Crowd heart health early warning device based on multi-source geographic space-time big data

Abstract

The invention provides a crowd heart health early warning device based on multi-source geographic space-time big data, and the device comprises a heart rate and blood pressure data acquisition device which is used for collecting the heart rate and blood pressure data of a patient; the heart rate and blood pressure vector acquisition device is used for acquiring heart rate and blood pressure vectors according to the heart rate and blood pressure data; the air environment related data acquisition device is used for acquiring air environment data around a holder of the cardiovascular disease early warning device, position data of the holder, road network data and current time data; the environment exposure level determining device is used for determining the environment exposure level of the holder according to the data obtained by the air environment related data obtaining device, the environment exposure vector obtaining device is used for determining environment exposure vectors according to the environment exposure level of the holder, and the vector merging device is used for merging the environment exposure vectors according to the environment exposure vector of the holder. Combining the heart rate and blood pressure vector with the environment exposure vector; and the early warning device is used for carrying out early warning according to the combined vectors.

Description

Technical Field

[0001] The present invention relates to the application of multi-source geographic spatiotemporal big data, and in particular to health warning using multi-source geographic spatiotemporal big data. Background Art

[0002] Heart diseases, including cardiovascular diseases, can seriously threaten the life safety of patients. Although various diagnostic and treatment methods have been developed, they are often detected after the disease has already occurred, that is, patients usually go to the hospital for diagnosis and treatment after the disease has occurred, and sometimes the best time for diagnosis is missed. At present, the early warning of cardiovascular disease cannot meet the heavy demand for cardiovascular health management of residents in reality.

[0003] Spatiotemporal big data is a data set based on a unified spatiotemporal reference (temporal reference system and spatial reference system) where activities are directly or indirectly related to locations in time and space. These data contain a lot of information, but there is currently no good way to use spatiotemporal big data for heart health warnings.

[0004] The description of the background technology is only for facilitating the understanding of the present invention, and does not mean that the description therein has been disclosed in the art. Summary of the invention

[0005] In view of the above situation, the present invention proposes a heart health early warning device based on multi-source geographic spatiotemporal big data, which is used to alleviate or eliminate one or more shortcomings existing in the prior art and at least provide a beneficial option.

[0006] According to one aspect of the present invention, the present invention provides a heart health warning device for people based on multi-source geographic spatiotemporal big data, including: a heart rate and blood pressure data acquisition device for collecting the heart rate and blood pressure data of the patient; a heart rate and blood pressure vector acquisition device for obtaining the heart rate and blood pressure vector according to the heart rate and blood pressure data; an air environment related data acquisition device for obtaining the air environment data around the holder of the cardiovascular disease warning device, the location data of the holder, the road network data, and the current time data; an environmental exposure level determination device for determining the environmental exposure level of the holder according to the data obtained by the air environment related data acquisition device, an environmental exposure vector acquisition device for determining the environmental exposure vector according to the environmental exposure level of the holder, a vector merging device for merging the heart rate and blood pressure vector with the environmental exposure vector; and an early warning device for early warning of heart health according to the merged vector. Terminal devices such as mobile phones and computers receive or monitor the specific health conditions of the device holder.

[0007] According to the present invention, a new feature extraction structure is provided, which includes left and right branches for feature extraction, and distributes weights to the left and right branches to extract more comprehensive and diverse features.

[0008] According to some embodiments of the present invention, combining data from multiple sources such as time and environment of spatiotemporal big data with basic health indicator data of an individual can effectively provide early warning of heart health. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The present invention can be better understood with reference to the accompanying drawings, but the accompanying drawings are merely illustrative and not drawn to scale, and are not intended to limit the scope of protection of the present invention.

[0010] Figure 1 A schematic block diagram of a cardiovascular disease early warning device based on spatiotemporal big data according to an embodiment of the present invention is shown.

[0011] Figure 2 A schematic block diagram of a heart rate and blood pressure vector acquisition device according to an embodiment of the present invention is shown.

[0012] Figure 3 A schematic block diagram of a timing processing layer according to an embodiment of the present invention is shown.

[0013] Figure 4 A schematic block diagram of a cardiovascular disease early warning device based on spatiotemporal big data according to another embodiment of the present invention is shown.

[0014] Figure 5 A schematic block diagram of a cardiovascular disease early warning device based on spatiotemporal big data according to another embodiment of the present invention is shown. DETAILED DESCRIPTION

[0015] The specific embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. These embodiments are exemplary and do not constitute any limitation to the present invention. The accompanying drawings do not show components that are related to the operation of the embodiments of the present invention but are irrelevant to the understanding of the present invention, nor are these components described in the specification. These components may adopt various technologies that are currently known or will be known in the future, all of which are within the scope of protection of the present invention. The specific embodiments of the present invention are not limitations on the claims, and the scope of the patent is defined by the claims. The features in the embodiments of the specification cannot be used to limit the scope of the claims.

[0016] Figure 1 FIG. 1 is a schematic block diagram of a cardiovascular disease early warning device based on spatiotemporal big data according to an embodiment of the present invention. Figure 1As shown, a cardiovascular disease warning device according to one embodiment of the present invention includes a heart rate and blood pressure data acquisition device 100, a heart rate and blood pressure vector acquisition device 200, an air environment related data acquisition device 300, an environmental exposure level determination device 400, an environmental exposure vector acquisition device 500, a vector merging device 600, and a warning device 700.

[0017] According to one embodiment, the heart rate and blood pressure data acquisition device 100 may be a wristband device. The wristband device can collect the patient's heart rate and blood pressure data in real time. There are already mature devices on the market that collect the heart rate and blood pressure of the holder and transmit the collected heart rate and blood pressure. The principles of collecting and transmitting heart rate and blood pressure are also known, so they will not be described in detail here.

[0018] The normal range of heart rate is 60-100 beats / minute. Data below 40 beats / minute or above 120 beats / minute may be abnormal values. The collected blood pressure data includes systolic pressure and diastolic pressure. The normal range is systolic pressure 90-120mmHg and diastolic pressure 60-80mmHg. Data below 80 / 50 or above 180 / 120 may be abnormal values.

[0019] The heart rate and blood pressure vector acquisition device 200 processes the heart rate and blood pressure data acquired by the heart rate and blood pressure data acquisition device 100 to obtain a heart rate and blood pressure vector.

[0020] Figure 2 FIG. 1 is a schematic block diagram of a heart rate and blood pressure vector acquisition device according to an embodiment of the present invention. Figure 2 As shown, according to one embodiment, the heart rate and blood pressure vector acquisition device 200 includes an input layer 210 , a timing processing layer 220 , and a fully connected layer 230 .

[0021] According to one embodiment, the input layer 210 inputs the heart rate and blood pressure data at three moments. These three moments respectively process the time t1 of the starting point l1 on the path L1->L2, the end time t3 of the path l2, and the middle moment t2=t1+(t1+t3) / 2. The input layer 210 collects the heart rate and blood pressure data at three time points on the path from t1 to t3. At t1, the heart rate is heart_ratet1 and the blood pressure is blood_pressuret1. At t2=(t1+t3) / 2, the heart rate is heart_ratet2 and the blood pressure is blood_pressuret2. At t3, the heart rate is heart_ratet3 and the blood pressure is blood_pressuret3. The input layer 210 processes the 6 data into (3,2) sequential data, also known as sequential matrix data:

[0022]

[0023] Then, the input features (heart rate and blood pressure) at each moment are input to the next moment processing unit together with the hidden state h and unit state c of the state of the processing unit at the current moment.

[0024] Therefore, the entire time series processing layer 220 obtains five parameters c1, c2, h1, h2, and h3. c1 is the output of the current layer obtained by processing the heart rate and blood pressure data at time t1 through the activation function, h1 is the hidden output of the heart rate and blood pressure data at time t1 without the activation function processing, similarly, h2 and h3 are the hidden outputs of the current layer obtained after the heart rate and blood pressure data at time t2 and t3 are not processed by the activation function, and c2 is the output of the current layer obtained after the heart rate and blood pressure data at time t2 is processed by the activation function.

[0025] Figure 3 FIG. 1 shows a schematic block diagram of a timing processing layer according to an embodiment of the present invention. Figure 3 As shown, according to one embodiment, the timing processing layer 220 is divided into a first time processing unit, a second time processing unit and a third time processing unit.

[0026] The first moment processing unit determines whether return_sequences is true (True). That is, it determines whether the moment processing unit returns a complete output sequence. Those skilled in the art know how to make such a determination, so it will not be described in detail.

[0027] If it is true, that is, return_sequences = True, the hidden state h1 is output to the fully connected layer and (here it is necessary to combine the second and third moment processing units (h2, h3), that is, (h1, h2, h3) to output to the fully connected layer, h1 will be temporarily stored in the storage unit, waiting for the second and third moment processing units to output h2, h3) the unit state c1 is passed to the next moment processing unit, that is, the second moment processing unit. If it is not true, that is, return_sequences = False, the hidden state h1 is used as the output of the entire sequence, not output to the fully connected layer, but output to the second moment processing unit processing layer. Here h1 is a sequence, that is, a vector.

[0028] When the second moment processing unit receives the hidden state h1 and the unit state c1, it determines whether return_sequences is true (True). If it is true, that is, return_sequences = True, the hidden state h2 is output to the fully connected layer (here it is necessary to combine all the previous outputs (h1, h3), that is, (h1, h2, h3) are combined and output to the fully connected layer, h2 will be temporarily stored in the memory, waiting for the third moment processing unit to output h3) The unit state c2 is passed to the next moment processing unit (that is, the third moment processing unit) and the fully connected layer. If it is not true, that is, return_sequences = False, only the hidden state h2 is output as the entire sequence to the next third moment processing unit processing layer.

[0029] When the third moment processing unit receives the hidden state h2 and the unit state c2, it determines whether return_sequences is true (True). If it is true, that is, return_sequences = True, the hidden state h3 is passed to the fully connected layer. Here, it is necessary to combine all the previous outputs (h1, h2), that is, (h1, h2, h3) and output them to the fully connected layer (there is no next one here, so c3 does not need to be passed to the next one). If it is not true, that is, return_sequences = False, only the hidden state h3 is used as the output of the entire sequence and the output of the entire time series processing layer. h3 is directly output to the fully connected layer. Because this is the last moment processing unit, it is directly output to the fully connected layer and does not need to be passed to the next moment processing unit. The shape of h3 here is a (1,3) vector. Because there are three moment processing units here, h3 is a (1,3) vector. This is the output shape of the entire time series processing layer after return_sequences = False is selected. It is a fixed output shape after we select return_sequences = False.

[0030] Although three time processing units are shown above, there may be more time processing units. The processing of each time processing unit can refer to the above description.

[0031] Therefore, when the return_sequences of the three time processing units are not true, the overall output shape of the timing layer where the time processing unit is located will be a (1,3) vector.

[0032] return_sequences = False, the fully connected layer 230 receives the output h3(1,3) vector of the time series processing layer as input. The output of the time series processing layer is subjected to feature extraction and transformation, useful information is extracted from the sequence data, and encoded into a hidden state. Then, the linear combination of the input features (plus the nonlinear activation function Relu) is learned to further extract and transform these features. The fully connected layer of the neural network can be used to extract and transform features using various currently known methods.

[0033] According to one embodiment, the fully connected layer includes 32 neurons, and the final output is 32 features t_sequences. These features are the relationship between the change characteristics of heart rate and blood pressure in the time series t1, t2, and t3 and time, as well as some difficult-to-capture intrinsic relationships, which are used to combine with other data later to analyze whether there is some correlation with other features.

[0034] According to one embodiment, the air environment related data acquisition device 300 acquires the air environment data around the device holder, the location data of the device holder (such as GPS data), the road network data, and the current time data.

[0035] According to one embodiment, the air environment related data acquisition device 300 uses an air quality sensor to feed back the air environment pollutants to the wristband interface in real time. The wristband interface displays the values ​​of various air pollutants. Here, the data of air environment pollutants are combined with road network data to indicate where the pollutant values ​​are obtained. For example, the road network data of OpenStreetMap can be used. "Road network data" generally refers to a spatial data set that describes the structure, attributes and relationships of a road network.

[0036] PM2.5: 0-12μg / m 3 (Excellent) to 250 μg / m 3 above (dangerous).

[0037] PM10: 0-54μg / m 3 (Excellent) to 420 μg / m 3 above (dangerous).

[0038] CO: 0-4.4ppm (excellent) to above 15ppm (dangerous).

[0039] NO2: 0-53ppb (excellent) to above 360ppb (dangerous).

[0040] O3: 0-54ppb (excellent) to over 300ppb (dangerous).

[0041] According to one embodiment, the range above the standard is defined as a dangerous range. This is the air quality data pushed by an air collection device or a mobile phone. Above this range, the vibration sensor of the bracelet will remind the holder not to go out, for example, through a vibration reminder similar to a mobile phone.

[0042] Road network data can provide path data related to roads, but calculating the estimated value of environmental exposure requires not only the path provided by the road network data, but also the starting point and the ending point. The road network data cannot directly provide this starting point and the ending point, so the wristband needs a GPS positioning device (which can obtain the current location of the device holder) to provide the starting point and the ending point. The air environment related data acquisition device 300 can obtain the air environment data of this path.

[0043] Assume that a set of spatially distributed observation points {x1,x2,...,x n}, each point x i Corresponding to an observation z(x i ). The environmental exposure level acquisition device 400 determines the environmental exposure level according to each observation point of the path and its corresponding air quality data observation value.

[0044] According to one embodiment, the environmental exposure level acquisition device 400 first performs spatial interpolation on the data obtained by the air environment related data acquisition device 300 (because the air environment sensor cannot continuously collect and obtain the current air), and then determines the environmental exposure level based on the result after the difference.

[0045] (1) Interpolation processing

[0046] According to one embodiment, each air pollutant gas is interpolated and the value of each air pollutant gas is calculated respectively, so as to calculate the continuous values ​​of various air pollutants along this path in the entire path area. According to the embodiment of the present invention, by performing difference, a value that changes continuously along this path is finally obtained and integrated, so as to obtain a better technical effect.

[0047] According to one embodiment of the present invention, interpolation is performed as follows.

[0048] Assume that a set of spatially distributed observation points {x1,x2,...,x n}, each point x i Corresponding to an observation z(x i ), the goal is to predict the value z(x0) at any unobserved point x0 in space;

[0049] Randomly select a point x0 on the path;

[0050] Calculate the distance between each known point and the randomly selected point x0, then select the N observation points with the smallest distance as the known numbers, and use formula (1) to infer the value z(x0) at x0:

[0051]

[0052] Among them, λ i is the spatial distance weight coefficient, and by solving the following equation group (2) we get

[0053]

[0054] In equation (2), u is the Lagrange multiplier, x i is the point closest to x0, γ(x i ,x j ) is the variation function, γ(h)=γ(x i ,x j ):

[0055]

[0056] Where N(h) is the value divided by x i In addition, all the remaining x j The number of points, h is x i and x j The distance between them. Using the above formulas (1), (2) and (3), the air pollutant concentration z(x0) at position x0 can be obtained. According to an embodiment of the present invention, by using spatial points with strong correlation and using Lagrange multipliers, the estimated pollutant concentration at the position point is more accurate.

[0057] (2) Calculate the estimated environmental exposure level

[0058] After the interpolation is completed, the estimated value of the environmental exposure level is obtained. The environmental exposure level determination device 400 obtains the environmental exposure level of the device holder based on the air pollutant concentration z(l) obtained by interpolation of the air pollutants at the location of the missing point l and the data obtained by the original air sensor.

[0059] According to one embodiment, the environmental exposure level of the device holder at a point l on his / her movement path during the time period (t1, t2) can be determined according to the following formula:

[0060] The intensity of exercise is determined based on the accelerometer and gyroscope of the wristband (these two sensors are already popular in sports wristbands), as well as the heart rate acquisition device. The intensity of exercise of the device holder at location l is determined based on these sensors or devices. Then, for example, the value of the device holder's exercise intensity is determined based on the short-term exposure to exercise-related exercise intensity recommended by the United States Environmental Protection Agency (USEPA) for respiratory rate. There are two forms of environmental exposure levels at location l, as shown in the following two formulas:

[0061] ① When the device holder does not move at location l, the environmental exposure level is as follows:

[0062]

[0063] ② When the device holder is moving at location l, the environmental exposure level is as follows:

[0064]

[0065] The environmental exposure level obtained by the above formula (4) or formula (5) is uniformly named as: environmental exposure level 有运动或无运动，人口尺度，地点l，时间t1-＞t2

[0066] Based on formulas (4) and (5), the following formula (6) can be obtained: the total environmental exposure level of the device holder on the path from l1 to l2 during the total time t5 to t6 at the individual scale ((t1, t2) is the time the device holder stays at point l, (t3, t4) is the total time (t5, t6) minus (t1, t2)), is:

[0067]

[0068] IR(age, gender) is the respiratory rate value obtained by querying the exposure summary table of recommended respiratory rate values ​​using age and gender. IR(age, gender, activity intensity) is the respiratory rate value obtained by querying the exposure summary table of recommended respiratory rate values ​​using age, gender and activity intensity. Activity intensity includes resting, sitting, light activity, moderate activity, heavy activity, high-limit activity, etc.

[0069] An exemplary US Environmental Protection Agency (USEPA) exposure summary table for breathing rate recommendations is as follows:

[0070] Table 1 Summary of recommended values ​​for respiratory rate

[0071]

[0072] US Environmental Protection Agency (USEPA) exposure summary table for respiratory rate recommendations

[0073] The pollutant concentration z(l) at location l is the concentration of air pollutants at a location l on the path from l1 to l2, and the population l represents the population at location l on l1 to l2 (this population is obtained through the online Chinese urban and rural resident population statistics), It is the sum of the populations of all locations on the path from l1 to l2. All the following descriptions are the values ​​X (replaced by X) of the total environmental exposure level of the device holder on the path from l1 to l2 during the period from t5 to t6 at the individual scale.

[0074] The obtained environmental exposure level X is normalized to a unified scale so that the value of the environmental exposure level X is converted from the original physical quantity to a value between 0 and 1, wherein the normalization benchmark takes the maximum value MAX(X) of the test training. Where XN is the normalized value of X.

[0075] The environmental exposure vector is obtained by the formula XN=XH*XL, where XH corresponds to the unit row vector (x1, x2...) h , where x1+x2....=1, here is the unit row vector. After getting the row vector, we can invert the column vector XL, which is (x1,x2......) l The obtained XL is the environmental exposure vector (that is, the environmental exposure characteristic quantity after the total transformation of the environmental exposure level).

[0076] Subsequently, the vector merging device 600 merges the heart rate and blood pressure vector obtained by the heart rate and blood pressure vector acquiring device 200 with the environmental exposure vector obtained by the environmental exposure vector acquiring device 500. The vector data obtained after processing the heart rate and blood pressure can be merged with the environmental exposure vector by splicing.

[0077] Finally, the early warning device 700 issues an early warning based on the merged vector. The merged vector is input into a neural network model that has been pre-trained using the spatial data and heart health data obtained in advance. It is first linearly transformed and then converted into the probability distribution of diseases related to heart health, thereby obtaining the predicted probability of various heart-related diseases. According to one embodiment, the category with the highest probability is selected as the prediction result of the model, and an early warning is issued based on the prediction result. The predicted result is used as a result of providing sexual advice. If the predicted result matches the type of a certain disease, the device holder or guardian can be warned to tell what type of cardiovascular disease the device holder may have, and it is recommended to go to the hospital for a check-up. Terminal devices such as mobile phones and computers can be used to receive or monitor the specific health status of the device holder.

[0078] Heart rate and blood pressure are two indicators that directly reflect cardiovascular or heart health, but sometimes they alone are not enough because the device holder is also affected by external factors. Air pollutant environment is a relatively large factor that can affect cardiovascular disease. According to an embodiment of the present invention, the air pollution environment and heart rate and blood pressure can be combined to provide more data sources for early warning, which greatly improves the accuracy of early warning.

[0079] Figure 4 FIG. 4 is a schematic block diagram of a cardiovascular disease early warning device based on spatiotemporal big data according to another embodiment of the present invention. Figure 4 As shown, according to another embodiment of the present invention, a cardiovascular disease early warning device based on spatiotemporal big data is Figure 1 The cardiovascular disease early warning device based on spatiotemporal big data shown also includes a personal information acquisition device 800 and a personal information vector generation device 900.

[0080] The personal information acquisition device 800 collects the personal information of the holder, such as gender, age, etc. The personal information acquisition device 800 can receive the personal information from the outside through, for example, a USB interface, or can be set up with a touch screen, a mouse, etc., to collect the personal information from the touch screen interface.

[0081] The personal information vector generation device 900 removes irrelevant characters, punctuation marks, stop words, special symbols, etc. from the collected personal information, and only retains words that are useful for subsequent processing. The text is divided into words, phrases or sub-word units (Token). Then the basic personal information is encoded, and the segmented text is converted into a digital code. This conversion can be achieved through a pre-established vocabulary (Vocabulary). Each unique word in the vocabulary corresponds to a unique index (Index). The segmented text is compared with the vocabulary so that it can be converted into a digital code.

[0082] According to one embodiment, the vocabulary is constructed as follows:

[0083] Collect vocabulary: Iterate through a text dataset (such as a patient's health record or diagnosis information) and record all the words that appear;

[0084] Deduplication, remove duplicate words and ensure that each word in the vocabulary is unique;

[0085] Assign indices, assigning each vocabulary word a unique index. This is usually a sequential integer starting from 0 or 1.

[0086] The vocabulary is a mapping where each unique word is assigned a unique index (integer). Once you have the vocabulary, you can convert the tokenized text into a digital encoding. This process involves looking up the vocabulary and replacing each word in the text with its corresponding index in the vocabulary, resulting in a sequence of digital encodings. This sequence is the personal information vector.

[0087] The vector merging device 600 merges the heart rate and blood pressure vector obtained by the heart rate and blood pressure vector acquiring device 200 , the environmental exposure vector obtained by the environmental exposure vector acquiring device 500 , and the personal information vector.

[0088] Figure 5 FIG. 4 is a schematic block diagram of a cardiovascular disease early warning device based on spatiotemporal big data according to another embodiment of the present invention. Figure 5 As shown, Figure 1 Compared with the illustrated embodiment, a cardiovascular disease early warning device based on spatiotemporal big data according to another embodiment of the present invention further includes a food picture acquisition unit 1000 and a food picture vector acquisition unit 1100.

[0089] The food picture acquisition unit 1000 can acquire food pictures taken by the device holder, or take food pictures of the device holder by himself (the food picture acquisition device here has a specific high-quality sensor, such as a camera with a DSLR sensor).

[0090] The food photo data can be enhanced by performing operations such as rotation, scaling, and cropping to increase the robustness of the model. The food image analysis model here is trained using a semi-supervised model with L2 regularization and voting. The main structure of the model draws on the residual model structure. We modify each convolution structure into three parts: left, middle, and right. The left side is a 3x3 convolution and then a 1x1 convolution on the feature result of the previous step (or the unprocessed original feature vector from the first step, that is, the unprocessed vectorized image). The right side is a 5x5 convolution and then a 3x3 convolution on the feature result of the previous step. The middle is the feature vector of the previous step, followed by the convolution structure on the left. The convolution result on the left is assigned a weight of 0.6, the convolution result on the right is assigned a weight of 0.4, and then the left, middle and right are added together (the results on both sides are in matrix form, so the assignment of weights and the addition are all additions in the numerical mathematical sense of the corresponding matrices. This is the result of each convolution structure, that is, the result of each convolution layer). The result of the addition is named RU here, and then it is added to the result RV of the next identical convolution structure, that is, RW=RU+RV, and this cycle is repeated until the last convolution structure, that is, the last convolution layer. This series of operations is used to enhance the model's ability to analyze richer features of food in images, such as the ability to analyze fat and other ingredients.

[0091] According to one embodiment of the present invention, the food picture vector acquisition unit 1100 includes a first convolution layer, a second convolution layer, a third convolution layer, a fourth convolution layer, a flattening layer, and a fully connected layer. According to one embodiment, it acquires the food picture vector as follows.

[0092] The first convolution layer performs convolution operation on the input image to obtain the 64-channel vector RA of the convolution feature of this image. This is to extract the low-level features of the food shape or color in the image, and obtain preliminary features through the nonlinear feature relationship of the nonlinear activation function and some strong features (here mainly for the extraction of features such as food surface oil or color texture that are not easy to analyze).

[0093] The second convolution layer: The second convolution layer convolves the output of the first convolution layer to obtain the result RB, and then adds them to get the result RW = RA + RB, thereby further extracting more complex features and obtaining the 128-channel vector of the convolution feature of this image. At this time, based on the features extracted in the previous layer, a deeper feature extraction can improve the accuracy of feature extraction, and the same nonlinear activation function is used to maintain nonlinearity and promote feature learning, and extract significant features.

[0094] The third convolution layer: The third convolution layer convolves the output of the second convolution layer to obtain the result RC, and then adds them to obtain the result RX = RW + RC, thereby further extracting more complex features and obtaining the vector of the 256-channel convolution feature of this image. At this time, based on the features extracted in the previous layer, a deeper feature extraction can be performed to improve the accuracy of feature extraction, and the same nonlinear activation function is used to maintain nonlinearity and promote feature learning. Extract more complex features.

[0095] The fourth convolution layer: The fourth convolution layer convolves the output of the third convolution layer to obtain the result RE. After adding, the result RY = RW + RE can be obtained, thereby further extracting more complex features and obtaining the vector of the 512-channel convolution feature of this image. At this time, based on the features extracted in the previous layer, a deeper feature extraction can be performed to improve the accuracy of feature extraction. The same nonlinear activation function is used to maintain nonlinearity and promote feature learning. The final feature vector of the convolution is obtained.

[0096] Average pooling layer: mainly reduces the complexity of image extraction results and speeds up the calculation of results (step size is 2).

[0097] Flattening layer: The flattening layer flattens the significant features extracted by the second convolutional layer into a one-dimensional array imgx(i1,i2......). Each component of the one-dimensional array here is a related feature.

[0098] Extract the fully connected layer to obtain the flattened data. Each neuron in the fully connected layer is connected to all neurons in the flattened layer. The output of each neuron in the fully connected layer is the weighted sum of the outputs of all neurons in the previous layer, and is nonlinearly transformed through the activation function to combine the learned "local" features into global features. This global feature is the food image vector.

[0099] According to this implementation, due to the selection of the number of convolutional layers and the convolutional layer structure, the obtained food image vector can better obtain food features, thereby contributing to the accuracy of the warning results.

[0100] The implementation mode of the present invention combines food pictures with air pollutants, basic personal information heart rate and blood pressure data, and can provide early warning more accurately. On the other hand, due to the processing of pictures and heart rate, blood pressure and other numerical values ​​by the present invention, pictures and heart rate, blood pressure and other numerical values ​​can be well combined, solving the problem that pictures and numerical values ​​cannot be well predicted together in the past.

[0101] The vector merging device 600 merges the heart rate and blood pressure vector obtained by the heart rate and blood pressure vector acquiring device 200 , the environmental exposure vector obtained by the environmental exposure vector acquiring device 500 , the food picture vector and the personal information vector.

[0102] Finally, the early warning device 700 issues an early warning according to the combined vectors. How to issue an early warning has been described above and will not be repeated here.

[0103] Those skilled in the art will understand that Figure 4 and Figure 5 The implementation methods can be combined, that is, the cardiovascular disease warning device based on spatiotemporal big data can simultaneously include a food picture acquisition unit 1000, a food picture vector acquisition unit 1100, a personal information acquisition device 800 and a personal information vector generation device 900.

[0104] The present invention can integrate environmental monitoring stations, smart phones, doctor's computers, smart bracelets, etc. to monitor people's behavior and the geographical environment, and can well detect heart health conditions.

[0105] Each step or unit of the present invention can be implemented by hardware, or by software in combination with hardware, or by a processor executing software stored in a memory. The above detailed description of the present invention only provides a further understanding of the content for those skilled in the art, for implementing the preferred aspects of the present invention, and does not limit the scope of the present invention. Only the claims are used to determine the scope of protection of the present invention. Therefore, the combination of features and steps in the foregoing detailed description is not necessary for implementing the present invention within the corresponding scope, and many are replaceable, and only representative embodiments of the detailed description of the present invention are taught here. In addition, in order to obtain additional useful embodiments of the present invention, the various features taught in the specification can be combined in a variety of ways, but these ways are not specifically listed.

Claims

1. A heart health warning device for people based on multi-source geographic spatiotemporal big data, characterized in that: include: A heart rate and blood pressure data acquisition device, used to collect the patient's heart rate and blood pressure data; A heart rate and blood pressure vector acquisition device, which acquires a heart rate and blood pressure vector according to the heart rate and blood pressure data; An air environment related data acquisition device, which acquires air environment pollutant gas data around the holder of the early warning device, the holder's location data, road network data, and current time data; an environmental exposure level determining device, for determining the environmental exposure level of the holder based on the data obtained by the air environment related data obtaining device; An environmental exposure vector acquisition device determines an environmental exposure vector according to the environmental exposure level of the holder; A vector merging device, merging the heart rate and blood pressure vector with the environmental exposure vector; The early warning device provides early warning of heart health based on the merged vectors.

2. The early warning device according to claim 1, characterized in that: The environmental exposure level determination device determines the environmental exposure level as follows: The air environment related data acquisition device performs spatial interpolation on the acquired air environment pollutant gases to obtain the values ​​of each air pollutant that continuously changes along the movement path of the holder; The environmental exposure level is determined according to the numerical values ​​of the air pollutants by an integration method.

3. The early warning device according to claim 2, characterized in that: The environmental exposure level determination device performs spatial interpolation as follows: Assume that a set of observation points {x1,x2,...,x n }, each point x i Corresponding to an observation z(x i ), the goal is to predict the value z(x0) at any unobserved point x0 in space; Randomly select a point x0 on the path; Calculate the distance between each known point and the randomly selected point x0, then select the N observation points with the smallest distance as the known numbers, and use formula (1) to infer the value z(x0) at x0: Among them, λ i is the spatial distance weight coefficient, and by solving the following equation group (2) we get In equation (2), u is the Lagrange multiplier, x i is the point closest to x0, γ(x i ,x j ) is the variation function, γ(h)=γ(x i ,x j ): Where N(h) is the value divided by x i In addition, all the remaining x j The number of points, h is x i and x j The distance between.

4. The early warning device according to claim 2, characterized in that: The environmental exposure level determination device determines the environmental exposure level of the device holder at location l on his / her moving path within the time range from total time t1 to t2 according to the following formula: ① When the device holder does not move at location l, the environmental exposure level is as follows: ② When the device holder is moving at location l, the environmental exposure level is as follows: The environmental exposure level obtained by the above formula is uniformly named as: environmental exposure level 有运动或无运动，人口尺度，地点l，时间t1-＞t2 , Then, the total environmental exposure level of the device holder on the path from l1 to l2 during the total time from t5 to t6 is determined based on the following formula: (6) Wherein, IR(age, gender) is the respiratory rate value obtained by querying the exposure summary table of recommended respiratory rate values ​​using age and gender, IR(age, gender, activity intensity) is the respiratory rate exposure value obtained by querying the exposure summary table of predetermined respiratory rate recommended values ​​using age, gender and activity intensity, and z(l) is the concentration value of the pollutant at point l; Among them, (t1, t2) is the time the holder stays at point l, (t3, t4) is the time period obtained by subtracting (t1, t2) from the total time (t5, t6), the pollutant concentration z(l) at point l is the concentration value of the air pollutant at point l on the path from l1 to l2, and the population l Represents the population at location l on l1 to l2 It is the sum of the populations of all locations on the path from l1 to l2.

5. The early warning device according to claim 3, characterized in that: The environment exposure vector acquisition device determines the environment exposure vector as follows: The obtained environmental exposure level X is normalized to a unified scale so that the value of the environmental exposure level X is converted from the original physical quantity to a value between 0 and 1, wherein the normalization benchmark takes the maximum value MAX(X) of the test training. Where XN is the normalized value of X; Then let XN = XH*XL, where XH corresponds to the unit row vector (x1, x2...) h , where x1+x2....=1, the obtained XL is the environmental exposure vector.

6. The early warning device according to claim 1, characterized in that: The early warning device also includes a food picture acquisition unit and a food picture vector acquisition unit. The food picture acquisition unit acquires the food picture of the holder; The food picture vector acquisition unit obtains a food picture vector according to the food photo; The vector merging device merges the heart rate and blood pressure vector, the environmental exposure vector and the food picture vector.

7. The early warning device according to claim 6, characterized in that: The food picture vector acquisition unit includes a first convolution layer, a second convolution layer, a third convolution layer, a fourth convolution layer, an average pooling layer, a flattening layer, and a fully connected layer; The first convolutional layer performs a convolution operation on the input image to obtain a 64-channel vector, thereby obtaining preliminary features of the input image; The second convolutional layer convolves the preliminary features and adds them to the preliminary features to obtain a 128-channel vector, thereby obtaining preliminary significant features; The third convolutional layer convolves the preliminary salient features and adds them to the preliminary salient features to obtain a 256-channel vector, thereby obtaining an intermediate salient feature; The fourth convolutional layer convolves the intermediate salient features and adds them to the intermediate features to obtain a 512-channel vector, thereby obtaining a final salient feature; Average pooling layer: pooling the final significant features; The flattening layer flattens the pooled final salient features into a one-dimensional array; The fully connected layer uses full connection to transform the one-dimensional array to obtain a global variable as the food image data.

8. The early warning device according to claim 7, characterized in that: The convolution structures of the first convolution layer, the second convolution layer, the third convolution layer and the fourth convolution layer are divided into three parts: left, middle and right. The left part is to perform 3x3 convolution and then 1x1 convolution on the feature vector of the previous step, the right part is to perform 5x5 convolution and then 3x3 convolution on the feature vector of the previous step, and the middle part obtains the feature vector of the previous step. The convolution result on the left is assigned a weight of 0.6, and the convolution result on the right is assigned a weight of 0.

4. Then the result of adding the left, middle and right is the result of the convolution layer.

9. The early warning device according to claim 1, characterized in that: The early warning device also includes a personal information acquisition device and a personal information vector generation device; The personal information acquisition device collects the personal information of the holder; The personal information vector generating device generates a personal information vector based on the personal information.

10. The early warning device according to claim 9, characterized in that: The personal information vector generating device divides the personal information into words, phrases or sub-word units, and then converts the words, phrases or sub-word units into digital codes according to a pre-established vocabulary, thereby obtaining a sequence composed of digital codes, namely the personal information vector.

Images