Pet emotion recognition system and method

Abstract

The invention discloses a pet emotion recognition system and method, and relates to the technical field of pet emotion recognition, and the system comprises a data acquisition module, a data analysis module, an environment analysis module, an emotion analysis module, an emotion comparison module, and an abnormality processing module. The data acquisition module is used for acquiring classified pet historical emotion data, pet emotion data acquired in real time and environment data acquired in real time, and then sorting the classified pet historical emotion data; the technical key points are as follows: analyzing and processing historical emotion data of the pet, identifying the emotion of the pet from two perspectives of sound and physiology, knowing the relationship between the emotion of the pet and the sound and physiology, and then identifying the emotion of the pet in real time based on the relationship between the emotion of the pet and the sound and physiology. The external environment and the state of the pet are analyzed in the pet emotion recognition process, the pet emotion recognition accuracy can be improved, and the good use prospect is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of pet emotion recognition, and in particular to a pet emotion recognition system and method. Background Art

[0002] The existing patent application publication number is CN110415727B, and the invention patent titled Pet Emotion Recognition Method and Device records that the following steps are included: extracting the acoustic features of the audio data of a specified pet, and specifying that the audio data of a specified pet is the audio data corresponding to a pet subject; inputting the acoustic features into an emotion representation model, and outputting the emotion judgment information corresponding to the pet subject, wherein the emotion representation model is obtained after training based on the acoustic features of the samples and the emotion judgment results corresponding to the acoustic features of the samples. The method provided in the embodiment of the present invention extracts the acoustic features of the audio data of the specified pet, inputs the acoustic features into the emotion representation model, and outputs the emotion judgment information corresponding to the pet subject. Since it is only necessary to extract the acoustic features of the audio data of the specified pet, and input the acoustic features into the emotion representation model, the emotion judgment information corresponding to the pet subject can be output without using a one-by-one comparison method, thereby reducing the amount of calculation and improving the efficiency of pet emotion recognition.

[0003] However, based on the above content and the prior art, the central idea of ​​the above patent is to obtain the audio data of the pet, and then compare the pet's audio data with the standard voiceprint data of its corresponding emotions one by one, and then calculate the similarity with the voiceprint data, and recognize the pet's emotions by judging the similarity.

[0004] However, the above scheme has great defects in actual use. There are many categories of emotions. Therefore, the number of comparisons is relatively large. At the same time, each frame of the audio data needs to be compared, and the amount of data for comparison is very large. In addition, affected by the external environment and the pet's own state, its sound will naturally change to a certain extent, causing the comparison result to change, resulting in comparison errors, making the identified pet emotions wrong, and not meeting people's usage requirements. For this reason, we have developed a pet emotion recognition system and method. Summary of the invention

[0005] 1. Technical issues to be resolved In view of the deficiencies in the prior art, the present invention provides a pet emotion recognition system and method, which analyzes and processes the pet's historical emotion data, identifies the pet's emotions from both sound and physiology perspectives, understands the relationship between the pet's emotions and sound and physiology, and then identifies the pet's emotions in real time based on the relationship between the pet's emotions and sound and physiology. In the process of identifying the pet's emotions, the external environment and the pet's own state are analyzed to further reduce the impact of the external environment and the pet's own state, thereby improving the accuracy of pet emotion recognition. The overall use effect is good and it has a good prospect for use.

[0006] (II) Technical solution To achieve the above objectives, the present invention is implemented through the following technical solutions: A pet emotion recognition system includes a data acquisition module, a data analysis module, an environment analysis module, an emotion analysis module, an emotion comparison module and an exception handling module: Data acquisition module: used to obtain classified pet historical emotion data, real-time collected pet emotion data and real-time collected environmental data, and then organize the classified pet historical emotion data, the historical emotion data includes historical sound index data and historical physiological index data, the historical sound index data is the pet sound data, and the historical physiological index data includes pet breathing rate data and pet heart rate data; Data analysis module: used to obtain the results sorted by the data acquisition module, and formulate the sound classification intervals corresponding to different emotions of the pet and the physiological classification intervals corresponding to different emotions of the pet according to the preset standards; Environmental analysis module: used to obtain the physiological index data of the pet when it is sleeping and the environmental data collected in real time, and compare the physiological index data of the pet with the standard physiological index data of the pet when it is sleeping, and then calculate the index deviation value in combination with the environmental data collected in real time, and adjust the sound classification interval and the physiological classification interval according to the index deviation value to obtain the real-time sound classification interval and the real-time physiological classification interval; Emotion analysis module: used to obtain the real-time collected pet emotion data, and further analyze and calculate the obtained pet emotion data, so as to obtain the pet emotion sound value and the pet emotion physiological value, and then compare and analyze the pet emotion sound value and the pet emotion physiological value with the real-time sound classification interval and the real-time physiological classification interval respectively, to obtain two emotion classifications; Emotion comparison module: used to obtain two emotion classifications and compare the two emotion classifications. If the two emotion classifications are the same, the pet emotion corresponding to the emotion classification is output; If the two emotion classifications are different, it is determined to be abnormal and an abnormal signal is issued; Exception handling module: used to receive abnormal signals and two emotion classifications, and use the voiceprint recognition algorithm to perform similarity analysis on the real-time collected pet sound data and the standard sound data of the two emotion classifications to obtain two sets of comprehensive similarity data, and then compare the two sets of comprehensive similarity data to output the pet emotion corresponding to the emotion classification with the larger comprehensive similarity data.

[0007] Furthermore, the steps of formulating the sound classification interval and the physiological classification interval according to the preset standard based on the sorted results are as follows: S1. Obtain the collection time corresponding to the classified pet historical emotion data; S2. Query the environmental data of the pet's historical emotion data collection time and calculate the environmental impact value. The environmental data includes air quality index, temperature and humidity; S3, comparing the environmental impact value with the standard threshold, extracting the pet historical emotion data with environmental impact values ​​less than or equal to the standard threshold to construct a benchmark data set, and extracting the pet historical emotion data with environmental impact values ​​greater than the standard threshold to construct a special data set; S4, analyzing the data in the benchmark data set, calculating the pet emotion sound value based on the historical sound index data, and calculating the pet emotion physiological value based on the historical physiological index data; S5. Use a clustering algorithm to process the pet emotion sound value and the pet emotion physiological value to obtain the threshold interval of the corresponding emotion, thereby obtaining the sound classification interval and the physiological classification interval.

[0008] Furthermore, the calculation formula of the environmental impact value is as follows: ;

[0009] Where IFh is the environmental impact value, n is the number of environmental data collected on the day corresponding to the time when the pet's historical emotional data was collected, and H j is the temperature data collected for the jth time, Bz W is the preset standard temperature data, xs W is the temperature proportionality coefficient, RH j is the humidity data collected for the jth time, Bz S is the preset standard humidity data, xs S is the influence coefficient of humidity ratio, AQI j is the air quality index data collected for the jth time, Bz AQI is the preset standard air quality index data, xs AQI is the influence coefficient of the air quality index ratio; The standard threshold is .

[0010] Furthermore, the steps for calculating the pet emotional sound value based on the historical sound index data are as follows: Process the historical sound index data to obtain a sound spectrum graph; The historical sound index data with a total duration of T is divided into m parts and the average value is calculated to obtain m groups of average amplitude data; The pet emotional sound value is calculated based on m groups of spectrum data. The specific calculation formula is as follows: ;

[0011] In the formula, Emo SY is the pet's emotional sound value, A i is the average amplitude data of the i-th group after division, A j is the average amplitude data of the jth group after division, A max is the maximum value among the m groups of amplitudes, α and β are both constant coefficients, α>β>1, and Tb is the preset standard amplitude data; The formula for calculating the pet's emotional physiological value based on historical physiological indicator data is as follows: ;

[0012] In the formula, Emo SL is the pet's emotional physiological value, HRj is the detected pet's heart rate data, HRb is the pet's heart rate data in a calm state, Rj is the detected pet's respiratory rate data, Rb is the pet's respiratory rate data in a calm state, and e is the natural constant data.

[0013] Furthermore, the steps of calculating the indicator deviation value are as follows: calculating the environmental impact value IFH´ based on the environmental data collected in real time, and comparing the environmental impact value IFH´ with the standard threshold; If IFH´≤ , at this time the calculation formula of the indicator deviation value is: ;

[0014] Wherein, Dev is the calculated index deviation value, HRs is the average heart rate data of the pet in the detected sleeping state, HRbs is the heart rate data of the pet in the standard sleeping state, Rs is the average respiratory rate data of the pet in the detected sleeping state, and Rbs is the average respiratory rate data of the pet in the standard sleeping state; If IFH´> , at this time, the calculation formula of the emotional index deviation value is: ;

[0015] Where K is the number of pet historical emotional data in the special data set with a difference ratio of IFh and IFH´ less than 5% and in the corresponding emotional state. The pet emotional sound value calculated by the number of pet historical emotional data corresponding to the emotional state of the i-th group, is the average value of the pet's emotional sound corresponding to the emotional state.

[0016] Furthermore, the sound classification interval is (ESymin i , ESymax i ), the physiological classification interval is (ESlmin i , ESlmax i ), the calculated real-time sound classification interval is (ESymin i +ESymin i ×Dev,ESymax i +ESymax i ×Dev), the calculated real-time physiological classification interval is (ESlmin i +ESlmin i ×Dev,ESlmax i +ESlmax i ×Dev), i in the sound classification interval, physiological classification interval, real-time sound classification interval and real-time physiological classification interval is the i-th pet emotion.

[0017] Further, when comparing and analyzing the pet emotion sound value and the pet emotion physiological value with the real-time sound classification interval and the real-time physiological classification interval respectively, the distance D1 between the pet emotion sound value and the real-time sound classification interval corresponding to each emotion category and the distance D2 between the pet emotion physiological value and the real-time physiological classification interval corresponding to each emotion category are calculated; D1=Emo SYS -(ESymin i +ESymin i ×Dev+ESymax i +ESymax i ×Dev) / 2; D2=Emo SLS -(ESlmin i +ESlmin i ×Dev+ESlmax i +ESlmax i ×Dev) / 2; In the formula, Emo SYS The pet emotional sound value is calculated based on the real-time collected sound index data. SLS It is the pet's emotional physiological value calculated based on the physiological indicator data collected in real time.

[0018] Furthermore, the standard sound data of the two emotion classifications are the historical sound index data closest to the middle value of the corresponding sound classification interval; The steps for using the voiceprint recognition algorithm to perform similarity analysis between the real-time collected pet sound data and the standard sound data of two emotion classifications are as follows: Preprocess the pet emotion data collected in real time to obtain sound spectrogram, speech spectrogram and mel spectrum graph; Obtain the sound spectrogram, speech spectrogram and mel-spectrogram obtained by preprocessing the standard sound data in two emotion intervals; Compare the three groups of spectra processed by the real-time collected pet emotion data with the three groups of spectra processed by two standard sound data, so as to obtain the sound spectrum similarity A1 and A2, the speech spectrum similarity B1 and B2, and the Mel spectrum similarity C1 and C2; Calculate two sets of comprehensive similarities, one set of comprehensive similarity is (A1+B1+C1) / 3, and the other set of comprehensive similarity is (A2+B2+C2) / 3.

[0019] Furthermore, the steps of preprocessing the real-time collected pet emotion data to obtain the spectrogram and mel-spectrogram are as follows: Pre-emphasize the audio signal of the sound indicator data in the pet emotion data collected in real time; The pre-emphasized audio signal is framed and then further processed using a Hanning window function; The further processed audio signal is subjected to a fast Fourier transform to obtain the spectrum of each frame, and the spectrum of the speech is modulo-squared to obtain the spectrogram of the audio signal; Perform logarithmic calculation on the spectrogram to obtain the Mel-spectrogram.

[0020] Furthermore, a pet emotion recognition method comprises the following steps: Obtain classified pet historical emotion data, real-time collected pet emotion data, and real-time collected environmental data, and then organize the classified pet historical emotion data, the historical emotion data includes historical sound index data and historical physiological index data, the historical sound index data is pet sound data, and the historical physiological index data includes pet breathing rate data and pet heart rate data; The sorted results are used to formulate sound classification intervals corresponding to different emotions of pets and physiological classification intervals corresponding to different emotions of pets according to preset standards; Compare the pet's physiological index data with the pet's standard sleep physiological index data, and then combine it with the real-time collected environmental data to calculate the index deviation value, and adjust the sound classification interval and the physiological classification interval according to the index deviation value to obtain the real-time sound classification interval and the real-time physiological classification interval; The obtained pet emotion data is further analyzed and calculated to obtain a pet emotion sound value and a pet emotion physiological value, and then the pet emotion sound value and the pet emotion physiological value are respectively compared and analyzed with the real-time sound classification interval and the real-time physiological classification interval to obtain two emotion classifications; The two emotion classifications are compared. If the two emotion classifications are the same, the pet emotion corresponding to the emotion classification is output; If the two emotion classifications are different, it is determined to be abnormal and an abnormal signal is issued; The voiceprint recognition algorithm is used to perform similarity analysis on the real-time collected pet sound data and the standard sound data of two emotion classifications to obtain two sets of comprehensive similarity data. The two sets of comprehensive similarity data are then compared to output the pet emotion corresponding to the emotion classification with the larger comprehensive similarity data.

[0021] (III) Beneficial effects The present invention provides a pet emotion recognition system and method, which has the following beneficial effects: The present invention provides a pet emotion recognition system and method, which analyzes and processes the pet's historical emotion data, identifies the pet's emotions from the two perspectives of sound and physiology, understands the relationship between the pet's emotions and sound and physiology, and then recognizes the pet's emotions in real time based on the relationship between the pet's emotions and sound and physiology. In the process of identifying the pet's emotions, the external environment and the pet's own state are analyzed to further reduce the influence of the external environment and the pet's own state, thereby improving the accuracy of pet emotion recognition, having a good overall use effect and having a good use prospect.

[0022] The present invention provides a pet emotion recognition system and method. In the process of recognizing the pet's emotions from the sound and physiological perspectives, the system analyzes the collected data, extracts the local sound and physiological features, and then uses the extracted features to recognize the pet's emotions. During the entire process, there is no need to compare the collected data with the standard emotion data one by one, which can reduce the amount of data comparison, thereby greatly improving the speed of pet emotion recognition, having a good use effect and a good use prospect. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A flowchart of a pet emotion recognition system according to the present invention; Figure 2 The present invention discloses an emotion graph based on the analysis of physiological data of a pet dog in a pet emotion recognition system. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] Research reasons The existing pet emotion recognition system solution is to obtain the pet's audio data, then frame the collected pet audio data, and then compare the pet's audio data with the standard voiceprint data of its corresponding emotion frame by frame, and then calculate the similarity with the voiceprint data, and realize the recognition of pet emotions by judging the similarity.

[0026] However, the above scheme has great defects in actual use. With the development of science and technology, the number of categories of pet emotion classification is constantly increasing (for example, common pet emotion recognition includes tension, relaxation, happiness, excitement, calmness and anger). Therefore, the number of audio data comparisons is constantly increasing. When comparing, every frame of each audio data is compared. Therefore, the audio data comparison takes a long time, which makes the speed of emotion recognition slow and the efficiency of emotion recognition is not high.

[0027] In addition, when the existing pet emotion recognition system is in use, it is affected by the external environment (temperature, humidity) and the pet's own condition (sickness, weakness, etc.), and its sound will naturally change to a certain extent, causing the audio data to change to a certain extent, so that the audio data is close in similarity to the standard audio data of multiple pet emotion categories, and the similarities are not high, which changes the comparison result, resulting in comparison errors, making the identified pet emotions wrong and not meeting people's usage requirements.

[0028] Design ideas Most existing systems process the collected pet audio data by frames, then compare the pet audio data with the standard voiceprint data of the corresponding emotion frame by frame, and then calculate the similarity with the voiceprint data. The pet's emotions are recognized by judging the similarity. The number of pet emotion classification categories is constantly increasing, and the time required for audio data comparison is relatively long, which makes the speed of emotion recognition slow and the efficiency of emotion recognition is low.

[0029] Therefore, during the research and development process, the initial idea was to change the existing idea and not use a single one-to-one comparison method to achieve emotion recognition. Therefore, based on the pet's historical emotional data, its relationship with sound and physiology was analyzed, and indicators that can be easily identified were studied, and the classification of pet emotions was achieved based on the new indicators.

[0030] However, in actual use, it is found that the pet's indicators will change with the external environment and the pet's own state. Therefore, the above scheme is used to identify the pet's emotions. After testing, the recognition accuracy is 50%-70%, which is not high and does not meet people's requirements.

[0031] Therefore, in the middle of the research and development, we changed our thinking and collected environmental data to study the impact of the environment on pets' sounds and physiological data. We made corresponding adjustments when analyzing the data, thereby improving the accuracy of identifying pets' emotions.

[0032] However, in actual use, it was found that during the analysis, since pets have similar physiological and sound data under different emotions, there will be inconsistencies in the sound and physiological analysis results. At this time, two emotions will be output, making it difficult to identify the pet's emotions.

[0033] Therefore, in the later stage of research and development, we fully combined the previous plans and developed a solution to deal with inconsistencies in sound and physiological analysis results, making the entire pet emotion recognition system plan more complete. After testing, the accuracy of pet emotion recognition reached more than 87%.

[0034] Study plan like Figure 1 As shown, a pet emotion recognition system includes a data acquisition module, a data analysis module, an environment analysis module, a emotion analysis module, an emotion comparison module and an exception handling module: This pet emotion recognition system consists of two parts: software and hardware.

[0035] The hardware part mainly includes the server that provides the computing basis for the software part and the data acquisition device corresponding to the data acquisition module.

[0036] The data collection device can be a bracelet or clothing equipped with a heart rate sensor and a respiratory rate sensor, which is worn on the pet and can collect data in real time.

[0037] Environmental data can be directly obtained from the data detected by the meteorological department.

[0038] The software part includes data analysis module, environmental analysis module, emotion analysis module, emotion comparison module and exception handling module.

[0039] Data collection In order to identify pet emotions, it is necessary to analyze the pet's historical data, understand the historical data, and then further analyze the real-time collected data to obtain more accurate results. The process of acquiring data is based on the data acquisition module.

[0040] The data acquisition module acquires classified pet historical emotion data, real-time collected pet emotion data and real-time collected environmental data, and then organizes the classified pet historical emotion data, wherein the historical emotion data includes historical sound index data and historical physiological index data, wherein the historical sound index data is pet sound data, and the historical physiological index data includes pet breathing rate data and pet heart rate data; The pet's historical emotion data is the data after the pet matures. The most basic one is the pet going through a voice changer. Because the pet's voice changes greatly before and after the voice change, the corresponding data must be used.

[0041] For example, a dog's voice is relatively small and crisp before it is one year old, and emotion recognition is generally not performed at this stage. This is mainly because there is little historical data at this stage, and the dog grows rapidly at this stage, and their physiology and voice change greatly, making it difficult to accurately identify emotions.

[0042] After a dog is one year old, its voice becomes louder, and its voice only becomes deeper when it reaches old age. This stage is relatively long and sufficient data is collected, so it is generally used to predict emotions at this stage.

[0043] As the amount of data increases, the pet's historical emotion data will also be filtered. The filtered pet's historical emotion data will be marked with the corresponding time and environmental data, rather than just the date corresponding to the data collection. Analysis based on this situation can effectively reduce the impact of the external environment, thereby improving the accuracy of emotion recognition.

[0044] Data processing After the relevant data is collected, it is necessary to conduct a detailed analysis of the collected pet historical data to analyze the relationship between emotions, physiology and sounds, and this process is based on the data analysis module.

[0045] The data analysis module obtains the results sorted by the data acquisition module, and formulates the sound classification intervals corresponding to different emotions of the pet and the physiological classification intervals corresponding to different emotions of the pet according to the preset standards; For example, there are six types of emotions: nervous, relaxed, happy, excited, calm, and angry.

[0046] Then there are 6 sound classification intervals and physiological classification intervals corresponding to different emotions of pets, which correspond one to one with the corresponding emotions.

[0047] There is overlap between the 6 sound classification intervals and the 6 physiological classification intervals.

[0048] For example, a dog's heart rate when angry is 110-140 beats per minute, and when excited it is 100-130 beats per minute. Also, a dog's breathing rate when angry is 26-35 beats per minute, and when excited it is 22-30 beats per minute.

[0049] Therefore, there is an overlap between the calculated anger physiological classification interval and the excitement physiological classification interval, and therefore, there will be a detection bias.

[0050] The steps for formulating the sound classification interval and physiological classification interval according to the preset standards are as follows: S1. Obtain the collection time corresponding to the classified pet historical emotion data; S2. Query the environmental data of the pet's historical emotion data collection time and calculate the environmental impact value. The environmental data includes air quality index, temperature and humidity; The calculation formula of environmental impact value is as follows: ;

[0051] Where IFh is the environmental impact value, n is the number of environmental data collected on the day corresponding to the time when the pet's historical emotional data was collected, and H j is the temperature data collected for the jth time, Bz W is the preset standard temperature data, xs W is the temperature proportionality coefficient, RH j is the humidity data collected for the jth time, Bz S is the preset standard humidity data, xs S is the influence coefficient of humidity ratio, AQI j is the air quality index data collected for the jth time, Bz AQI is the preset standard air quality index data, xs AQI is the influence coefficient of the air quality index ratio; Since there is less historical emotional data on pets and usually only the date of collection is recorded, in the initial stage, the air quality index, temperature and humidity in the environmental data are based on the difference between the average of the day and the preset standard data.

[0052] For example, temperature, humidity, and air quality index are usually collected every two hours, so n is 12. You can also collect 16 hours of temperature, humidity, and air quality data (excluding data corresponding to sleep time) every two hours, then n is 8.

[0053] S3, comparing the environmental impact value with the standard threshold, extracting the pet historical emotion data with environmental impact values ​​less than or equal to the standard threshold to construct a benchmark data set, and extracting the pet historical emotion data with environmental impact values ​​greater than the standard threshold to construct a special data set; The standard threshold is .

[0054] When the environmental impact value is less than the standard threshold, it means that the environmental data of that day is close to the standard environment and has little impact on the pet's sound and physiological data.

[0055] The data stored in the benchmark data set is less affected by environmental interference and is more accurate. The more data in the benchmark data set, the more accurate it is.

[0056] Special data sets are mainly affected by the external environment. Using them as benchmark analysis data will affect the accuracy of the analysis. Therefore, they cannot be used as benchmarks.

[0057] S4, analyzing the data in the benchmark data set, calculating the pet emotion sound value based on the historical sound index data, and calculating the pet emotion physiological value based on the historical physiological index data; The steps to calculate the pet emotional sound value based on historical sound indicator data are as follows: Process the historical sound index data to obtain a sound spectrum graph; The historical sound index data with a total duration of T is divided into m parts and the average value is calculated to obtain m groups of average amplitude data; The pet emotional sound value is calculated based on m groups of spectrum data. The specific calculation formula is as follows: ;

[0058] In the formula, Emo SY is the pet's emotional sound value, A i is the average amplitude data of the i-th group after division, A j is the average amplitude data of the jth group after division, A max is the maximum value among the m groups of amplitudes, α and β are both constant coefficients, α>β>1, and Tb is the preset standard amplitude data; Emo SY It mainly analyzes the sharpness of the amplitude data, the overall sound duration and the loudness of the sound, which can meet the accurate identification of pet emotions in most cases.

[0059] The weight coefficient is determined by the coefficient of variation method, which is a method of weighting each indicator according to the degree of variation between the current value of each evaluation indicator and the target value. If the numerical difference of an indicator is large, it can clearly distinguish the evaluated objects, indicating that the indicator has rich discrimination information, and thus the indicator should be given a larger weight. On the contrary, if the numerical difference of each evaluated object on a certain indicator is small, then the ability of this indicator to distinguish the evaluation objects is weak, and thus the indicator should be given a smaller weight. This method directly uses the information contained in each indicator and obtains the weight of the indicator through calculation, so it is objective.

[0060] The formula for calculating the pet's emotional physiological value based on historical physiological indicator data is as follows: ;

[0061] In the formula, Emo SL is the pet's emotional physiological value, HRj is the detected pet's heart rate data, HRb is the pet's heart rate data in a calm state, Rj is the detected pet's respiratory rate data, Rb is the pet's respiratory rate data in a calm state, and e is the natural constant data.

[0062] When the pet's mood changes, the pet's emotional physiological value will change significantly. The pet's mood can be analyzed more accurately through the pet's emotional physiological value.

[0063] S5. Use a clustering algorithm to process the pet emotion sound value and the pet emotion physiological value to obtain the threshold interval of the corresponding emotion, thereby obtaining the sound classification interval and the physiological classification interval.

[0064] The clustering algorithm used is K-means clustering algorithm.

[0065] The sound classification interval is (ESymin i , ESymax i ), the physiological classification interval is (ESlmin i , ESlmax i ).

[0066] Environmental treatment After understanding the relationship between pets' emotions, physiology and sounds, we also discovered that the external environment affects changes in physiology and sounds. Therefore, we need to understand changes in external environmental factors and the pet's own state in order to obtain relatively accurate real-time sound classification intervals and real-time physiological classification intervals. This process is based on the environmental analysis module.

[0067] The environmental analysis module obtains the physiological index data of the pet when it is sleeping and the environmental data collected in real time, and compares the physiological index data of the pet with the standard physiological index data of the pet when it is sleeping, and then calculates the index deviation value in combination with the environmental data collected in real time, and adjusts the sound classification interval and the physiological classification interval according to the index deviation value to obtain the real-time sound classification interval and the real-time physiological classification interval; The steps for calculating the indicator deviation value are as follows: calculating the environmental impact value IFH´ based on the environmental data collected in real time, and comparing the environmental impact value IFH´ with the standard threshold; If IFH´≤ , at this time the calculation formula of the indicator deviation value is: ;

[0068] Wherein, Dev is the calculated index deviation value, HRs is the average heart rate data of the pet in the detected sleeping state, HRbs is the heart rate data of the pet in the standard sleeping state, Rs is the average respiratory rate data of the pet in the detected sleeping state, and Rbs is the average respiratory rate data of the pet in the standard sleeping state; In this case, the environment has basically no effect, and the impact of the pet's own state is analyzed at this time.

[0069] The calculation formula of IFH´ is the same as that of IFH, only the data are different.

[0070] The changing proportion can directly reflect the changes in the pet itself. The original data corresponding to HRs and Rs collected are more than 6 groups, and the interval between each collection is 20min-40min, which can effectively reduce the physiological changes affected by other conditions (dreaming).

[0071] If IFH´> , at this time, the calculation formula of the emotional index deviation value is: ;

[0072] Where K is the number of pet historical emotional data in the special data set with a difference ratio of IFh and IFH´ less than 5% and in the corresponding emotional state. The pet emotional sound value calculated by the number of pet historical emotional data corresponding to the emotional state of the i-th group, is the average value of the pet's emotional sound corresponding to the emotional state.

[0073] For example, if a dog is in an angry mood, K is the number of historical emotional data of pets in an angry mood whose difference ratio between IFh and IFH´ is less than 5% in the special data set. The pet emotional sound value calculated for the number of pet historical emotional data under the i-th group of angry emotions, is the average value of the pet emotional sound value of anger, that is, the average value of the pet emotional sound value corresponding to the K group data.

[0074] By using the corresponding environment and the corresponding emotion value for analysis instead of unified calculation, the accuracy of the emotion index deviation value obtained is greatly improved, thereby improving the accuracy of subsequent calculation results, and the use effect is good.

[0075] In this case, it indicates that there is an environmental impact, and both environmental and physiological factors need to be considered. After analysis, the environment will also cause physiological changes. After long-term comparative analysis, the above formula is obtained.

[0076] The sound classification interval is (ESymin i , ESymax i ), the physiological classification interval is (ESlmin i , ESlmax i ), the calculated real-time sound classification interval is (ESymin i +ESymin i ×Dev,ESymax i +ESymax i ×Dev), the calculated real-time physiological classification interval is (ESlmin i +ESlmin i ×Dev,ESlmax i +ESlmax i ×Dev), i in the sound classification interval, physiological classification interval, real-time sound classification interval and real-time physiological classification interval is the i-th pet emotion.

[0077] For example, the order of emotions is anger, excitement, tension, happiness, calmness and relaxation, so i is 3, that is, (ESymin3, ESymax3) represents the tension sound classification interval.

[0078] The present invention provides a pet emotion recognition system for analyzing and processing the pet's historical emotion data, identifying the pet's emotions from the two perspectives of sound and physiology, understanding the relationship between the pet's emotions and sound and physiology, and then identifying the pet's emotions in real time based on the relationship between the pet's emotions and sound and physiology. In the process of identifying the pet's emotions, the external environment and the pet's own state are analyzed to further reduce the influence of the external environment and the pet's own state, thereby improving the accuracy of pet emotion recognition. The overall use effect is good and has a good prospect for use.

[0079] Sentiment Analysis When understanding the external environment and the pet's own status, it is necessary to further analyze the pet's emotional data collected in real time and identify the pet's emotions in real time, so as to help users better raise pets. This process is based on the emotion analysis module.

[0080] The emotion analysis module obtains the pet emotion data collected in real time, and further analyzes and calculates the obtained pet emotion data, thereby obtaining the pet emotion sound value and the pet emotion physiological value, and then compares and analyzes the pet emotion sound value and the pet emotion physiological value with the real-time sound classification interval and the real-time physiological classification interval respectively, to obtain two emotion classifications; When comparing and analyzing the pet emotion sound value and the pet emotion physiological value with the real-time sound classification interval and the real-time physiological classification interval respectively, the distance D1 between the pet emotion sound value and the real-time sound classification interval corresponding to each emotion category and the distance D2 between the pet emotion physiological value and the real-time physiological classification interval corresponding to each emotion category are calculated; D1=Emo SYS -(ESymin i +ESymin i ×Dev+ESymax i +ESymax i ×Dev) / 2; D2=Emo SLS -(ESlmin i +ESlmin i ×Dev+ESlmax i +ESlmax i ×Dev) / 2; In the formula, Emo SYS The pet emotional sound value is calculated based on the real-time collected sound index data. SLS It is the pet's emotional physiological value calculated based on the physiological indicator data collected in real time.

[0081] ESymin i +ESymin i ×Dev<Emo SYS <ESymax i +ESymax i ×Dev,ESlmin i +ESlmin i ×Dev<Emo SLS <ESlmax i +ESlmax i ×Dev; Emo SYS The calculation formula of Emo SY Same, Emo SLSThe calculation formula of Emo SL same.

[0082] The above calculation method can eliminate the interference caused by the overlap between most real-time physiological classification intervals, thereby obtaining more accurate emotions.

[0083] Figure 2 It is a pet emotion classification based on the analysis of physiological indicator data collected in real time.

[0084] Emotional judgment Emotions are judged from two categories, physiological and acoustic, so the analysis results may be different. Therefore, after obtaining two emotion classifications, it is necessary to compare whether the emotions corresponding to the two emotion classifications are the same, and the comparison process is based on the emotion comparison module.

[0085] The emotion comparison module obtains two emotion classifications and compares the two emotion classifications. If the two emotion classifications are the same, the pet emotion corresponding to the emotion classification is output; If the two emotion classifications are different, it is judged as abnormal and an abnormal signal is issued.

[0086] Anomaly Analysis When the two emotion classifications are different, it indicates that the above-mentioned fast calculation method cannot accurately obtain the data. At this time, the collected data needs to be analyzed and processed in detail, and this process is based on the exception handling module.

[0087] Exception handling module: used to receive abnormal signals and two emotion classifications, and use the voiceprint recognition algorithm to perform similarity analysis on the real-time collected pet sound data and the standard sound data of the two emotion classifications to obtain two sets of comprehensive similarity data, and then compare the two sets of comprehensive similarity data to output the pet emotion corresponding to the emotion classification with the larger comprehensive similarity data.

[0088] The standard sound data of the two emotion classifications are the historical sound index data closest to the middle value of the corresponding sound classification interval; The steps for using the voiceprint recognition algorithm to perform similarity analysis between the real-time collected pet sound data and the standard sound data of two emotion classifications are as follows: Preprocess the pet emotion data collected in real time to obtain sound spectrogram, speech spectrogram and mel spectrum graph; The steps for preprocessing the real-time collected pet emotion data to obtain the spectrogram and mel-spectrogram are as follows: Pre-emphasize the audio signal of the sound indicator data in the pet emotion data collected in real time; The pre-emphasized audio signal is framed and then further processed using a Hanning window function; The further processed audio signal is subjected to a fast Fourier transform to obtain the spectrum of each frame, and the spectrum of the speech is modulo-squared to obtain the spectrogram of the audio signal; Perform logarithmic calculation on the spectrogram to obtain the Mel-spectrogram.

[0089] Obtain the sound spectrogram, speech spectrogram and mel-spectrogram obtained by preprocessing the standard sound data in two emotion intervals; Emo calculated from standard sound data SY and emo SL The difference between the sound classification interval and the physiological classification interval of the corresponding emotion interval is the same as the middle value of the sound classification interval and the physiological classification interval, or the difference between the sound classification interval and the physiological classification interval is less than 3%.

[0090] Compare the three groups of spectra processed by the real-time collected pet emotion data with the three groups of spectra processed by two standard sound data, so as to obtain the sound spectrum similarity A1 and A2, the speech spectrum similarity B1 and B2, and the Mel spectrum similarity C1 and C2; Calculate two sets of comprehensive similarities, one set of comprehensive similarity is (A1+B1+C1) / 3, and the other set of comprehensive similarity is (A2+B2+C2) / 3.

[0091] Due to the pre-order calculation, it is only necessary to perform similarity analysis between the pet sound data and the standard sound data of two emotion classifications, and there is no need to perform similarity analysis with the standard sound data of other emotion classifications, so the amount of data to be compared is relatively low.

[0092] However, as the previous analysis shows, ordinary comparison can no longer accurately identify emotions, it is necessary to further process and analyze the pet sound data collected in real time. The use of sound spectrogram, speech spectrogram and Mel spectrum can clearly reflect the detailed characteristics of the sound data, thereby achieving accurate recognition and good results.

[0093] The present invention provides a pet emotion recognition system. In the process of identifying the emotions of a pet from the sound and physiological perspectives, the collected data is analyzed to extract local sound and physiological features, and then the extracted features are used to identify the pet's emotions. During the entire process, there is no need to compare the collected data with standard emotion data one by one, which can reduce the amount of data comparison, thereby greatly improving the speed of pet emotion recognition, having good use effect and having a good use prospect.

[0094] A pet emotion recognition method comprises the following steps: Obtain classified pet historical emotion data, real-time collected pet emotion data, and real-time collected environmental data, and then organize the classified pet historical emotion data, the historical emotion data includes historical sound index data and historical physiological index data, the historical sound index data is pet sound data, and the historical physiological index data includes pet breathing rate data and pet heart rate data; The sorted results are used to formulate sound classification intervals corresponding to different emotions of pets and physiological classification intervals corresponding to different emotions of pets according to preset standards; Compare the pet's physiological index data with the pet's standard sleep physiological index data, and then combine it with the real-time collected environmental data to calculate the index deviation value, and adjust the sound classification interval and the physiological classification interval according to the index deviation value to obtain the real-time sound classification interval and the real-time physiological classification interval; The obtained pet emotion data is further analyzed and calculated to obtain a pet emotion sound value and a pet emotion physiological value, and then the pet emotion sound value and the pet emotion physiological value are respectively compared and analyzed with the real-time sound classification interval and the real-time physiological classification interval to obtain two emotion classifications; The two emotion classifications are compared. If the two emotion classifications are the same, the pet emotion corresponding to the emotion classification is output; If the two emotion classifications are different, it is determined to be abnormal and an abnormal signal is issued; The voiceprint recognition algorithm is used to perform similarity analysis on the real-time collected pet sound data and the standard sound data of two emotion classifications to obtain two sets of comprehensive similarity data. The two sets of comprehensive similarity data are then compared to output the pet emotion corresponding to the emotion classification with the larger comprehensive similarity data.

[0095] The above formula is a formula obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0096] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0097] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0098] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A pet emotion recognition system, characterized in that: include: Data acquisition module: used to obtain classified pet historical emotion data, real-time collected pet emotion data and real-time collected environmental data, and then organize the classified pet historical emotion data, the historical emotion data includes historical sound index data and historical physiological index data, the historical sound index data is the pet sound data, and the historical physiological index data includes pet breathing rate data and pet heart rate data; Data analysis module: used to obtain the results sorted by the data acquisition module, and formulate the sound classification intervals corresponding to different emotions of the pet and the physiological classification intervals corresponding to different emotions of the pet according to the preset standards; Environmental analysis module: used to obtain the physiological index data of the pet when it is sleeping and the environmental data collected in real time, and compare the physiological index data of the pet with the standard physiological index data of the pet when it is sleeping, and then calculate the index deviation value in combination with the environmental data collected in real time, and adjust the sound classification interval and the physiological classification interval according to the index deviation value to obtain the real-time sound classification interval and the real-time physiological classification interval; Emotion analysis module: used to obtain the real-time collected pet emotion data, and further analyze and calculate the obtained pet emotion data, so as to obtain the pet emotion sound value and the pet emotion physiological value, and then compare and analyze the pet emotion sound value and the pet emotion physiological value with the real-time sound classification interval and the real-time physiological classification interval respectively, to obtain two emotion classifications; Emotion comparison module: used to obtain two emotion classifications and compare the two emotion classifications. If the two emotion classifications are the same, the pet emotion corresponding to the emotion classification is output; If the two emotion classifications are different, it is determined to be abnormal and an abnormal signal is issued; Exception handling module: used to receive abnormal signals and two emotion classifications, and use the voiceprint recognition algorithm to perform similarity analysis on the real-time collected pet sound data and the standard sound data of the two emotion classifications to obtain two sets of comprehensive similarity data, and then compare the two sets of comprehensive similarity data to output the pet emotion corresponding to the emotion classification with the larger comprehensive similarity data.

2. A pet emotion recognition system according to claim 1, characterized in that: The steps for formulating the sound classification interval and physiological classification interval according to the preset standards are as follows: S1. Obtain the collection time corresponding to the classified pet historical emotion data; S2. Query the environmental data of the pet's historical emotion data collection time and calculate the environmental impact value. The environmental data includes air quality index, temperature and humidity; S3, comparing the environmental impact value with the standard threshold, extracting the pet historical emotion data with environmental impact values ​​less than or equal to the standard threshold to construct a benchmark data set, and extracting the pet historical emotion data with environmental impact values ​​greater than the standard threshold to construct a special data set; S4, analyzing the data in the benchmark data set, calculating the pet emotion sound value based on the historical sound index data, and calculating the pet emotion physiological value based on the historical physiological index data; S5. Use a clustering algorithm to process the pet emotion sound value and the pet emotion physiological value to obtain the threshold interval of the corresponding emotion, thereby obtaining the sound classification interval and the physiological classification interval.

3. A pet emotion recognition system according to claim 2, characterized in that: The calculation formula of environmental impact value is as follows: ； Where IFh is the environmental impact value, n is the number of environmental data collected on the day corresponding to the time when the pet's historical emotional data was collected, and H j is the temperature data collected for the jth time, Bz W is the preset standard temperature data, xs W is the temperature proportionality coefficient, RH j is the humidity data collected for the jth time, Bz S is the preset standard humidity data, xs S is the influence coefficient of humidity ratio, AQI j is the air quality index data collected for the jth time, Bz AQI is the preset standard air quality index data, xs AQI is the influence coefficient of the air quality index ratio; The standard threshold is .

4. A pet emotion recognition system according to claim 3, characterized in that: The steps to calculate the pet emotional sound value based on historical sound indicator data are as follows: Process the historical sound index data to obtain a sound spectrum graph; The historical sound index data with a total duration of T is divided into m parts and the average value is calculated to obtain m groups of average amplitude data; The pet emotional sound value is calculated based on m groups of spectrum data. The specific calculation formula is as follows: ； In the formula, Emo SY is the pet's emotional sound value, A i is the average amplitude data of the i-th group after division, A j is the average amplitude data of the jth group after division, A max is the maximum value among the m groups of amplitudes, α and β are both constant coefficients, α>β>1, and Tb is the preset standard amplitude data; The formula for calculating the pet's emotional physiological value based on historical physiological indicator data is as follows: ； In the formula, Emo SL is the pet's emotional physiological value, HRj is the detected pet's heart rate data, HRb is the pet's heart rate data in a calm state, Rj is the detected pet's respiratory rate data, Rb is the pet's respiratory rate data in a calm state, and e is the natural constant data.

5. A pet emotion recognition system according to claim 4, characterized in that: The steps for calculating the indicator deviation value are as follows: calculating the environmental impact value IFH´ based on the environmental data collected in real time, and comparing the environmental impact value IFH´ with the standard threshold; If IFH´≤ , at this time the calculation formula of the indicator deviation value is: ； Wherein, Dev is the calculated index deviation value, HRs is the average heart rate data of the pet in the detected sleeping state, HRbs is the heart rate data of the pet in the standard sleeping state, Rs is the average respiratory rate data of the pet in the detected sleeping state, and Rbs is the average respiratory rate data of the pet in the standard sleeping state; If IFH´> , at this time, the calculation formula of the sentiment index deviation value is: ； Where K is the number of pet historical emotional data in the special data set with a difference ratio of IFh and IFH´ less than 5% and in the corresponding emotional state. The pet emotional sound value calculated by the number of pet historical emotional data corresponding to the emotional state of the i-th group, is the average value of the pet's emotional sound corresponding to the emotional state.

6. A pet emotion recognition system according to claim 5, characterized in that: The sound classification interval is (ESymin i , ESymax i ), the physiological classification interval is (ESlmin i , ESlmax i ), the calculated real-time sound classification interval is (ESymin i +ESymin i ×Dev,ESymax i +ESymax i ×Dev), the calculated real-time physiological classification interval is (ESlmin i +ESlmin i ×Dev,ESlmax i +ESlmax i ×Dev), i in the sound classification interval, physiological classification interval, real-time sound classification interval and real-time physiological classification interval is the i-th pet emotion.

7. A pet emotion recognition system according to claim 6, characterized in that: When comparing and analyzing the pet emotion sound value and the pet emotion physiological value with the real-time sound classification interval and the real-time physiological classification interval respectively, the distance D1 between the pet emotion sound value and the real-time sound classification interval corresponding to each emotion category and the distance D2 between the pet emotion physiological value and the real-time physiological classification interval corresponding to each emotion category are calculated; D1=Emo SYS -(ESymin i +ESymin i ×Dev+ESymax i +ESymax i ×Dev) / 2; D2=Emo SLS -(ESlmin i +ESlmin i ×Dev+ESlmax i +ESlmax i ×Dev) / 2; In the formula, Emo SYS The pet emotional sound value is calculated based on the real-time collected sound index data. SLS It is the pet's emotional physiological value calculated based on the physiological indicator data collected in real time.

8. A pet emotion recognition system according to claim 7, characterized in that: The standard sound data of the two emotion classifications are the historical sound index data closest to the middle value of the corresponding sound classification interval; The steps for using the voiceprint recognition algorithm to perform similarity analysis between the real-time collected pet sound data and the standard sound data of two emotion classifications are as follows: Preprocess the pet emotion data collected in real time to obtain sound spectrogram, speech spectrogram and mel spectrum graph; Obtain the sound spectrogram, speech spectrogram and mel-spectrogram obtained by preprocessing the standard sound data in two emotion intervals; Compare the three groups of spectra processed by the real-time collected pet emotion data with the three groups of spectra processed by two standard sound data, so as to obtain the sound spectrum similarity A1 and A2, the speech spectrum similarity B1 and B2, and the Mel spectrum similarity C1 and C2; Calculate two sets of comprehensive similarities, one set of comprehensive similarity is (A1+B1+C1) / 3, and the other set of comprehensive similarity is (A2+B2+C2) / 3.

9. A pet emotion recognition system according to claim 8, characterized in that: The steps for preprocessing the real-time collected pet emotion data to obtain the spectrogram and mel-spectrogram are as follows: Pre-emphasize the audio signal of the sound indicator data in the pet emotion data collected in real time; The pre-emphasized audio signal is framed and then further processed using a Hanning window function; The further processed audio signal is subjected to a fast Fourier transform to obtain the spectrum of each frame, and the spectrum of the speech is modulo-squared to obtain the spectrogram of the audio signal; Perform logarithmic calculation on the spectrogram to obtain the Mel-spectrogram.

10. A pet emotion recognition method, using any system described in claims 1 to 9, characterized in that: The steps include: Obtain classified pet historical emotion data, real-time collected pet emotion data, and real-time collected environmental data, and then organize the classified pet historical emotion data, the historical emotion data includes historical sound index data and historical physiological index data, the historical sound index data is pet sound data, and the historical physiological index data includes pet breathing rate data and pet heart rate data; The sorted results are used to formulate sound classification intervals corresponding to different emotions of pets and physiological classification intervals corresponding to different emotions of pets according to preset standards; Compare the pet's physiological index data with the pet's standard sleep physiological index data, and then combine it with the real-time collected environmental data to calculate the index deviation value, and adjust the sound classification interval and the physiological classification interval according to the index deviation value to obtain the real-time sound classification interval and the real-time physiological classification interval; The obtained pet emotion data is further analyzed and calculated to obtain a pet emotion sound value and a pet emotion physiological value, and then the pet emotion sound value and the pet emotion physiological value are respectively compared and analyzed with the real-time sound classification interval and the real-time physiological classification interval to obtain two emotion classifications; The two emotion classifications are compared. If the two emotion classifications are the same, the pet emotion corresponding to the emotion classification is output; If the two emotion classifications are different, it is determined to be abnormal and an abnormal signal is issued; The voiceprint recognition algorithm is used to perform similarity analysis on the real-time collected pet sound data and the standard sound data of two emotion classifications to obtain two sets of comprehensive similarity data. The two sets of comprehensive similarity data are then compared to output the pet emotion corresponding to the emotion classification with the larger comprehensive similarity data.

Images