein the brain wave original data comprise three attention levels, namely, the attention is not difficult to concentrate for a long time, the attention is difficult to concentrate selectively and the attention is difficult to concentrate highly; the 7-inch RGB touch screen is used for displaying the classification result and the training result; the auxiliary information is provided for doctors.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular to a children's attention deficit hyperactivity disorder attention detection and training device based on a non-invasive brain-computer interface. Background Art

[0002] In modern society, attention deficit is becoming more and more important. With the growing demand for comprehensive medical care, the diagnosis and treatment of attention deficit has gradually received widespread attention and recognition. According to reports, the overall prevalence of Attention Deficit and Hyperactivity Disorder (ADHD) in children in my country is as high as 6%, and there are more than 23 million children and adolescents with ADHD. More than 60% of parents of children with ADHD have reservations about the current drug treatment plan, and the digital evaluation methods of a few attention training software lack a closed control loop. Summary of the invention

[0003] Purpose of the invention: The purpose of the present invention is to provide a children's attention hyperactivity disorder attention detection training instrument based on a non-invasive brain-computer interface, which solves the problem of objectifying and visualizing the attention detection process by introducing brain wave signal analysis technology, reducing the influence of doctors' subjective judgment, and improving the accuracy and consistency of diagnosis; providing auxiliary information for doctors to help doctors more accurately diagnose and treat children's attention disorders, especially ADHD patients in young age.

[0004] Technical solution: The present invention discloses a non-invasive brain-computer interface-based attention detection and training device for children with ADHD, comprising: a brain wave module, a Bluetooth communication module, an STM32 single-chip microcomputer, and a 7-inch RGB touch screen; wherein the brain wave module, the Bluetooth communication module, the STM32 single-chip microcomputer, and the 7-inch RGB touch screen are connected in sequence; the brain wave module is used to extract raw brain wave data and input it into the STM32 single-chip microcomputer through the Bluetooth module; the STM32 single-chip microcomputer is used to process the raw brain wave data and classify it, including three attention levels: not difficult to concentrate for a long time, difficult to concentrate selectively, and difficult to concentrate highly; the 7-inch RGB touch screen is used to display the classification results and training results; wherein the processing of the raw brain wave data comprises the following steps: (1) Transform and reconstruct wavelet packets; (2) Use the AR model to obtain the power spectrum density of frequency domain features; (3) Calculate time domain characteristics, including peak-to-peak value, mean value, and RMS value; (4) The power spectral density of the obtained frequency domain features and the calculated time domain features are subjected to multi-level classification through a decision tree model to obtain the classification results.

[0005] Furthermore, the EEG module extracts raw EEG data including frequency domain features and time domain features. The specific process is as follows: the EEG module sends 513 packets per second to the Bluetooth module, and the packets include small packets and large packets. The format of the small packet is AA AA 04 80 02 xxHigh xxLow xxCheckSum, where AA AA 04 80 02 is constant, and the last three bytes are always changing. xxHigh and xxLow constitute the raw data, and xxCheckSum is the checksum. The calculation formula of the raw data is as follows: rawdata = (xxHigh<<8) | xxLow; if(rawdata>32768){ rawdata =65536;} Where the checksum is calculated as follows: sum = ((0x80 + 0x02 + xxHigh + xxLow)^ 0xFFFFFFFF)&0xFF That is, add up the four bytes after 04, invert them, and then take the lower eight bits; if the calculated sum and xxCheckSum are equal, the packet is correct, and then the accuracy of the data extracted by the EEG module is corrected.

[0006] Furthermore, step (1) is specifically as follows: using the Daubechies wavelet basis and the maximum decomposition level, the original brain wave data is decomposed by wavelet packets, the decomposition process starts from the top layer, and the original signal is gradually decomposed into signals of different frequency sub-bands according to the wavelet decomposition based on the Daubechies wavelet basis; the original signal is decomposed into a tree structure, and each node represents a sub-band signal in a specific frequency range; after the decomposition is completed, the brain wave signal of the specific frequency band is reconstructed, that is, the signal frequency corresponding to alpha, beta, theta, and delta is reconstructed; wherein, the frequency range corresponding to the alpha brain wave is 8-12 Hz, the frequency range corresponding to the beta brain wave is 12-30 Hz, the frequency range corresponding to the theta brain wave is 4-8 Hz, and the frequency range corresponding to the delta brain wave is 1.5-4 Hz. The four brain wave signals are reconstructed according to the frequency ranges of the four electroencephalographic rhythms.

[0007] Furthermore, step (2) is as follows: first, the input EEG signals of four frequency bands are stimulated to establish a linear system, and then the power spectrum is calculated by estimating the AR model parameters.

[0008] Further, the specific process is as follows: Assume that the EEG signal sequence 𝑋 n , the AR model assumes that: ; where a1, a2, …, a p is a real constant, and a p ≠0, The mean is 0 and the variance is The white noise sequence is the EEG signal Considered as a white noise signal, the formula After output; among them, The expression of A(z) is as follows: ; Because X n The autocorrelation function of itself satisfies: ; in, Yes X n The autocorrelation function, especially when 0≤m≤p, further derives the Yule-Walker equation for solving the model parameters: ; in, , 0≤m≤p is the autocorrelation function of Xn, am(m=1,2,3,…,p) is the model parameter, The variance of the EEG signal input to the AR model; By solving the Yule-Walker equation to determine the parameters of the model, we can further deduce , -∞≤m≤+∞, and finally the power spectrum density of the EEG signals in the four frequency bands of alpha, beta, theta, and delta is calculated according to the power spectrum density function. The formula is as follows: .

[0009] Furthermore, step (3) is as follows: The peak-to-peak value is the difference between the maximum and minimum values ​​of the signal in one cycle. For brain wave signals, the discrete sequence is x[n], where n=1,2.3,…,N-1. The formula for calculating the peak-to-peak value is: ; Where x[n] represents a discrete sequence of brain wave signals, where n=1,2.3,…,N-1, and N is the length of the discrete sequence; represents the maximum value obtained in this discrete sequence, It represents the minimum value obtained, and the difference between the two is the peak-to-peak value, which reflects the amplitude range of the brain wave signal fluctuation in one cycle; The mean is the average amplitude of the signal over a period of time; for a brain wave signal sequence, the calculation formula for the mean is: ; Where N is the length of the discrete sequence of brain wave signals; x[n] is the element in the discrete sequence, It means to sum all values ​​from n=0 to n=N-1 and divide by N to get the average amplitude value of the signal value, which represents the average level of the brain wave signal over a period of time; The root mean square is a measure of signal energy. For brain wave signals, the root mean square calculation formula is: ; Where N is the length of the discrete sequence; x[n] is the discrete sequence element of the brain wave signal. Each x[n] is first squared and then summed. , then divide N and take the square root to get the root mean square; the root mean square is used to measure the energy of the signal.

[0010] Furthermore, step (4) is as follows: first, the Gini index of each feature is calculated based on the EEG signal feature data set, and the data set is recursively divided to minimize uncertainty; specifically, in the EEG three-classification problem, let the categories be C1, C2, and C3, and the probabilities of the sample points belonging to these three categories are p 1、 p 2、 p3, then: According to the definition of the Gini index, the Gini index of the probability distribution is: ; For a given EEG feature data sample set D, its Gini index is: ; Where C1 is the sample subset of D that belongs to category C1, C2 is the sample subset of D that belongs to category C2, C3 is the sample subset of D that belongs to category C3, and |D| represents the total number of samples in sample set D; Among them, D can be divided into D1 and D2 according to a certain feature. Then, under the condition of feature A, the Gini index of the set is defined as: .

[0011] Furthermore, the formula for step (4) is as follows: Input feature vector ,in are the time domain features, namely, mean, RMS, and peak-to-peak values. is the frequency domain feature, i.e., the power spectrum density of different brain waves; through decision tree training, a tree structure is obtained, in which each node contains the feature division criteria: ; When performing classification, given a new input sample , which is judged by the decision tree and finally outputs the corresponding attention level category; where y is the predicted attention level, belonging to the category set {low, medium, high}.

[0012] Furthermore, the STM32 microcontroller is also used to design three game animations for targeted training based on the classification results; the flight altitude of the aircraft, whether the mole is hit, and Mario's jump are used as visual feedback of the attention level.

[0013] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: It comprehensively uses a variety of advanced technical means such as wavelet packet transform and reconstruction, AR model to calculate frequency domain feature power spectrum density, and calculate time domain features (peak-to-peak value, mean value and root mean square), to fully extract rich feature information of brain wave signals. Then, the frequency domain and time domain features are multi-classified through a decision tree model. Compared with a single feature or a simple method, this multi-feature fusion classification method can more accurately identify the attention level of children, including the three states of not having difficulty concentrating for a long time, having difficulty selectively concentrating, and having difficulty highly concentrating, providing doctors with more detailed and accurate auxiliary diagnostic information, especially helpful for early diagnosis and intervention of young ADHD patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic diagram of the structure of the present invention; Figure 2 It is the original data extraction diagram of the brain wave of the present invention; Figure 3 It is a schematic diagram of the AR model of the present invention. DETAILED DESCRIPTION

[0015] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0016] The embodiment of the present invention provides a children's attention detection and training instrument for attention deficit hyperactivity disorder based on a non-invasive brain-computer interface, including: an electroencephalogram module, a Bluetooth communication module, an STM32 single-chip microcomputer, and a 7-inch RGB touch screen; wherein the electroencephalogram module, the Bluetooth communication module, the STM32 single-chip microcomputer, and the 7-inch RGB touch screen are connected in sequence; the electroencephalogram module is used to extract raw electroencephalogram data and input it into the STM32 single-chip microcomputer through the Bluetooth module; the STM32 single-chip microcomputer is used to process the raw electroencephalogram data and classify it, including: it is not difficult to concentrate for a long time, it is difficult to concentrate selectively The brainwave module extracts raw brainwave data including frequency domain features and time domain features. The specific process is as follows: The brainwave module sends 513 packets per second to the Bluetooth module, and the packets sent include small packets and large packets. The format of the small packet is AA AA 04 80 02 xxHigh xxLow xxCheckSum, in which AA AA 0480 02 is unchanged, and the last three bytes are always changing. xxHigh and xxLow constitute the raw data, and xxCheckSum is the checksum. The calculation formula of the raw data is as follows: rawdata = (xxHigh<<8) | xxLow; if(rawdata>32768){ rawdata =65536;} Before calculating the raw data, you should check the checksum first. The checksum is calculated as follows: sum = ((0x80 + 0x02 + xxHigh + xxLow)^ 0xFFFFFFFF)&0xFF That is, add up the four bytes after 04, invert them, and then take the lower eight bits; if the calculated sum and xxCheckSum are equal, the packet is correct, and then the accuracy of the data extracted by the EEG module is corrected.

[0017] Processing the raw EEG data includes the following steps: (1) Transform and reconstruct the wavelet packet; specifically, the original brain wave data is decomposed by wavelet packet using the Daubechies wavelet basis and the maximum decomposition level. The decomposition process starts from the top layer and gradually decomposes the original signal into signals of different frequency sub-bands according to specific algorithm rules. The original signal is decomposed into a tree structure, and each node represents a sub-band signal of a specific frequency range. The formula is as follows: ; Where L represents the level of wavelet packet decomposition. n is the index related to the node, which is used to distinguish the positions of different sub-band signals in the decomposition process. t is the time variable, which reflects the change of the signal in the time domain. and It is the wavelet filter coefficient, which is determined according to the selected Daubechies wavelet basis and plays the role of extracting and transforming different frequency components of the original signal during the decomposition process. Used to calculate the coefficients of the low-frequency part, Used to calculate the coefficients of the high-frequency part, and use these coefficients to compare with the signals of different levels and nodes Perform convolution operations to achieve signal decomposition and reconstruction.

[0018] After the decomposition is completed, the brain wave signals of a specific frequency band are reconstructed, that is, they are reconstructed according to the signal frequencies corresponding to alpha, beta, theta, and delta; among them, the frequency range corresponding to alpha brain waves is 8-12 Hz, the frequency range corresponding to beta brain waves is 12-30 Hz, the frequency range corresponding to theta brain waves is 4-8 Hz, and the frequency range corresponding to delta brain waves is 1.5-4 Hz. The four brain wave signals are reconstructed according to the frequency ranges of the four EEG rhythms.

[0019] (2) The power spectrum density of the frequency domain features is obtained using the AR model. First, the input EEG signals of four frequency bands are stimulated to establish a linear system, and then the power spectrum is calculated by estimating the AR model parameters.

[0020] The specific process is as follows: Suppose the EEG signal sequence 𝑋 n , the AR model assumes that: ; where a1, a2, …, a p is a real constant, and a p ≠0, The mean is 0 and the variance is The white noise sequence is the EEG signal Considered as a white noise signal, the formula After output; Figure 3 As shown, The expression of A(z) is as follows: ; Because X n The autocorrelation function of itself satisfies: ; in, Yes X nThe autocorrelation function, especially when 0≤m≤p, further derives the Yule-Walker equation for solving the model parameters: ; in, , 0≤m≤p is the autocorrelation function of Xn, am(m=1,2,3,…,p) is the model parameter, The variance of the EEG signal input to the AR model.

[0021] From the above solution, we can see that as long as we get , 0≤m≤p, we can determine the parameters of the model by solving the Yule-Walker equation, and further deduce , -∞≤m≤+∞, and finally the power spectrum density of the EEG signals in the four frequency bands of alpha, beta, theta, and delta is calculated according to the power spectrum density function. The formula is as follows: ;

[0022] (3) Calculate the time domain characteristics, including peak-to-peak value, mean value, and RMS value; the details are as follows: The peak-to-peak value is the difference between the maximum and minimum values ​​of the signal in one cycle. For brain wave signals, the discrete sequence is X[n], where n=1,2.3,…,N-1. The formula for calculating the peak-to-peak value is: ; The mean is the average amplitude of the signal over a period of time; for a brain wave signal sequence, the calculation formula for the mean is: ; The root mean square is a measure of signal energy. For brain wave signals, the root mean square calculation formula is: ; (4) The power spectral density of the obtained frequency domain features and the calculated time domain features are subjected to multi-level classification through a decision tree model to obtain the classification results. Specifically, in the EEG three-classification problem, let the categories be C1, C2, and C3, and the probabilities of the sample points belonging to these three categories are p and c respectively. 1、 p 2、 p3, then: According to the definition of the Gini index, the Gini index of the probability distribution is: ; For a given EEG feature data sample set D, its Gini index is: ; Where C1 is the sample subset of D that belongs to category C1, C2 is the sample subset of D that belongs to category C2, C3 is the sample subset of D that belongs to category C3, and |D| represents the total number of samples in sample set D; Among them, D can be divided into D1 and D2 according to a certain feature. Then, under the condition of feature A, the Gini index of the set is defined as: ; When building a decision tree to solve the EEG attention three-classification problem, the steps of the CART algorithm are as follows: Input: EEG feature dataset D, feature A, sample number threshold, Gini coefficient threshold; Output: CART decision tree T; S1. For the data set D of the current node, if the number of samples is less than the threshold or there is no feature, the decision subtree is returned and the recursion of the current node stops; S2. Calculate the Gini coefficient of sample set D. If the Gini coefficient is less than the threshold, return to the decision tree subtree and stop recursion at the current node. S3, calculate the Gini coefficient of each feature value of each feature of the current node for the data set D; S4. Select the feature with the smallest Gini coefficient and the corresponding eigenvalue from the calculated Gini coefficients of the eigenvalues ​​of each feature to the data set D. Based on the optimal feature and the optimal eigenvalue, divide the data set into two parts D1 and D2, and establish the left and right nodes of the current node. The data set D of the left node is D1, and the data set D of the right node is D2. S5. Recursively call the previous S1-S4 steps for the left and right child nodes to generate a decision tree.

[0022] Then, according to the gain or Gini index of each feature, the feature that can most effectively segment the data is selected for node division; after each division, the same calculation is continued on the subset until the stopping condition is met, that is, the information gain threshold is less than a preset threshold and all samples of a node in the leaf node purity belong to the same category; after the decision tree training is completed, the new EEG data is classified by the constructed decision tree; among them, each new sample starts from the root node and is judged along the tree structure according to the feature value.

[0023] The formula is as follows: Input feature vector ,in is the time domain feature, i.e., mean, RMS, and peak-to-peak value), is the frequency domain feature, i.e., the power spectrum density of different brain waves; through decision tree training, a tree structure is obtained, in which each node contains the feature division criteria:

[0024] When performing classification, given a new input sample , which is judged by the decision tree and finally outputs the corresponding attention level category; where y is the predicted attention level, belonging to the category set {low, medium, high}.

Claims

1. A children's attention deficit hyperactivity disorder attention detection and training instrument based on a non-invasive brain-computer interface, characterized in that: include: Brain wave module, Bluetooth communication module, STM32 single chip microcomputer, 7-inch RGB touch screen; wherein, the brain wave module, Bluetooth communication module, STM32 single chip microcomputer, 7-inch RGB touch screen are connected in sequence; the brain wave module is used to extract raw brain wave data and input it into the STM32 single chip microcomputer through the Bluetooth module; the STM32 single chip microcomputer is used to process the raw brain wave data and classify it, including: three attention levels: not difficult to concentrate for a long time, difficult to concentrate selectively, and difficult to concentrate highly; the 7-inch RGB touch screen is used to display the classification results and training results; wherein, processing the raw brain wave data includes the following steps: (1) Transform and reconstruct wavelet packets; (2) Use the AR model to obtain the power spectrum density of frequency domain features; (3) Calculate time domain characteristics, including peak-to-peak value, mean value, and RMS value; (4) The power spectral density of the obtained frequency domain features and the calculated time domain features are subjected to multi-level classification through a decision tree model to obtain the classification results.

2. According to claim 1, a non-invasive brain-computer interface-based attention detection and training instrument for children with ADHD, characterized in that: The EEG module extracts raw EEG data including frequency domain features and time domain features. The specific process is as follows: the EEG module sends 513 packets per second to the Bluetooth module. The packets include small packets and large packets. The format of the small packet is AA AA 04 80 02 xxHigh xxLow xxCheckSum. AA AA 04 80 02 is constant, and the last three bytes are always changing. xxHigh and xxLow constitute the raw data, and xxCheckSum is the checksum. The calculation formula of the raw data is as follows: rawdata = (xxHigh << 8) | xxLow; if(rawdata > 32768){ rawdata =65536;} Where the checksum is calculated as follows: sum = ((0x80 + 0x02 + xxHigh + xxLow)^ 0xFFFFFFFF) & 0xFF That is, add up the four bytes after 04, invert them, and then take the lower eight bits; if the calculated sum and xxCheckSum are equal, the packet is correct, and then the accuracy of the data extracted by the EEG module is corrected.

3. According to claim 1, a non-invasive brain-computer interface-based attention detection and training instrument for children with ADHD, characterized in that: Step (1) is as follows: using the Daubechies wavelet basis and the maximum decomposition level, the original brain wave data is decomposed by wavelet packets. The decomposition process starts from the top layer and the wavelet decomposition is performed according to the Daubechies wavelet basis to gradually decompose the original signal into signals of different frequency sub-bands. The original signal is decomposed into a tree structure, and each node represents a sub-band signal of a specific frequency range. After the decomposition is completed, the brain wave signals of a specific frequency band are reconstructed, that is, they are reconstructed according to the signal frequencies corresponding to alpha, beta, theta, and delta; among them, the frequency range corresponding to alpha brain waves is 8-12Hz, the frequency range corresponding to beta brain waves is 12-30Hz, the frequency range corresponding to theta brain waves is 4-8Hz, and the frequency range corresponding to delta brain waves is 1.5-4 Hz. The four brain wave signals are reconstructed according to the frequency ranges of the four electroencephalographic rhythms.

4. The attention detection and training device for children with ADHD based on non-invasive brain-computer interface according to claim 1, characterized in that: Step (2) is as follows: first, the input EEG signals of four frequency bands are stimulated to establish a linear system, and then the power spectrum is calculated by estimating the AR model parameters.

5. A child attention hyperactivity disorder attention detection and training instrument based on non-invasive brain-computer interface according to claim 4, characterized in that: The specific process is as follows: Suppose the EEG signal sequence 𝑋 n , the AR model assumes that: ； where a1, a2, …, a p is a real constant, and a p ≠0, The mean is 0 and the variance is The white noise sequence is the EEG signal Considered as a white noise signal, the formula After output; among them, The expression of A(z) is as follows: ； Because X n The autocorrelation function of itself satisfies: ； in, Yes X n The autocorrelation function, especially when 0≤m≤p, further derives the Yule-Walker equation for solving the model parameters: ； in, , 0≤m≤p is the autocorrelation function of Xn, am(m=1,2,3,…,p) is the model parameter, The variance of the EEG signal input to the AR model; By solving the Yule-Walker equation to determine the parameters of the model, we can further deduce , -∞≤m≤+∞, and finally the power spectrum density of the EEG signals in the four frequency bands of alpha, beta, theta, and delta is calculated according to the power spectrum density function. The formula is as follows: 。 6. The attention detection and training device for children with ADHD based on non-invasive brain-computer interface according to claim 1, characterized in that: Step (3) is as follows: The peak-to-peak value is the difference between the maximum and minimum values ​​of the signal in one cycle; for brain wave signals, its discrete sequence is x[n], where n=1,2.3,…,N-1, and the formula for calculating the peak-to-peak value is: ； Where x[n] represents a discrete sequence of brain wave signals, where n=1,2.3,…,N-1, and N is the length of the discrete sequence; represents the maximum value obtained in this discrete sequence, It represents the minimum value obtained, and the difference between the two is the peak-to-peak value, which reflects the amplitude range of the brain wave signal fluctuation in one cycle; The mean is the average amplitude of the signal over a period of time; for a brain wave signal sequence, the calculation formula for the mean is: ； Where N is the length of the discrete sequence of brain wave signals; x[n] is the element in the discrete sequence, It means to sum all values ​​from n=0 to n=N-1 and divide by N to get the average amplitude value of the signal value, which represents the average level of the brain wave signal over a period of time; The root mean square is a measure of signal energy. For brain wave signals, the root mean square calculation formula is: ； Where N is the length of the discrete sequence; x[n] is the discrete sequence element of the brain wave signal. Each x[n] is first squared and then summed. , then divide N and take the square root to get the root mean square; the root mean square is used to measure the energy of the signal.

7. The attention detection and training device for children with ADHD based on non-invasive brain-computer interface according to claim 1, characterized in that: Step (4) is as follows: First, the Gini index of each feature is calculated based on the EEG signal feature data set, and the data set is recursively divided to minimize uncertainty; specifically, in the EEG three-classification problem, let the categories be C1, C2, and C3, and the probabilities of the sample points belonging to these three categories are p 1、 p 2、 p3, then: According to the definition of the Gini index, the Gini index of the probability distribution is: ； For a given EEG feature data sample set D, its Gini index is: ； Where C1 is the sample subset of D that belongs to category C1, C2 is the sample subset of D that belongs to category C2, C3 is the sample subset of D that belongs to category C3, and |D| represents the total number of samples in sample set D; Among them, D is divided into D1 and D2 according to a certain feature. Under the condition of feature A, the Gini index of the set is defined as: 。 8. The attention detection and training device for children with ADHD based on non-invasive brain-computer interface according to claim 7, characterized in that: The formula for step (4) is as follows: Input feature vector ,in are the time domain features, namely, mean, RMS, and peak-to-peak values. is the frequency domain feature, i.e., the power spectrum density of different brain waves; through decision tree training, a tree structure is obtained, in which each node contains the feature division criteria: ； When performing classification, given a new input sample , which is judged by the decision tree and finally outputs the corresponding attention level category; where y is the predicted attention level, belonging to the category set {low, medium, high}.

9. The attention detection and training device for children with ADHD based on non-invasive brain-computer interface according to claim 1, characterized in that: The STM32 microcontroller is also used to design three game animations for targeted training based on the classification results; the flight altitude of the aircraft, whether the mole is hit, and Mario's jump are used as visual feedback on the attention level.

Images