CN112541547B - Weighted fusion weight determination method supporting underwater rapid target recognition - Google Patents
Weighted fusion weight determination method supporting underwater rapid target recognition Download PDFInfo
- Publication number
- CN112541547B CN112541547B CN202011466539.8A CN202011466539A CN112541547B CN 112541547 B CN112541547 B CN 112541547B CN 202011466539 A CN202011466539 A CN 202011466539A CN 112541547 B CN112541547 B CN 112541547B
- Authority
- CN
- China
- Prior art keywords
- sensor
- weight
- criterion
- calculating
- sensors
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
- G01H3/00—Measuring characteristics of vibrations by using a detector in a fluid
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Measurement Of Velocity Or Position Using Acoustic Or Ultrasonic Waves (AREA)
Abstract
The invention discloses a weighted fusion weight determination method for supporting underwater rapid target identification, and relates to a weighted fusion weight determination method. The invention relates to a weighted fusion weight determination method. The process is as follows: 1: the method comprises the following steps that (1) sound data are collected by n sensors, and the sound data collected by each sensor are divided into m sections; 2: calculating the sensor support degree and the sensor similarity among the sensors; 3: calculating local stability and local support in the sensor; 4: calculating an entropy weight coefficient based on an entropy weight method; 5: calculating the cumulative contribution rate of the criterion index; 6: determining a criterion index of each sound signal segment; 7: determining the weight corresponding to each sound signal segment; 8: dividing original sound signal data into acceleration data and constant speed data; respectively giving a weight to the acceleration and the uniform speed, and multiplying the weight of the acceleration and the uniform speed by the weight corresponding to the sound signal segment to be used as a second-order weight; 9: and determining the category of the underwater rapid target. The method is used for the field of underwater rapid target identification.
Description
Technical Field
The invention relates to a weighted fusion weight determination method.
Background
The information fusion method is always the key point in information fusion, and the feature fusion is based on a plurality of feature extraction methods, and is used for extracting features of the same identification object, so that the aim of improving the target identification capability and the model generalization capability is achieved. Feature fusion has been widely applied in the field of image recognition, such as face recognition, content-based image retrieval, etc., and with the continuous and intensive research of various scholars, many achievements have been obtained in the field of image recognition, but the field of voice recognition has few related researches, and most of the research directions are in the field of voice recognition, such as emotion recognition. The related research based on non-voice signals is still in the field of conventional single feature extraction and classifier combination, more research is on the improvement of a new feature extraction method and a classifier of the sound signals, and the research of fusion recognition based on the existing features is seriously insufficient. The feature level information fusion is now gradually becoming the research focus in the information fusion field.
At present, a great deal of research on weight determination is based on the decision of criteria, such as correlation function, distance and the like, the single criteria has small operand and high calculation speed, but no accepted criteria can be widely accepted, so the decision based on the single criteria is too subjective and unilateral.
In the traditional weighted fusion algorithm, researchers assign corresponding weights to the sensors according to the quality of data or after judging from multiple angles. However, in conjunction with the background of the present study, this uniform weighting of sensor data for the identification of underwater fast targets is too coarse, as shown, for example. According to the background of the research, the sensors collect data by using hydrophones, each hydrophone has an area with high quality of collected information, so that the variation of the quality of the collected data caused by the arrangement area of the hydrophones can be caused along with the variation of the orientation of an object. Although the area with high signal collected by the hydrophone is set by the specific model at the factory, the area with high signal quality collected by the hydrophone is a sector area, so the signal quality collected by the hydrophone may frequently change underwater. For underwater fast targets, the advantage of fast speed is the characteristic. The underwater target dynamic weight method is different from other underwater targets, particularly when the underwater target dynamic weight method is close to an object to be attacked, the underwater target is fast in azimuth change due to the fact that the underwater target dynamic weight method is fast in speed, and the significance of the dynamic weight is obvious.
Disclosure of Invention
The invention aims to solve the problem of low identification accuracy of the existing underwater quick target, and provides a weighting fusion weight determination method for supporting underwater quick target identification.
The method for determining the weighted fusion weight for supporting underwater rapid target identification comprises the following specific processes:
step 1: n number of sensors s 1 ,s 2 ,...,s n Collecting sound data, and equally dividing the sound data collected by each sensor into m sections;
the sensors are hydrophones, and n hydrophones form a hydrophone array;
step 2: calculating the sensor support degree and the sensor similarity among the sensors;
and 3, step 3: calculating local stability and local support in the sensor;
and 4, step 4: calculating an entropy weight coefficient based on an entropy weight method;
and 5: calculating the cumulative contribution rate of the criterion index;
and 6: determining the criterion index of each sound signal segment according to the accumulated contribution rate;
and 7: determining the weight corresponding to each sound signal segment according to the entropy weight coefficient calculated in the step 4;
and 8: dividing original sound signal data into acceleration and uniform speed according to waveforms;
respectively giving a weight to each of the acceleration and the uniform speed, multiplying the weight of the acceleration by the weight corresponding to the sound signal segment determined in the step 7, and multiplying the weight of the uniform speed by the weight corresponding to the sound signal segment determined in the step 7 to serve as a second-order weight;
the weight for acceleration is greater than 1 and less than or equal to 3;
the weight of the uniform speed is more than or equal to 0 and less than or equal to 1;
and step 9: and determining the category of the underwater fast target based on the second-order weight obtained in the step 8.
The beneficial effects of the invention are as follows:
the invention provides an improved dynamic weight determination method by combining the characteristics of a quick target after considering the application background as the underwater quick target.
The invention adopts a weighting method to perform multi-sensor feature fusion, and the weighting method has the advantages of simplicity, high efficiency and the like. Each sensor is given a weight by a corresponding algorithm, and then the extracted features are multiplied by the weights and then fused. Compared with series fusion, the weighted feature fusion has small calculation amount, and the fused features have smaller dimensionality. Although simple to understand, the weighted average method is complex and important in determining the weight. How to determine the weight of each sensor is the central weight of the weighted average method. Therefore, the invention adopts the multi-criterion index tree to make various decisions to calculate the weight more comprehensively and more finely.
The traditional weight assignment is equivalent to setting the segment weight to be the same for each segment weight, and the invention aims to assign the corresponding weight to each small segment according to the specific condition of each segment and the comparison between sensors, so that the error can be reduced, the accuracy is improved, and the traditional weighting method is more necessary to be improved based on the characteristic of the underwater rapid target.
When the sensor array monitors an underwater fast target, the quality of data received by the sensor array can be greatly influenced by the position of the target. Because the present invention is concerned with underwater fast targets, the sensors use hydrophones, each of which has an angular range in which the received signal quality is high. For example, the effect of the sensor 2 at the starting point is better, the effect of the sensor 3 at the end point is better, and in practical application, the situation is much more complicated because the moving direction of the underwater fast target changes rapidly, and a segmented dynamic weight is provided based on the situation.
The invention mainly completes the algorithm design of the dynamic second-order weight method based on the multi-criterion index tree. Firstly, a dynamic weight is provided according to the characteristics of the underwater target, the target is segmented, and a refined weight is expected to be given. The weight calculation method adopts a multi-criterion tree mode to judge the sound quality from multiple angles, and meanwhile, according to the characteristics of the underwater rapid target, a high weight is given to the acceleration process, and a low weight is given to the uniform speed process for improvement. And then, judging the importance of the criterion by an entropy weight-principal component analysis method, and finally giving the score of each section of sound to determine the weight.
The dynamic weight of the invention can solve the problem that the static weight does not consider the target direction.
The target is moving, which results in different sound quality for each sensor, so the invention divides into small time segments and gives him a dynamic weight.
The multi-criterion method solves the problem of one-sided comparison of single criteria, and then the multi-criterion method can enable different data set effects to be stable.
According to the characteristics of the underwater rapid target, the method provides the speed characteristic weight to optimize the weight.
Therefore, the underwater rapid target identification accuracy is improved, and the problem of low identification accuracy of the existing underwater rapid target is solved.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a diagram illustrating the weight criteria of the present invention.
Detailed Description
The first embodiment is as follows: the weighted fusion weight determination method supporting underwater rapid target identification in the embodiment specifically comprises the following processes:
underwater fast targets such as torpedoes;
step 1: n sensors s 1 ,s 2 ,...,s n Collecting sound data, and equally dividing the sound data collected by each sensor into m sections;
the sensor is a hydrophone, and n hydrophones form a hydrophone array;
step 2: calculating the sensor support degree and the sensor similarity among the sensors;
and step 3: calculating local stability and local support in the sensor;
and 4, step 4: calculating an entropy weight coefficient based on an entropy weight method;
and 5: calculating the cumulative contribution rate of the criterion index;
obtaining an accumulated contribution rate according to the characteristic root;
in the formula of lambda k Is the root of the feature of matrix R in step 412;
step 6: determining the criterion index of each sound signal segment according to the accumulated contribution rate;
F t ＝E 1t X 1 +E 2t X 2 +…+E nt X n
and 7: determining the weight corresponding to each sound signal segment according to the entropy weight coefficient calculated in the step 4;
F＝w 1 F 1 +w 2 F 2 +…+w p F p
and 8: dividing original sound signal data into acceleration and uniform speed according to waveforms;
respectively giving a weight to each of the acceleration and the uniform speed, multiplying the weight of the acceleration by the weight corresponding to the sound signal segment determined in the step 7, and multiplying the weight of the uniform speed by the weight corresponding to the sound signal segment determined in the step 7 to serve as a second-order weight;
the weight for acceleration is greater than 1 and less than or equal to 3;
the weight of the uniform speed is more than or equal to 0 and less than or equal to 1;
the specific process is as follows:
weighted optimization based on underwater fast target
The weight importance calculated by the method is multiplied by the characteristics to carry out characteristic fusion, so that higher weight is necessary for important characteristics, the characteristics of the underwater rapid target cannot be ignored except the importance calculated according to the indexes, and the underwater rapid target has some unique characteristics. The underwater fast target outlet pipe generally comprises an underwater fast target outlet pipe with a launching pipe opened and a back-filling water outlet pipe. After the underwater fast target is launched out of the pipe, the pipe is ignited and then accelerated for a period of time, and the underwater fast target finally accelerates to a very high speed due to the fast target, so that the acceleration process can be obviously shown on a line spectrum of a time-frequency plane. Generally, underwater fast objects are accelerated by a propeller, which has to reach a high rotational speed in order to finally enter a constant speed stage at a high speed, and the acceleration process lasts for several seconds. The acceleration stage of the underwater fast target can generate extremely complex radiation noise, which is extremely significant for target identification, namely, the important part in a section of underwater fast target sound signal, namely, the underwater fast target is subjected to a process of accelerating firstly and then keeping the speed constant from the exit pipe to the hit target, and the acceleration process is more significant for detection and identification of the underwater fast target. And (4) providing a speed characteristic weight according to the characteristic of the underwater rapid target, and taking the speed characteristic weight and the calculated weight together as a second-order weight. The idea of dynamic weighting is proposed to segment the radiated noise and calculate the weight of each segment, whereas for the velocity characteristic weight, the sound signal acquired by the whole sensor is divided into two parts of acceleration and uniform velocity according to the oscillogram. The speed characteristic weight in the acceleration process is a value greater than 1, while the speed characteristic weight in the constant speed process is a value less than 1, and the specific value needs to be determined through experiments.
The underwater rapid target exit pipe is firstly accelerated, and the fusion formula is shown as the following formula:
wherein L is a post-fusion feature; sigma 1 Is the acceleration stage velocity feature weight; w n The weight of the nth section of the sound signal of the sensor is calculated according to the weight; l is a radical of an alcohol n Extracting features of the nth section of the sound signal section of the sensor through a convolutional neural network;
after the acceleration process, the process enters a constant speed process, and the formula is shown as the following formula
In the formula, σ 2 The speed characteristic weight of the uniform speed stage.
And step 9: and determining the type (ship or torpedo) of the underwater fast target based on the second-order weight obtained in the step 8.
The second embodiment is as follows: the difference between the present embodiment and the first embodiment is that, in the step 2, the sensor support degree and the sensor similarity between the sensors are calculated; the specific process is as follows:
the weight is determined by adopting an algorithm based on a multi-criterion index tree, the one-sidedness of a single criterion can be avoided by determining the weight through multiple criteria, and rich criteria can prevent a sensor which should not be endowed with high weight from playing an important role in weighted fusion because of the excellence of a certain aspect. The idea of segment weights is introduced in step one, so the weights calculated in this step are the weights of the individual sensors at a certain time after segmentation, e.g., { σ } in step one 12 ,σ 22 ,σ 32 ...,σ n2 }. The comparison of the sound inside the analyzed section of sound and the sound of the corresponding section of the sound of other sensors is used as the criteria of two major aspects inside the sensors and between the sensors, and the two major criteria are divided into a plurality of sub-criteria. Due to the fact that underwater environment is variable, noise mixed in the obtained sound is inevitable, but the fact is just the meaning of the existence of multiple sensors, and errors are reduced as much as possible through the multiple sensors. The inside of the sensor is compared with the inside of the analyzed sound segment, and the noise of the sound segment sensor is judged according to the performance of the inside of the analyzed sound segment, so that the influence of the noise is reduced by giving smaller weight to the sound segment with more noise and giving larger weight to the sound segment with less noise. The criterion between the sensors is to analyze the analyzed sound segment with the sound segments of other sensors in the same time segment, and if the other sensors support the analyzed sound segment, it indicates that the sound segment is supported highly, and a higher weight can be given.
The basic idea of the inter-sensor criterion is to determine the quality of the sound signal measured by the sensor to be analyzed by the sound signal measured by the other sensors. In section 4.2, dynamic weights are proposed and their significance is elucidated, so that the quality of the sensor after segmentation in a time period is also studied in this chapter. The inter-sensor criterion is to compare the quality of other sensors with the quality of the target sensor in the same time period to obtain a result. The purpose of the chapter is information fusion, the final purpose of the fusion is target identification, the data source is a sound signal measured in a sensor array, but the problem at present is that no standard exists, and what sound signal is most meaningful for identification can not be found through standard comparison. The sensor array for detecting the underwater fast target must meet the condition that most of signals measured by the sensors are high in quality, and on the basis of the condition, comparison and verification among the sensors are effective.
Step 21: calculating the sensor support degree among the sensors; the specific process is as follows:
the sensor support degree represents the support degree of other sensors to the sensor to be analyzed at a certain moment in the sensor array; the DTW distance is used as an index for judging the support degree; the DTW distance is a typical optimization problem, has a good effect on measuring the similarity of two time sound time sequence signals, and does not require that the two signals are always equal in length.
The method adopts the idea of dynamic programming, uses a time warping function to express the corresponding relation between a target sequence and the sequence, and finally obtains a minimum accumulated distance which is the result. The DTW distance calculation process is roughly to start matching from the beginning, and assuming that two time series a and B now, the current shortest distance is selected from each point to the beginning for superposition until the last end distance and the shortest distance between the two time series a and B are the required DTW distance, and the DTW distance represents the similarity between the time series a and B.
Constructing a distance matrix D by means of DTW distances between the sensor to be analyzed and other sensors n Distance matrix D n Representing the support of each sensor in the sensor array at a certain time;
in the formula, d 12 DTW distance between the first and second sensors; the sum of the ith row elements of the matrix is the sensor support value of the ith sensor; 1,2, n;
step 22: calculating sensor similarity between sensors; the specific process is as follows:
the sensor similarity is a criterion for judging the data quality collected by the sensor at a certain moment from a characteristic level;
a sensor collects sound signals and extracts MFCC characteristic vectors from the sound signals;
the MFCC is a Mel frequency cepstrum coefficient;
the similarity of the feature vectors of two sensors collecting sound signals at a certain moment is defined by using a radial basis function, and the traditional Euclidean distance is replaced by the DTW distance between the two feature vectors, as shown in the following formula
Wherein σ' is the base width; s ij Collecting similarity of feature vectors of sound signals for two sensors at a certain time; s (Q) i ,Q j ) Is the DTW distance between two feature vectors;
when s (Q) i ,Q j ) When the value is 0, the two eigenvectors are completely similar, namely the similarity is equal to 1;
the average similarity of the ith sensor after normalization with other sensors is as follows:
in the formula, n is the number of sensors; i is not equal to j;
other steps and parameters are the same as those in the first embodiment.
The third concrete implementation mode: the difference between the present embodiment and the first or second embodiment is that, in the step 3, the local stability and the local support degree in the sensor are calculated; the specific process is as follows:
internal criteria of sensor
The determination of the criteria in the multi-criteria model is the most important, and since the research background of the text is an underwater rapid target, a category of criteria based on the inside of the sensor is provided. Because the underwater fast target is in a complex marine environment, interference on target radiation noise and various marine noises and air noises generated by the complex and variable environment generate great interference for passive identification through the hydrophone. Therefore, under the condition that the underwater radiation noise is not obvious generally nowadays, the evaluation of the noise in a section of indexes with high and low radiation noise quality plays an important role.
Step 31: calculating local stability within the sensor; the specific process is as follows:
the local stability represents the consistency of a section of sound signals of a sensor in time and reflects the stability degree of the sensor in a period of time. Firstly, in order to realize the non-dimensionalization of the measurement data, the normalized variance of the f-th measurement parameter of a section of sound signal of the i-th sensor is set as follows:
wherein, f is 1,2,., z; u is the number of samples into which one of the m sound signal segments to be analyzed is again sliced,
is the mean value of the f-th measured parameter, v f,a 'is the a' th measurement of the f-th measurement parameter; v. of f,max ,v f,min Maximum and minimum values of the f-th measurement parameter in the sound signal segment to be analyzed (maximum and minimum values of one segment in the m segments); sigma f The variance of the f-th measured parameter;
the variance of a sound signal segment is represented by the mean value of the measured parameter, the local stability of which can be expressed as
Wherein St (ε) is the normalized variance ε with the measured parameter f Measure of interest, mapping local stability to [0, 1%]The larger the value is, the higher the local stability of the sound signal segment to be analyzed is; st is integrated, and the contrast of local stability can be increased by adjusting S';
the mean normalized variance for the z measurement parameters; s' is a constant;
step 32: calculating local support degree in the sensor; the specific process is as follows:
the local support can show the noise in a section of sound signal, and is similar to the sensor support. Firstly, dividing sound data collected by an ith sensor into m sections; one of the m sound signal segments to be analyzed is cut into a segments { l }again 1 ,l 2 ,…l a Calculating the DTW distance between the sections, and calculating the local support value of the ith sensor according to the DTW distance between the sections:
in the formula (I), the compound is shown in the specification,
is the first 1 And l 2 The DTW distance between the segments; SUP (super) i Local support value of the ith sensor;
the larger the numerical value is, the lower the support degree of the inner parts is, and the worse the local support degree is;
determining the importance of the criterion according to an entropy weight-principal component analysis method, and multiplying the normalized criterion result to obtain a final weight;
in the index evaluation method based on the entropy weight-principal component analysis method, under the condition that the prior knowledge of a decision maker is little or not available, accurate evaluation of each index which is high or low is difficult to be given, so that the corresponding active judgment method cannot achieve the desired effect, such as a statistical analysis method, a network analysis method and the like. This section contemplates the use of objective weighting to accomplish the evaluation of criteria. The objective weighting method is to obtain results through some mathematical methods, and to ensure complete objectivity without subjective intention judgment, the results are obtained through analysis according to the characteristics of data. The main idea of the principal component analysis method is dimension reduction, and the original index is converted into a few irrelevant indexes by calculating the variance contribution rate of the index. The principal component analysis method has the advantages that firstly, redundancy among indexes can be solved by reducing the number of the indexes, and the original indexes are converted into a group of new indexes with less number according to contribution rate; secondly, objectivity, the principal component analysis method is one of objective weighting methods, naturally has the advantages of the objective weighting methods, calculates results only according to variance contribution rate, and has no subjective intention; thirdly, the method can adapt to different data, and the number of indexes can be adjusted according to the situation. The basic idea of the entropy weight method is to reflect the degree of distinction of the index from the sample through the degree of disorder of the index. The smaller the entropy value, the more ordered the data, the greater the difference between samples, the greater the ability to discriminate between the objects to be evaluated, and a greater weight should be given if only the entropy value is considered. The entropy weight method has the following advantages: the first is that the sample discrimination ability is only scored according to the judged index, so that the calculation is easy, and the understanding is simple and easy; the second is also strong objectivity and has no influence of subjective factors.
Other steps and parameters are the same as those in the first or second embodiment.
The fourth concrete implementation mode: the difference between this embodiment and one of the first to third embodiments is that, in the step 4, an entropy weight coefficient is calculated based on an entropy weight method; the specific process is as follows:
step 41: dividing the data of the criterion index into positive and negative indexes, and normalizing the data of the criterion index;
the data of the criterion index are: sensor support and sensor similarity between sensors; and local stability and local support within the sensor;
local stability within the sensor is a positive indicator;
the sensor support among sensors, the sensor similarity and the local support in the sensors are negative indexes;
for example, the DTW distance used in the sensor support is a negative index, i.e., the larger the value, the better the result, and the sensor similarity is a positive index, i.e., the larger the value, the better the result.
For the forward indicator:
for the negative indicators:
in the formula, t ik The ith evaluation object is shown (the evaluation object is a sound signal break; the evaluation object is at the same time, the same time is provided with a plurality of sensors, the same time is provided with a plurality of sound signal breaks, the same time is provided with N sensors, and the moment is provided with N sound signal breaks);
the segmented sound signal segment m segments of the sensor) in the kth criterion index (the sensor support degree and the sensor similarity among the sensors; and local stability and local support within the sensor; ) Index value of (1), t 11 A first criterion representing a first sensor, namely a value of sensor support;
the index value is normalized;
meanwhile, since logarithm calculation is needed when the entropy weight is calculated, one-step operation is needed
The value becomes positive, the formula:
where σ "is the amplitude of the transform;
step 42: calculating the proportion of the ith evaluation object criterion value under the kth criterion index to obtain a matrix R, wherein each element R ik Comprises the following steps:
wherein k is 1,2, …, m ', m' is a standard number;
step 43: according to the proportion of the ith evaluation object criterion value under the kth criterion index, calculating the entropy weight of the kth criterion through an entropy weight formula; the concrete formula is as follows:
step 44: calculating a difference coefficient of the criterion according to the entropy weight of the k-th criterion;
step 45: and calculating the entropy weight coefficient according to the difference coefficient of the criterion.
Other steps and parameters are the same as those in one of the first to third embodiments.
The fifth concrete implementation mode: the difference between this embodiment and the first to the fourth embodiment is that, in step 43, the entropy weight of the kth criterion is calculated through an entropy weight formula according to the specific gravity of the ith evaluation object criterion value under the kth criterion index; the concrete formula is as follows:
wherein K is a constant.
Other steps and parameters are the same as in one of the first to fourth embodiments.
The sixth specific implementation mode: this embodiment is different from one of the first to fifth embodiments in that in the step 44
Calculating a difference coefficient of the criterion according to the entropy weight of the kth criterion; the concrete formula is as follows:
q k ＝1-e k
in the formula, q k Is a coefficient of difference;
a large q value indicates that the discrimination capability is stronger and the effect is larger.
Other steps and parameters are the same as those in one of the first to fifth embodiments.
The seventh embodiment: the difference between this embodiment and the first to sixth embodiments is that, in the step 45, the entropy weight coefficient is calculated according to the difference coefficient of the criterion; the specific process is as follows:
other steps and parameters are the same as those in one of the first to sixth embodiments.
The present invention is capable of other embodiments and its several details are capable of modifications in various obvious respects, all without departing from the spirit and scope of the present invention.
Claims (6)
1. The method for determining the weighted fusion weight for supporting underwater rapid target identification is characterized by comprising the following steps: the method comprises the following specific processes:
step 1: n number of sensors s 1 ,s 2 ,...,s n Collecting sound data, and dividing the sound data collected by each sensor into m sections;
the sensors are hydrophones, and n hydrophones form a hydrophone array;
step 2: calculating the sensor support degree and the sensor similarity among the sensors;
and step 3: calculating local stability and local support in the sensor;
and 4, step 4: calculating an entropy weight coefficient based on an entropy weight method;
and 5: calculating the cumulative contribution rate of the criterion index;
step 6: determining the criterion index of each sound signal segment according to the accumulated contribution rate;
and 7: determining the weight corresponding to each sound signal segment according to the entropy weight coefficient calculated in the step 4;
and 8: dividing original sound signal data into acceleration and uniform speed according to waveforms;
respectively giving a weight to each of the acceleration and the uniform speed, multiplying the weight of the acceleration by the weight corresponding to the sound signal segment determined in the step 7, and multiplying the weight of the uniform speed by the weight corresponding to the sound signal segment determined in the step 7 to serve as a second-order weight;
the weight for acceleration is greater than 1 and less than or equal to 3;
the weight of the uniform speed is more than or equal to 0 and less than or equal to 1;
and step 9: determining the category of the underwater rapid target based on the second-order weight obtained in the step 8;
calculating local stability and local support degree in the sensor in the step 3; the specific process is as follows:
step 31: calculating local stability within the sensor; the specific process is as follows:
the normalized variance of the f-th measurement parameter of a section of sound signal of the i-th sensor is set as follows:
wherein, f is 1, 2.., z; u is the number of samples into which one of the m sound signal segments to be analyzed is again sliced,
is the mean value of the f-th measured parameter, v f,a′ Is the a' th measured value of the f-th measured parameter; v. of f,max ,v f,min Maximum value and minimum value of the f-th measurement parameter in the sound signal segment to be analyzed; sigma f For the f-th measurementVariance of the parameter;
the variance of a sound signal segment is represented by the mean value of the measured parameter, the local stability of the sound signal segment being represented by
In the formula (I), the compound is shown in the specification,
the mean normalized variance of the z measurement parameters; s' is a constant;
step 32: calculating local support within the sensor; the specific process is as follows:
one of the m sound signal segments to be analyzed is cut into a segments { l }again 1 ,l 2 ,…l a Calculating the DTW distance between the sections, and calculating the local support value of the ith sensor according to the DTW distance between the sections:
in the formula (I), the compound is shown in the specification,
is the first 1 And l 2 The DTW distance between the segments; SUP (super) i Is the local support value of the ith sensor.
2. The method for determining weighted fusion weights supporting underwater fast object recognition according to claim 1, wherein: calculating the sensor support degree and the sensor similarity among the sensors in the step 2; the specific process is as follows:
step 21: calculating the sensor support degree among the sensors; the specific process is as follows:
the sensor support degree represents the support degree of other sensors to the sensor to be analyzed at a certain moment in the sensor array; the DTW distance is used as an index for judging the support degree;
constructing a distance matrix D by means of DTW distances between the sensor to be analyzed and other sensors n Distance matrix D n Representing the support of each sensor in the sensor array at a certain time;
in the formula (d) 12 DTW distance between the first and second sensors; the sum of the ith row elements of the matrix is the sensor support value of the ith sensor; 1,2, n;
step 22: calculating sensor similarity between sensors; the specific process is as follows:
a sensor collects sound signals and extracts MFCC characteristic vectors from the sound signals;
the MFCC is a Mel frequency cepstrum coefficient;
the similarity of the feature vectors of the two sensors collecting sound signals at a certain moment is defined by using the radial basis function, and the traditional Euclidean distance is replaced by the DTW distance between the two feature vectors, which is shown in the following formula
Wherein σ' is the base width; s ij Collecting similarity of feature vectors of sound signals for two sensors at a certain time; s (Q) i ,Q j ) Is the DTW distance between two feature vectors;
when s (Q) i ,Q j ) When the value is 0, the two eigenvectors are completely similar, namely the similarity is equal to 1;
the average similarity of the ith sensor after normalization with other sensors is as follows:
in the formula, n is the number of sensors; i ≠ j.
3. The method for determining weighted fusion weight supporting underwater fast target recognition according to claim 2, wherein: in the step 4, entropy weight coefficients are calculated based on an entropy weight method; the specific process is as follows:
step 41: dividing data of the criterion index into positive and negative indexes, and normalizing the data of the criterion index;
the data of the criterion index are: sensor support and sensor similarity between sensors; and local stability and local support within the sensor;
local stability within the sensor is a positive indicator;
the sensor support among the sensors, the sensor similarity and the local support in the sensors are negative indexes;
for the forward indicator:
for the negative indicators:
in the formula, t ik An index value representing the ith evaluation object on the kth criterion index;
the index value is normalized;
step 42: calculating the proportion of the ith evaluation object criterion value under the kth criterion index to obtain a matrix R, wherein each element R ik Comprises the following steps:
wherein k is 1,2, …, m ', m' is a standard number;
step 43: according to the proportion of the ith evaluation object criterion value under the kth criterion index, calculating the entropy weight of the kth criterion through an entropy weight formula;
and step 44: calculating a difference coefficient of the criterion according to the entropy weight of the k-th criterion;
step 45: and calculating the entropy weight coefficient according to the difference coefficient of the criterion.
4. The method for determining weighted fusion weights supporting underwater fast object recognition according to claim 3, wherein: in the step 43, according to the proportion of the ith evaluation object criterion value under the kth criterion index, the entropy weight of the kth criterion is calculated through an entropy weight formula; the concrete formula is as follows:
wherein K is a constant.
5. The method for determining weighted fusion weights supporting underwater fast object recognition according to claim 4, wherein: calculating a difference coefficient of the criterion according to the entropy weight of the k-th criterion in the step 44; the concrete formula is as follows:
q k ＝1-e k
in the formula, q k Is the coefficient of difference.
6. The method for determining weighted fusion weights supporting underwater fast object recognition according to claim 5, wherein: in the step 45, the entropy weight coefficient is calculated according to the difference coefficient of the criterion; the concrete formula is as follows:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011466539.8A CN112541547B (en) | 2020-12-14 | 2020-12-14 | Weighted fusion weight determination method supporting underwater rapid target recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011466539.8A CN112541547B (en) | 2020-12-14 | 2020-12-14 | Weighted fusion weight determination method supporting underwater rapid target recognition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112541547A CN112541547A (en) | 2021-03-23 |
| CN112541547B true CN112541547B (en) | 2022-09-02 |
Family
ID=75018595
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011466539.8A Active CN112541547B (en) | 2020-12-14 | 2020-12-14 | Weighted fusion weight determination method supporting underwater rapid target recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112541547B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114220458B (en) * | 2021-11-16 | 2024-04-05 | 武汉普惠海洋光电技术有限公司 | Voice recognition method and device based on array hydrophone |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104504740A (en) * | 2015-01-23 | 2015-04-08 | 天津大学 | Image fusion method of compressed sensing framework |
| CN111093244A (en) * | 2019-12-25 | 2020-05-01 | 华北电力大学 | Wireless sensor network route optimization method and system |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7694555B2 (en) * | 2007-03-27 | 2010-04-13 | Gm Global Technology Operations, Inc. | Brake pad prognosis system |
| CN101883425A (en) * | 2010-06-04 | 2010-11-10 | 哈尔滨工程大学 | Target tracking and identification device and method based on entropy-weighted gray correlation |
| US8577640B2 (en) * | 2011-08-17 | 2013-11-05 | Invensense, Inc. | Magnetometer bias and anomaly detector |
| CN109444728A (en) * | 2018-09-21 | 2019-03-08 | 国网河南省电力公司济源供电公司 | A kind of circuit breaker failure diagnostic method based on dynamic weighting Hybrid Clustering Algorithm |
| CN109858526B (en) * | 2019-01-08 | 2023-08-18 | 沈阳理工大学 | Multi-target track fusion method based on sensor in target tracking |
-
2020
- 2020-12-14 CN CN202011466539.8A patent/CN112541547B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104504740A (en) * | 2015-01-23 | 2015-04-08 | 天津大学 | Image fusion method of compressed sensing framework |
| CN111093244A (en) * | 2019-12-25 | 2020-05-01 | 华北电力大学 | Wireless sensor network route optimization method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112541547A (en) | 2021-03-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Gerosa et al. | Scream and gunshot detection in noisy environments | |
| US8271283B2 (en) | Method and apparatus for recognizing speech by measuring confidence levels of respective frames | |
| US6178399B1 (en) | Time series signal recognition with signal variation proof learning | |
| US8428945B2 (en) | Acoustic signal classification system | |
| US9984677B2 (en) | Bettering scores of spoken phrase spotting | |
| Becker et al. | Forensic speaker verification using formant features and Gaussian mixture models. | |
| Eryildirim et al. | Pulse Doppler radar target recognition using a two-stage SVM procedure | |
| US5761639A (en) | Method and apparatus for time series signal recognition with signal variation proof learning | |
| CN113284513B (en) | Method and device for detecting false voice based on phoneme duration characteristics | |
| CN115457966B (en) | Pig cough sound identification method based on improved DS evidence theory multi-classifier fusion | |
| CN112541547B (en) | Weighted fusion weight determination method supporting underwater rapid target recognition | |
| CN117727307B (en) | Bird voice intelligent recognition method based on feature fusion | |
| Sigmund | Statistical analysis of fundamental frequency based features in speech under stress | |
| JP3004023B2 (en) | Voice recognition device | |
| Cunha et al. | Applying non-destructive testing and machine learning to ceramic tile quality control | |
| CN113345443A (en) | Marine mammal vocalization detection and identification method based on mel-frequency cepstrum coefficient | |
| KR101671305B1 (en) | Apparatus for extracting feature parameter of input signal and apparatus for recognizing speaker using the same | |
| CN109410968B (en) | Efficient detection method for initial position of voice in song | |
| CN111681671B (en) | Abnormal sound identification method and device and computer storage medium | |
| CN114373453A (en) | Voice keyword detection method based on motion trail and discriminative information | |
| US7292981B2 (en) | Signal variation feature based confidence measure | |
| Yamamoto et al. | Using acoustic dissimilarity measures based on state-level distance vector representation for improved spoken term detection | |
| Bouffaut et al. | Automated blue whale song transcription across variable acoustic contexts | |
| Yerramreddy et al. | Speaker Identification Using MFCC Feature Extraction: A Comparative Study Using GMM, CNN, RNN, KNN and Random Forest Classifier | |
| Kasahara et al. | Weakly supervised defect detection using acoustic data based on positive and negative constraints |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |