A target detection and classification system and method based on data fusion

Abstract

The present invention discloses a target detection and classification system and method based on data fusion, which relates to the field of target recognition and detection technology. The system comprises a target detection and classification module, a waiting-for-work area planning module, a flight data analysis module and a target detection and recognition optimization module. The target is image-recognized and voiceprint-recognized by the target detection and classification module to determine the type of the target. The data collection effective area is spatially segmented by the waiting-for-work area planning module, and the historical image recognition and voiceprint recognition data of the target in the segmented space are analyzed to plan an optimal fusion and recognition area. The probability of the target flying into the optimal fusion and recognition area is analyzed by the flight data analysis module. The target detection reference model is established by the target detection and recognition optimization module to predict the probability that the current target will fly into the optimal fusion and recognition area. The detection and recognition method for the current target is planned, thereby reducing the collection of invalid data in the target detection and recognition process.

Description

Technical Field

[0001] The present invention relates to the technical field of data recognition, and in particular to a target detection and classification system and method based on data fusion. Background Art

[0002] Through target detection technology, the activities of flying targets around certain areas can be monitored and identified in real time. In order to reduce the missed recognition rate and improve the recognition accuracy, a combination of image recognition and voiceprint recognition can be used for target recognition and detection. Image recognition technology uses a camera to capture images and video footage of the target, and compares the captured information with the target features in the database to determine the type of target. Voiceprint recognition technology uses a sound recorder to record the target's calls, and compares the collected target's calls with the sound features in the known voiceprint database to determine the type of target.

[0003] However, image recognition and voiceprint recognition each have their own optimal recognition areas or ranges, and the two optimal recognition areas may overlap. Taking images and collecting sounds after the target appears in the overlapping area can improve the accuracy of subsequent target identification. When the existing technology performs image recognition and voiceprint recognition on the target, the working matching degree between the data collection devices is not high. Generally, the target data will be collected when the target appears within the recognizable range of the device, and then the respective target recognition work will be carried out. At this time, the collected target data may contain a large amount of data that leads to low recognition accuracy, which is not conducive to subsequent recognition work, and the target recognition efficiency and accuracy cannot be improved. Summary of the Invention

[0004] The purpose of the present invention is to provide a target detection and classification system and method based on data fusion to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solutions: a target detection and classification system based on data fusion, the system comprising a target detection and classification module, a waiting area planning module, a flight data analysis module, and a target detection and recognition optimization module;

[0006] Performing image recognition and voiceprint recognition on the target through the target detection and classification module to determine the type of the target;

[0007] The image acquisition effective area and the voiceprint acquisition effective area are spatially segmented by the to-be-worked area planning module, the historical image recognition and voiceprint recognition data of the target in the segmented space are analyzed, and the optimal fusion recognition area is planned based on the analysis results;

[0008] The flight data analysis module obtains the flight trajectory data of the historical target, selects the reference trajectory, and analyzes the probability of the target flying into the optimal fusion recognition area when the overlap between the target and the reference trajectory is different;

[0009] The target detection and recognition optimization module is used to establish a target detection reference model, predict the probability that the current target will fly to the optimal fusion recognition area, and plan the detection and recognition method for the current target based on the prediction result.

[0010] Preferably, the target detection and classification module includes an image data acquisition unit, a sound data collection unit, and a target detection and classification unit;

[0011] The image data acquisition unit uses an image capture device to capture a target image and transmits the target image data to the target detection and classification unit;

[0012] The sound data collection unit records the target's call using a sound recorder, and transmits the collected target's call data to the target detection and classification unit;

[0013] The target detection and classification unit compares the target image data with the target features in the database to determine the type of the target; and compares the target's call with the sound features in the known voiceprint database to determine the type of the target.

[0014] Preferably, the to-be-worked area planning module includes an identification area segmentation unit, a sub-region information statistics unit, and an optimal fusion identification area analysis unit;

[0015] The identification area segmentation unit divides a plurality of spherical spaces with the position of the image capture device as the sphere center, and uses the largest spherical space among them as a first spherical space. The first spherical space is the maximum data acquisition range of the image capture device, that is, the image capture device cannot capture images outside the first spherical space; divides a plurality of spherical spaces with the position of the sound recorder as the sphere center, and uses the largest spherical space among them as a second spherical space. The second spherical space is the maximum data acquisition range of the sound recorder, that is, the sound recorder cannot collect sounds outside the second spherical space;

[0016] The regional information statistics unit counts the number of historical target images captured by the image capture device when the historical target appears in a plurality of spherical spaces, as well as the number of target calls collected by the sound recorder, and counts the number of times the target species is correctly determined after feature comparison of the historical target images, and counts the number of times the target species is correctly determined after feature comparison of a plurality of collected target calls;

[0017] The optimal fusion recognition area analysis unit selects the optimal image acquisition area of the image capture device and the optimal sound collection area of the sound recorder based on the statistical results, and uses the overlapping area of the optimal image acquisition area and the optimal sound collection area as the optimal fusion recognition area.

[0018] Preferably, the flight data analysis module includes a historical flight data acquisition unit, a coincidence analysis unit, and a position and area matching probability analysis unit;

[0019] The historical flight data acquisition unit acquires the total number of historical targets that have flown into the first spherical space and the second spherical space, as well as the number of historical targets that have flown into the optimal fusion recognition area, to acquire the flight trajectories of the historical targets;

[0020] The coincidence analysis unit retrieves a random flight trajectory of a historical target that flew into the optimal fusion recognition area as a reference trajectory, and analyzes the coincidence between the flight trajectories of the remaining targets and the reference trajectory, wherein the remaining targets refer to the targets that flew into the optimal fusion recognition area except the retrieved target;

[0021] The position and area matching probability analysis unit counts the ratio of the number of targets in different spaces when the overlap with the reference trajectory is different, and analyzes the probability of the target flying into the optimal fusion recognition area when the overlap with the reference trajectory is different.

[0022] Preferably, the target detection and recognition optimization module includes a target detection reference model establishment unit and a detection mode planning unit;

[0023] The target detection reference model establishment unit establishes a target detection reference model using the overlap and probability of the flight trajectories between targets as training samples;

[0024] The detection method planning unit obtains the overlap between the generated flight trajectory of the current target and the reference trajectory, inputs it into the target detection reference model, predicts the probability that the current target will fly to the optimal fusion recognition area, and plans the detection and recognition method for the current target based on the prediction result.

[0025] A target detection and classification method based on data fusion includes the following steps:

[0026] S100: Perform image recognition and voiceprint recognition on the target to determine the type of the target;

[0027] S101: spatially segmenting the effective image acquisition area and the effective voiceprint acquisition area, analyzing the historical image recognition and voiceprint recognition data of the target within the segmented space, and planning the optimal fusion recognition area based on the analysis results;

[0028] S102: Acquire historical target flight trajectory data, select a reference trajectory, and analyze the probability of the target flying into the optimal fusion recognition area when the overlap between the target and the reference trajectory is different;

[0029] S103: Establish a target detection reference model, predict the probability that the current target will fly to the optimal fusion recognition area, and plan the detection and recognition method for the current target based on the prediction result.

[0030] Preferably, S100 includes: using an image capture device to capture a target image, comparing the target image data with target features in a database, and determining the type of the target; using a sound recorder to record the target's call, and comparing the collected target's call with sound features in a known voiceprint database to determine the type of the target.

[0031] Preferably, the step S101 includes: dividing n spherical spaces with the location of the image capturing device as the sphere center, wherein the diameters of the n spherical spaces are {R1, R2, ... R n}, where R i >R i-1 The difference between two adjacent diameters is a fixed value, i refers to the i-th spherical space, i=2,3,...n, that is, the difference between the diameters of any two adjacent items is equal. The spherical space with the largest diameter among the n spaces is taken as the first spherical space, and the first spherical space is the maximum data acquisition range of the image capture device. The m spherical spaces with the location of the sound recorder as the sphere center are divided in the same way, and the spherical space with the largest diameter among the m spaces is taken as the second spherical space, and the second spherical space is the maximum data acquisition range of the sound recorder. When the historical target appears in a random spherical space in the n spaces, the number of images of the corresponding historical target captured by the image capture device is U. After recognizing K images out of the U images, the type of the corresponding historical target can be correctly determined. The recognition accuracy w of the target appearing in a random spherical space is calculated. j , w j =K / U, and the set of recognition accuracy for capturing and recognizing the target appearing in the n spherical space is w={w1,w2,...w j ,...w n}, screen out the spherical space with the highest recognition accuracy as the best image acquisition area of the image shooting device, and screen out the spherical space with the highest recognition accuracy as the best sound collection area of the sound recorder in the same way. If there is an overlapping area between the best image acquisition area and the best sound collection area, the overlapping area of the best image acquisition area and the best sound collection area is used as the best fusion recognition area; if there is no overlapping area, the best fusion recognition area is not divided, and the target is recognized by a separate recognition judgment method, that is, when the current target appears in the first spherical space, the target image is captured, and when the current target appears in the second spherical space, the target call is collected, and image recognition and voiceprint recognition are performed respectively, and the respective recognition results are transmitted to the monitoring center respectively;

[0032] Based on the purpose of planning the image and sound data acquisition and detection method, taking into account that the target can be correctly identified when it does not appear in all positions within the effective acquisition range of the data acquisition device, the acquisition space is divided into several spherical spaces with the device as the center, and the diameter difference between each two is equal, which ensures the fairness and rationality of the subsequent setting of the best data acquisition area. After the space segmentation is completed, the recognition accuracy of the target data collected and identified in all spherical spaces is analyzed, and the best image acquisition area and the best sound collection area are set according to the recognition accuracy. If there is an overlapping area between the areas, it is judged that the accuracy of data collection and target recognition at two levels is higher when the target appears in the overlapping area, so the overlapping area is set as the best fusion recognition area.

[0033] Preferably, S102 includes: obtaining a total number of historical targets A that have flown into the first spherical space or the second spherical space in the past, obtaining a number of historical targets that have flown into the optimal fusion recognition area in the past, obtaining flight trajectories of the historical targets, retrieving a random flight trajectory of the historical target that has flown into the optimal fusion recognition area as a reference trajectory, wherein the length of the reference trajectory is L1, setting k discrete points on the reference trajectory, obtaining a set of discrete point coordinates, traversing the flight trajectories of the remaining targets, counting the number of coordinates in the flight trajectory of the remaining random target that exist in the set of discrete point coordinates as g, and calculating the degree of overlap W between the flight trajectory of the remaining random target and the reference trajectory. e :

[0034] W e =[g / min(D e ,L1)]*100%;

[0035] Among them, D e Represents the flight trajectory length of another random target, and the obtained overlap set is W={W1,W2,...W e ,...W k}, k is the number of remaining targets. The flight trajectories of B1 targets among A historical targets have a degree of overlap with the reference trajectory of W1. The probability P1 of the target flying into the optimal fusion recognition area when the overlap with the reference trajectory is W1 is obtained. P1=B1 / A, and the probability set is P={P1,P2,...P k}, P k The degree of overlap with the reference trajectory is W k The probability that the target will fly into the optimal fusion recognition area when .

[0036] Preferably, the step S103 includes: forming a training sample {(W1, P1), (W2, P2), ... (W k ,P k )}, after fitting the training samples, the target detection reference model is established: y=φ1*x+φ2, φ1 and φ2 represent the fitting coefficients, and the overlap between the generated flight trajectory of the current target and the reference trajectory is obtained as W ’ , let x=W ’ , the probability that the current target will fly to the best fusion recognition area is φ1*W ’ +φ2, set the probability threshold to P ’ , if φ1*W ’ +φ2>P ’ , predict that the current target will fly to the best fusion recognition area, suspend the image and sound data collection work, and wait until the current target flies to the best fusion recognition area and then collect image and sound data at the same time to identify the type of the current target. If the image recognition result and the voiceprint recognition result are the same, the type of the current target is confirmed; otherwise, re-collect image and sound data until the type of the current target is confirmed, and transmit the confirmed type information of the current target to the monitoring center; if φ1*W ’ +φ2≤P ’ , predicting that the current target will not fly to the optimal fusion recognition area, the image and sound data collection work will not be suspended, and image recognition and voiceprint recognition will be performed respectively, and the respective recognition results will be transmitted to the monitoring center;

[0037] By collecting historical data and analyzing the target's flight trajectory, the flight trajectory of one of the targets appearing in the optimal fusion recognition area is used as the baseline trajectory. The probability that the target will fly into the optimal fusion recognition area when the overlap with the baseline trajectory is different is analyzed. The overlap and probability data are combined into training samples, and a target detection reference model is established. The model is used as a reference to predict whether the current target will fly into the optimal fusion recognition area. If so, data collection is suspended for a period of time. After the target flies into the optimal fusion recognition area, image and sound data collection is performed simultaneously. This reduces invalid data that leads to low recognition accuracy and improves target recognition efficiency and accuracy.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] The present invention divides the acquisition space into several spherical spaces with the device as the center, and the diameter difference between each two is equal, which ensures the fairness and rationality of the subsequent setting of the optimal data acquisition area. After the space segmentation is completed, the recognition accuracy of the target data collected and identified in all spherical spaces is analyzed, and the optimal image acquisition area and the optimal sound collection area are set according to the recognition accuracy. If there is an overlapping area between the areas, it is determined that the accuracy of data acquisition and target recognition at two levels is higher when the target appears in the overlapping area, so the overlapping area is set as the optimal fusion recognition area.

[0040] After setting up the optimal fusion recognition area, historical data is collected and the flight trajectory of the target is analyzed. The flight trajectory of one of the targets appearing in the optimal fusion recognition area is used as the baseline trajectory. The probability that the target will fly into the optimal fusion recognition area when the overlap with the baseline trajectory is different is analyzed. The overlap and probability data are combined into training samples, and a target detection reference model is established. The model is used as a reference to predict whether the current target will fly into the optimal fusion recognition area. If so, data collection is suspended for a period of time. After the target flies into the optimal fusion recognition area, image and sound data collection is performed simultaneously. This reduces invalid data that leads to low recognition accuracy and improves target recognition efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a structural diagram of a target detection and classification system based on data fusion according to the present invention;

[0042] Figure 2 The figure is a flow chart of a target detection and classification method based on data fusion according to the present invention. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0044] Example 1: Figure 1 As shown, this embodiment provides a target detection and classification system based on data fusion, the system includes: a target detection and classification module, a waiting work area planning module, a flight data analysis module and a target detection and recognition optimization module;

[0045] The target detection and classification module performs image recognition and voiceprint recognition on the target to determine the type of the target;

[0046] The working area planning module is used to spatially segment the effective image acquisition area and the effective voiceprint acquisition area, analyze the historical image recognition and voiceprint recognition data of the target in the segmented space, and plan the optimal fusion recognition area based on the analysis results;

[0047] The flight data analysis module obtains historical target flight trajectory data, selects a baseline trajectory, and analyzes the probability of the target flying into the optimal fusion recognition area when the overlap between the target and the baseline trajectory is different.

[0048] The target detection and recognition optimization module establishes a target detection reference model, predicts the probability that the current target will fly to the optimal fusion recognition area, and plans the detection and recognition method for the current target based on the prediction results.

[0049] The target detection and classification module includes an image data acquisition unit, a sound data collection unit and a target detection and classification unit; the image data acquisition unit uses an image capture device to capture target images and transmits the target image data to the target detection and classification unit; the sound data collection unit uses a sound recorder to record the target's call and transmits the collected target's call data to the target detection and classification unit; the target detection and classification unit compares the target image data with the target features in the database to determine the type of the target; the target's call is compared with the sound features in the known voiceprint database to determine the type of the target.

[0050] The area planning module to be worked on includes an identification area segmentation unit, a sub-area information statistics unit and an optimal fusion identification area analysis unit; the identification area segmentation unit divides a number of spherical spaces with the position of the image shooting device as the sphere center, and takes the largest spherical space as the first spherical space. The first spherical space is the maximum data acquisition range of the image shooting device, that is, the image shooting device cannot capture images outside the first spherical space; divides a number of spherical spaces with the position of the sound recorder as the sphere center, and takes the largest spherical space as the second spherical space. The second spherical space is the maximum data acquisition range of the sound recorder, that is, the sound recorder cannot collect images outside the second spherical space. sound; using a regional information statistics unit to count the number of historical target images captured by the image capture device when the historical targets appear in several spherical spaces and the number of target calls collected by the sound recorder, the number of times the target type is correctly judged after feature comparison of the historical target images, and the number of times the target type is correctly judged after feature comparison of several collected target calls; using an optimal fusion recognition area analysis unit to screen out the optimal image collection area of the image capture device and the optimal sound collection area of the sound recorder based on the statistical results, and the area where the optimal image collection area and the optimal sound collection area overlap is used as the optimal fusion recognition area.

[0051] The flight data analysis module includes a historical flight data acquisition unit, an overlap analysis unit, and a position and area matching probability analysis unit; the total number of historical targets that have flown into the first spherical space and the second spherical space in the past, as well as the number of historical targets that have flown into the best fusion recognition area in the past are obtained through the historical flight data acquisition unit, and the flight trajectories of the historical targets are obtained; the overlap analysis unit randomly retrieves the flight trajectory of a historical target that has flown into the best fusion recognition area as a reference trajectory, and analyzes the overlap between the flight trajectories of the remaining targets and the reference trajectory, where the remaining targets refer to the targets that fly into the best fusion recognition area except the retrieved targets; the position and area matching probability analysis unit counts the proportion of the number of targets in different spaces when the overlap with the reference trajectory is different, and analyzes the probability of the target flying into the best fusion recognition area when the overlap with the reference trajectory is different.

[0052] The target detection and recognition optimization module includes a target detection reference model establishment unit and a detection method planning unit; the target detection reference model establishment unit uses the overlap and probability of the flight trajectories between targets as training samples to establish a target detection reference model; the detection method planning unit obtains the overlap between the generated flight trajectory of the current target and the reference trajectory, inputs it into the target detection reference model, predicts the probability that the current target will fly to the optimal fusion recognition area, and plans the detection and recognition method for the current target based on the prediction result.

[0053] Example 2: Figure 2 As shown, this embodiment provides a target detection and classification method based on data fusion, which is implemented based on the detection system in the embodiment and specifically includes the following steps:

[0054] S100: Perform image recognition and voiceprint recognition on the target to determine the type of the target: use an image capture device to capture the target image, compare the target image data with the target features in the database, and determine the type of the target; use a sound recorder to record the target's call, and compare the collected target's call with the sound features in the known voiceprint database to determine the type of the target;

[0055] S101: Perform spatial segmentation on the effective image acquisition area and the effective voiceprint acquisition area, analyze the historical image recognition and voiceprint recognition data of the target in the segmented space, and plan the best fusion recognition area based on the analysis results: divide the image capture device location into n spherical spaces with the diameter of the n spherical spaces being {R1, R2, ... R n}, where R i >R i-1 The difference between two adjacent diameters is a fixed value, i refers to the i-th spherical space, i=2,3,...n, that is, the difference between the diameters of any two adjacent items is equal. The spherical space with the largest diameter among the n spaces is taken as the first spherical space, and the first spherical space is the maximum data acquisition range of the image capture device. The m spherical spaces with the location of the sound recorder as the sphere center are divided in the same way, and the spherical space with the largest diameter among the m spaces is taken as the second spherical space, and the second spherical space is the maximum data acquisition range of the sound recorder. When the historical target appears in a random spherical space in the n spaces, the number of images of the corresponding historical target captured by the image capture device is U. After recognizing K images out of the U images, the type of the corresponding historical target can be correctly determined. The recognition accuracy w of the target appearing in a random spherical space is calculated. j , w j =K / U, and the set of recognition accuracy for capturing and recognizing the target appearing in the n spherical space is w={w1,w2,...w j ,...w n}, select the spherical space with the highest recognition accuracy as the best image acquisition area of the image shooting device, and select the spherical space with the highest recognition accuracy as the best sound collection area of the sound recorder in the same way. The screening method of the best sound collection area is the same as the screening method of the best image acquisition area: when the historical target appears in a random spherical space in m spaces, the sound recorder collects a total of a calls corresponding to the historical target. After performing voiceprint recognition on b of the collected calls, the target type is correctly judged. The recognition accuracy rate of collecting and identifying the target appearing in a random spherical space among m spherical spaces is b / a, and the accuracy rate of collecting and identifying the target appearing in m spherical spaces is obtained. The accuracy of sound collection and recognition of the target in the spherical space is used to select the spherical space with the highest recognition accuracy as the optimal sound collection area of the sound recorder. If there is an overlapping area between the optimal image collection area and the optimal sound collection area, the overlapping area of the optimal image collection area and the optimal sound collection area is used as the optimal fusion recognition area. If there is no overlapping area, the optimal fusion recognition area is not divided, and the target is recognized by a separate recognition judgment method, that is, when the current target appears in the first spherical space, the target image is captured, and when the current target appears in the second spherical space, the target call is collected, and image recognition and voiceprint recognition are performed separately, and the respective recognition results are transmitted to the monitoring center.

[0056] S102: Obtain flight trajectory data of historical targets, select a reference trajectory, and analyze the probability of the target flying into the optimal fusion recognition area when the overlap between the target and the reference trajectory is different: obtain the total number of historical targets that have flown into the first spherical space or the second spherical space in the past as A, obtain the number of historical targets that have flown into the optimal fusion recognition area in the past, obtain the flight trajectory of the historical targets, retrieve the flight trajectory of a random historical target that has flown into the optimal fusion recognition area as the reference trajectory, the length of the reference trajectory is L1, set k discrete points on the reference trajectory, obtain the discrete point coordinate set, traverse the flight trajectories of the remaining targets, count the number of coordinates in the discrete point coordinate set in the flight trajectory of the remaining random target as g, and calculate the overlap W between the flight trajectory of the remaining random target and the reference trajectory. e :

[0057] W e =[g / min(D e ,L1)]*100%;

[0058] Among them, D e Represents the flight trajectory length of another random target, and the obtained overlap set is W={W1,W2,...W e ,...W k}, k is the number of remaining targets. The flight trajectories of B1 targets among A historical targets have a degree of overlap with the reference trajectory of W1. The probability P1 of the target flying into the optimal fusion recognition area when the overlap with the reference trajectory is W1 is obtained. P1=B1 / A, and the probability set is P={P1,P2,...P k}, P k The degree of overlap with the reference trajectory is W k The probability of the target flying into the best fusion recognition area when ;

[0059] S103: Build a target detection reference model, predict the probability that the current target will fly to the optimal fusion recognition area, and plan the detection and recognition method for the current target based on the prediction result: form a training sample {(W1, P1), (W2, P2), ... (W k ,P k )}, after fitting the training samples, the target detection reference model is established: y=φ1*x+φ2, φ1 and φ2 represent fitting coefficients, * represents the multiplication sign, and the overlap between the generated flight trajectory of the current target and the reference trajectory is obtained as W ’ , let x=W ’ , the probability that the current target will fly to the best fusion recognition area is φ1*W ’ +φ2, set the probability threshold to P ’ , if φ1*W ’ +φ2>P ’ , predict that the current target will fly to the best fusion recognition area, suspend the image and sound data collection work, and wait until the current target flies to the best fusion recognition area and then collect image and sound data at the same time to identify the type of the current target. If the image recognition result and the voiceprint recognition result are the same, the type of the current target is confirmed; otherwise, re-collect image and sound data until the type of the current target is confirmed, and transmit the confirmed type information of the current target to the monitoring center; if φ1*W ’ +φ2≤P ’ , predicting that the current target will not fly to the optimal fusion recognition area, the image and sound data collection work will not be suspended, and image recognition and voiceprint recognition will be performed respectively, and the respective recognition results will be transmitted to the monitoring center;

[0060] For example, taking the target as a flying bird, when detecting and identifying the target, the overlap between the current bird's flight trajectory and the reference trajectory is 0.83. After inputting 0.83 into the target detection reference model, it is predicted that the probability that the current bird will fly to the best fusion recognition area is 0.90. The probability threshold is set to 0.85, 0.90>0.85, and it is predicted that the current bird will fly to the best fusion recognition area. The image and sound data collection work is suspended. When the current bird flies into the best fusion recognition area, the image and sound data are collected simultaneously to identify the species of the current bird. If the image recognition result and the voiceprint recognition result are the same, the species of the current bird is confirmed.

[0061] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A target detection and classification system based on data fusion, characterized by: The system includes: Target detection and classification module, work area planning module, flight data analysis module, and target detection and recognition optimization module; Performing image recognition and voiceprint recognition on the target through the target detection and classification module to determine the type of the target; The image acquisition effective area and the voiceprint acquisition effective area are spatially segmented by the to-be-worked area planning module, the historical image recognition and voiceprint recognition data of the target in the segmented space are analyzed, and the optimal fusion recognition area is planned based on the analysis results; The flight data analysis module obtains the flight trajectory data of the historical target, selects the reference trajectory, and analyzes the probability of the target flying into the optimal fusion recognition area when the overlap between the target and the reference trajectory is different; The target detection and recognition optimization module establishes a target detection reference model, predicts the probability that the current target will fly to the optimal fusion recognition area, and plans the detection and recognition method for the current target based on the prediction result; The working area planning module includes an identification area segmentation unit, a sub-region information statistics unit and an optimal fusion identification area analysis unit; The identification area segmentation unit divides a plurality of spherical spaces with the position of the image capture device as the sphere center, and uses the largest spherical space among them as a first spherical space, which is the maximum data acquisition range of the image capture device; divides a plurality of spherical spaces with the position of the sound recorder as the sphere center, and uses the largest spherical space among them as a second spherical space, which is the maximum data acquisition range of the sound recorder; The regional information statistics unit counts the number of historical target images captured by the image capture device when the historical target appears in a plurality of spherical spaces, as well as the number of target calls collected by the sound recorder, and counts the number of times the target species is correctly determined after feature comparison of the historical target images, and counts the number of times the target species is correctly determined after feature comparison of a plurality of collected target calls; The optimal fusion recognition area analysis unit selects the optimal image acquisition area of the image capture device and the optimal sound collection area of the sound recorder based on the statistical results, and uses the overlapping area of the optimal image acquisition area and the optimal sound collection area as the optimal fusion recognition area.

2. The target detection and classification system based on data fusion according to claim 1, characterized in that: The target detection and classification module includes an image data acquisition unit, a sound data collection unit and a target detection and classification unit; The image data acquisition unit uses an image capture device to capture a target image and transmits the target image data to the target detection and classification unit; The sound data collection unit records the target's call using a sound recorder, and transmits the collected target's call data to the target detection and classification unit; The target detection and classification unit compares the target image data with the target features in the database to determine the type of the target; Compare the target's call with the sound features in the known voiceprint database to determine the type of target.

3. The target detection and classification system based on data fusion according to claim 1, characterized in that: The flight data analysis module includes a historical flight data acquisition unit, a coincidence analysis unit, and a position and area matching probability analysis unit; The historical flight data acquisition unit acquires the total number of historical targets that have flown into the first spherical space and the second spherical space, as well as the number of historical targets that have flown into the optimal fusion recognition area, to acquire the flight trajectories of the historical targets; The coincidence analysis unit retrieves a random flight trajectory of a historical target that flies to the optimal fusion recognition area as a reference trajectory, and analyzes the coincidence between the flight trajectories of the remaining targets and the reference trajectory; The position and area matching probability analysis unit counts the ratio of the number of targets in different spaces when the overlap with the reference trajectory is different, and analyzes the probability of the target flying into the optimal fusion recognition area when the overlap with the reference trajectory is different.

4. The target detection and classification system based on data fusion according to claim 3, characterized in that: The target detection and recognition optimization module includes a target detection reference model establishment unit and a detection mode planning unit; The target detection reference model establishment unit establishes a target detection reference model using the overlap and probability of flight trajectories between targets as training samples; The detection method planning unit obtains the overlap between the generated flight trajectory of the current target and the reference trajectory, inputs it into the target detection reference model, predicts the probability that the current target will fly to the optimal fusion recognition area, and plans the detection and recognition method for the current target based on the prediction result.

5. A target detection and classification method based on data fusion, characterized by: The following steps are involved: S100: Perform image recognition and voiceprint recognition on the target to determine the type of the target; S101: spatially segmenting the effective image acquisition area and the effective voiceprint acquisition area, analyzing the historical image recognition and voiceprint recognition data of the target within the segmented space, and planning the optimal fusion recognition area based on the analysis results; S102: Acquire historical target flight trajectory data, select a reference trajectory, and analyze the probability of the target flying into the optimal fusion recognition area when the overlap between the target and the reference trajectory is different; S103: Establishing a target detection reference model, predicting the probability that the current target will fly to the optimal fusion recognition area, and planning a detection and recognition method for the current target based on the prediction result; The step S101 includes: dividing n spherical spaces with the location of the image capturing device as the sphere center, wherein the diameters of the n spherical spaces are {R1, R2, ... R n }, where R i >R i-1 The difference between two adjacent diameters is a fixed value, i refers to the i-th spherical space, i=2,3,...n, the spherical space with the largest diameter among the n spaces is taken as the first spherical space, and the first spherical space is the maximum data acquisition range of the image capture device. The m spherical spaces with the location of the sound recorder as the sphere center are divided in the same way, and the spherical space with the largest diameter among the m spaces is taken as the second spherical space, and the second spherical space is the maximum data acquisition range of the sound recorder. When the historical target appears in a random spherical space in the n spaces, the number of images of the corresponding historical target captured by the image capture device is U. After recognizing K images out of the U images, the type of the corresponding historical target can be correctly determined. The recognition accuracy w of capturing and recognizing the target appearing in the random spherical space is calculated. j , w j =K / U, and the set of recognition accuracy for capturing and recognizing the target appearing in the n spherical space is w={w1,w2,...w j ,...w n }, screen out the spherical space with the highest recognition accuracy as the best image collection area of the image shooting device, and screen out the spherical space with the highest recognition accuracy as the best sound collection area of the sound recorder in the same way. If there is an overlapping area between the best image collection area and the best sound collection area, the overlapping area between the best image collection area and the best sound collection area will be used as the best fusion recognition area; if there is no overlapping area, the best fusion recognition area will not be divided.

6. The target detection and classification method based on data fusion according to claim 5, characterized in that: The S100 includes: using an image capture device to capture a target image, comparing the target image data with target features in a database, and determining the type of the target; using a sound recorder to record the target's call, and comparing the collected target's call with sound features in a known voiceprint database to determine the type of the target.

7. The target detection and classification method based on data fusion according to claim 5, characterized in that: The step S102 includes: obtaining a total number of historical targets A that have flown into the first spherical space or the second spherical space in the past, obtaining a number of historical targets that have flown into the optimal fusion recognition area in the past, obtaining flight trajectories of the historical targets, retrieving a random flight trajectory of the historical target that has flown into the optimal fusion recognition area as a reference trajectory, wherein the length of the reference trajectory is L1, setting k discrete points on the reference trajectory, obtaining a set of discrete point coordinates, traversing the flight trajectories of the remaining targets, counting the number of coordinates in the flight trajectory of the remaining random target that exist in the set of discrete point coordinates as g, and calculating a degree of overlap W between the flight trajectory of the remaining random target and the reference trajectory. e : W e =[g / min(D e ,L1)]*100%; Among them, D e Represents the flight trajectory length of another random target, and the obtained overlap set is W={W1,W2,...W e ,...W k }, k is the number of remaining targets. The flight trajectories of B1 targets among A historical targets have a degree of overlap with the reference trajectory of W1. The probability P1 of the target flying into the optimal fusion recognition area when the overlap with the reference trajectory is W1 is obtained. P1=B1 / A, and the probability set is P={P1,P2,...P k }, P k The degree of overlap with the reference trajectory is W k The probability that the target will fly into the optimal fusion recognition area when .

8. The target detection and classification method based on data fusion according to claim 7, characterized in that: The step S103 includes: forming a training sample {(W1, P1), (W2, P2), ... (W k ,P k )}, after fitting the training samples, the target detection reference model is established: y=φ1*x+φ2, φ1 and φ2 represent the fitting coefficients, and the overlap between the generated flight trajectory of the current target and the reference trajectory is obtained as W ’ , let x=W ’ , the probability that the current target will fly to the best fusion recognition area is φ1*W ’ +φ2, set the probability threshold to P ’ , if φ1*W ’ +φ2>P ’ , predict that the current target will fly to the best fusion recognition area, suspend the image and sound data collection work, and wait until the current target flies to the best fusion recognition area and then collect image and sound data at the same time to identify the type of the current target. If the image recognition result and the voiceprint recognition result are the same, the type of the current target is confirmed; otherwise, re-collect image and sound data until the type of the current target is confirmed, and transmit the confirmed type information of the current target to the monitoring center; if φ1*W ’ +φ2≤P ’ , predicting that the current target will not fly to the optimal fusion recognition area, without suspending the image and sound data collection work, perform image recognition and voiceprint recognition respectively, and transmit their respective recognition results to the monitoring center.

Images