A target agglomeration implementation method of dynamic connectivity processing

Abstract

The application provides a target condensation implementation method of dynamic connectivity processing, and belongs to the technical field of radar signal processing, and specifically comprises the following steps: in combination with a processing period and a space use condition of engineering implementation, a distance length coding algorithm is selected as a real-time implementation of two-dimensional matrix connectivity, wherein a distance length coding obtaining step is that threshold crossing points in a single pulse are firstly merged according to distance, and the merging of the single pulse is coded through a start distance and an end distance; weighted fast merging, when calculating a root node of a certain node each time, a node checked along a path is also pointed to the root node, a tree is maximized to be flattened, and through a recursive method, the root node is found while an index number thereof is pointed to the root node. Through the processing scheme of the application, the threshold crossing detection points are subjected to target condensation processing under the premise of meeting the real-time requirement of the radar, and the distance and azimuth accuracy of target measurement is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of radar signal processing, in particular to a target condensation implementation method of dynamic connectivity processing. BACKGROUND

[0002] The main function of the sea search radar is to realize the detection of the sea surface target. The RCS of the sea surface target has a large span, ranging from 1 to 100000 m 2 , spanning 5 orders of magnitude. In order to improve the detection performance of small targets and reduce the influence of sea clutter, the sea search radar begins to use high-resolution waveforms. With the improvement of resolution, multiple threshold detection points of the sea surface target are more likely to appear, becoming an extended target. The extended target will cause a substantial increase in the subsequent target tracking processing data, so after detection, all threshold detection points need to be condensed into the same target, not only to reduce the pressure of subsequent data processing, but also to calculate the centroid of the target in the condensation process, improve the distance and azimuth accuracy of target measurement.

[0003] The master thesis "Implementation of Constant False Alarm Rate and Target Condensation Algorithm Based on TMS320C6678" searched on China National Knowledge Infrastructure, which describes the multi-core parallel development technology, mainly introduces the multi-core synchronization technology based on IPC interrupt and the multi-core synchronization technology based on semaphore module. Both methods need to complete the synchronization and communication between cores, as well as the update of shared data. The overall design of the thesis processes by frame period, which takes a long time, needs a large amount of cached data, and has a large hardware space resource overhead. For fast timing, large sampling real-time systems, this method is not suitable, especially for dense sea surface small targets, which cannot be effectively distinguished after target condensation. SUMMARY

[0004] Therefore, the present application provides a target condensation implementation method of dynamic connectivity processing, which solves the problems in the prior art and improves the distance and azimuth accuracy of target measurement under the premise of meeting the real-time requirements of the radar.

[0005] The target condensation implementation method of dynamic connectivity processing provided by the present application adopts the following technical scheme:

[0006] A target condensation implementation method of dynamic connectivity processing includes the following steps:

[0007] According to the processing period and space usage of the engineering implementation, the distance length coding algorithm is selected as the real-time implementation of the two-dimensional matrix connectivity, wherein the distance length coding acquisition step is to first merge the threshold points in a single pulse according to the distance, and encode the merging of a single pulse through the start distance and the end distance.

[0008] Weighted fast merge, each time the root node of a certain node is calculated, the nodes along the path are also directed to the root node, the tree is maximally flattened, and through the recursive method, the index number of the root node is directed to the root node while finding the root node.

[0009] Optionally, the weighted fast merge uses the halving path compression algorithm, and the nodes are constantly moved upwards until all the nodes are connected to the root node.

[0010] Optionally, in the inter-pulse merging process, the OpenMP parallel processing mode is used, each core is allocated a certain number of distance segments, and multiple cores are parallel to perform the inter-pulse merging process, and the boundaries of the segmented edges are processed separately.

[0011] Optionally, in the engineering implementation, the path compression weighted merging method is used to establish and delete the tree in a real-time dynamic process, and the root buffer pool and the leaf buffer pool are created for dynamic management; the leaf buffer pool uses the space of the new and old buffer pools to alternately perform dynamic management, the root buffer pool is managed by the number of leaf nodes, the management of the root buffer pool includes the initialization process, the increase of the number of leaf nodes, the decrease of the number of leaf nodes and the cleaning, and the root buffer pool is dynamically managed by setting the state, in the pulse merging process, the idle position in the root buffer pool is dynamically searched according to the state mark, and dynamic management is realized.

[0012] Optionally, the storage space of the multi-core processor includes the LL2 memory space unique to each core and the shared SL2 memory space, through decomposition of the algorithm steps, the processing data of the root buffer pool and the leaf buffer pool are placed in the LL2 memory space independent of each core in the parallel process, and finally the output data of the target condensation is uniformly synchronized to the shared space, and the same data is processed in the shared space through this design.

[0013] Optionally, it includes:

[0014] Step 1, data initialization, the measurement information of the threshold detection point is input and arranged into a new leaf buffer pool through OpenMP multi-core parallel processing;

[0015] Step 2, inter-pulse association processing, realized through OpenMP multi-core parallel processing, using binary search method to associate the old and new leaf buffers, if the association condition is met, it is judged whether it has been associated before, if it has been associated, the root node association pair needs to be further obtained, finally the pulse number of the old node is updated, the position index number of the old node is assigned to the position index number of the new node, and the distance association pair and the root node association pair are output.

[0016] Step 3, new leaf node is associated to root processing, using OpenMP multi-core parallel processing to traverse all new leaf nodes, if the new leaf node is associated, the current leaf node is connected to the root, and the number of leaf nodes of the root is added by 1, if the new leaf node is not associated, the minimum number of position with state vector being 0 is selected in the root buffer pool, the current position root node is initialized, and the new node is associated to the root buffer pool;

[0017] Step 4, old leaf node arrangement processing, using OpenMP multi-core parallel processing to realize old leaf node arrangement function, including deletion, reservation and update, traversing all old leaf nodes, connecting the old leaf node to the root, calculating the pulse number difference, and processing according to the pulse number difference;

[0018] Step 5, update operation feature processing, using OpenMP multi-core parallel processing to realize traversal of all new nodes and root associated pairs, assigning effective nodes to corresponding root node positions, recording angular resolution forced splitting record maximum power vector, and land segment processing operation;

[0019] Step 6, segment boundary processing. Through OpenMP multi-core parallel processing, the boundary between segments is found, and all valid updates in the buffer pool are processed; all distance gate number start and end that meet the conditions are found; the minimum value of the start distance gate and the maximum value of the end distance gate are found, and the number and distance gate number start and end are stored in the array sequence;

[0020] Step 7, segment boundary merging processing, single-core serial loop traverses all segments to get the node at the end of the current segment and the node at the beginning of the next segment, if the two nodes meet the association condition and one of them is in the output state, the output node is updated to the other node, and the output node is cleaned up;

[0021] Step 8, target output and cleaning processing, using OpenMP multi-core parallel processing to traverse all root buffers, if the index number in the root buffer pool is not equal to the position itself, cleaning is performed, if the state in the root buffer pool is output, indicating that the output operation needs to be performed, the target number of the output buffer pool is added by 1, the centroid of the distance and azimuth is calculated, and the azimuth is expanded, the false alarm and target of the output target are judged, and finally the effective target information attribute is output.

[0022] Optionally, in step 4, if the pulse number difference is equal to 0, the number of leaf nodes of the corresponding root is reduced by 1;

[0023] If the pulse number difference is greater than the deletion threshold value, the leaf node deletion condition is met, the root node number is reduced by 1, and if the current root leaf node number is 0, it indicates that all leaf nodes of the current root node have been deleted, and need to be output, and the root node state is set to output 2;

[0024] If the pulse node difference is greater than 0 and less than the deletion threshold value, the current old leaf node has no new leaf node associated with it, and also does not meet the deletion condition, the old node is added to the free place in the new buffer pool, and in the next pulse processing process, the new buffer becomes the old buffer, and can still be retained in the buffer pool.

[0025] In summary, the present application includes the following beneficial technical effects:

[0026] The OpenMP parallel model is simple and easy to use. The main parallelization work can be realized by 3-4 instructions, and the ability of coarse-grained and fine-grained parallelization can be realized. When multi-threaded task synchronization, instructions can be used to complete, or the area enclosed by the braces {} can be used to complete.

[0027] The method has short processing time and fully meets the real-time requirements of radar signal processing. Taking the processing capacity of target condensation input as 1056 measurement points as an example, the target condensation using the dynamic connection processing mode takes about 450us, which meets the real-time requirements of fast time sequence radar system.

[0028] The memory space resources of the hardware platform are saved. Through dynamic management, the memory consumption is reduced as much as possible, and the idle memory space is used in time. Taking the processing capacity of target condensation input as 1056 measurement points as an example, the local LL2 space has more than 90% of the margin, and the shared SL2 space has more than 90% of the margin.

[0029] The algorithm has good portability. The bottom layer uses the component encapsulation method, so when used between models, it has good portability. Only the parameters need to be configured, and the algorithm effect can be easily realized, saving the cost of secondary development of manpower. BRIEF DESCRIPTION OF DRAWINGS

[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating laborious work.

[0031] Figure 1 The flow chart for target condensation processing of the present application is realized. DETAILED DESCRIPTION

[0032] The embodiments of the present application will be described in detail below with reference to the drawings.

[0033] The present application now will be described more fully hereinafter with reference to the accompanying drawings, in which some, but not all embodiments of the application are shown. Indeed, these various embodiments and what is described herein of the application can be implemented independently of one another, and of other embodiments, as will be readily appreciated by those of ordinary skill in the art. Likewise, the application can be implemented using any number of technologies and approaches. The various embodiments disclosed herein can each be implemented independently of one another and of other embodiments described herein. The scope of the present application is not intended to be limited to the embodiments described herein but is intended to include any suitable structure or application that incorporates employees of the present application and its principles. The application incorporates the scope of the claims and their equivalents.

[0034] To illustrate, various aspects of embodiments within the scope of the appended claims are described hereinafter. It should be apparent to those skilled in the art that the aspects described herein can be practiced without the specific details disclosed herein. Indeed, the aspects described herein can be practiced with any number of variations within the scope of the present application. Based on the teachings herein, one skilled in the art should appreciate that an aspect disclosed herein can be implemented independently of any other aspects and that various

[0035] It is also important to note that the present application can be carried out in ways not specifically set forth herein without departing from essential characteristics of the application. Thus, the present application should not be limited to the embodiments discussed herein, but should be given the broadest possible interpretation within its scope, as set forth in the appended claims.

[0036] Moreover, in the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be recognized by one of ordinary skill in the art that the various embodiments of the present application, including its methods, can be practiced without the specific details set forth in the following description.

[0037] The embodiment of the present application provides a target condensation implementation method of dynamic connection processing.

[0038] In the implementation of the domestic multi-core processor platform, the use of memory space is designed in combination with the memory characteristics of the processor. Software programming adopts the OpenMP fork-join model to realize parallel algorithm design. In the design and implementation, the single pulse timing is processed, the hardware resource is a domestic processor, and the hardware platform and software programming model are combined

[0039] A target condensation implementation method of dynamic connectivity processing includes the following steps:

[0040] In combination with the processing cycle and space usage of engineering implementation, a distance length coding algorithm is selected as a real-time implementation of two-dimensional matrix connectivity, wherein the distance length coding obtains the step of merging the threshold points in a single pulse according to the distance, and the merging of the single pulse is coded by the start distance and the end distance. In the inter-pulse merging process, each core is allocated several distance segments through the OpenMP parallel processing mode, and the multi-core parallel simultaneously performs the inter-pulse merging process, and then the segmented boundaries are separately processed. The parallel design time consumption is only 1 / 4 of the serial processing time consumption, which is particularly critical for real-time systems.

[0041] The establishment and deletion of the tree through the path compression weighted merging method are a real-time dynamic process, and therefore, in engineering implementation, how to dynamically manage the data buffer pool is a key to implementation. The purpose of dynamic management is to reduce memory loss as much as possible and use the idle memory space in time. The dynamic management is specifically performed through the creation of a root buffer pool and a leaf buffer pool.

[0042] In engineering implementation, the establishment and deletion of the tree through the path compression weighted merging method are a real-time dynamic process, and the dynamic management is performed through the creation of a root buffer pool and a leaf buffer pool. The leaf buffer pool uses the space of the new and old buffer pools to alternately perform dynamic management, the root buffer pool is managed through the number of leaf nodes, and the management of the root buffer pool includes an initialization process, an increase in the number of leaf nodes, a decrease in the number of leaf nodes, and cleaning. The root buffer pool is dynamically managed through the state, and in the pulse merging process, the idle position in the root buffer pool is dynamically searched according to the state mark to achieve dynamic management. The dynamic management of the root buffer pool runs through the entire inter-pulse merging process, and the core is to manage through the number of leaf nodes, including the initialization process, the increase in the number of leaf nodes, the decrease in the number of leaf nodes, and the cleaning. The dynamic management is performed through the state, and in the pulse merging process, the idle position in the root buffer pool is dynamically searched according to the state mark to achieve dynamic management.

[0043] Weighted fast merging, each time the root node of a certain node is calculated, the nodes along the path are also pointed to the root node, the tree is maximized to be flattened, and the number of nodes to be traversed is greatly reduced when checking the connectivity state. Through the recursive method, the index number of the root node is pointed to the root node while finding the root node, but it is difficult to implement in the engineering implementation process. The weighted fast merging uses the halving path compression algorithm to continuously move the nodes upwards until all the nodes are connected to the root node. The engineering implementation of this way is simpler.

[0044] The rationality optimization design of memory space allocation, the storage space of multi-core processor includes the LL2 memory space unique to each core and the shared SL2 memory space, through the decomposition of algorithm steps, the processing data of root buffer pool and leaf buffer pool are placed in the independent LL2 memory space of each core in parallel process, and finally the output data of target condensation are uniformly synchronized to the shared space, through this kind of design and the same data volume uniformly processed in the shared space, the efficiency is improved by 30%.

[0045] In one embodiment, a target condensation implementation method of dynamic connection processing.

[0046] The input of target condensation is set to 1056 measurements, and the processing space is allocated to local LL2 space and shared SL2 space. Since multiple threshold detection points are more likely to appear on the sea surface target, which becomes an extended target, and there are also sea clutter and false alarms, target condensation processing is required after detection. The specific processing steps are as follows:

[0047] As shown in Figure 1 Step 1, data initialization, through OpenMP multi-core parallel processing, the measurement information of the threshold detection point is inputted and arranged into a new leaf buffer pool.

[0048] Step 2, inter-pulse correlation processing, realized by OpenMP multi-core parallel mode, using binary search method to associate the old and new leaf buffers, if the association condition is met, it is judged whether it has been associated before, if it has been associated, the root node association pair needs to be further obtained, finally the pulse number of the old node is updated, the position index number of the old node is assigned to the position index number of the new node, and the distance association pair and the root node association pair are outputted. The processing capacity of the target condensation input is set to 1056 measurement points, and the inter-pulse correlation is searched according to the conventional traversal cycle mode, which takes milliseconds. After multi-core parallel, it still cannot meet the demand of fast time sequence real-time system, therefore, the classical binary search method is used to realize pulse correlation processing, which improves the correlation efficiency, greatly reduces the processing time, and meets the demand of fast time sequence and large sample real-time system.

[0049] Step 3, new leaf node association to root processing, using OpenMP multi-core parallel processing to traverse all new leaf nodes, if the new leaf node is associated, the current leaf node is connected to the root, and the leaf node number of the root is increased by 1, if the new leaf node is not associated, the position with the smallest number and state vector of 0 is selected in the root buffer pool, the root node of the current position is initialized, and the new node is associated to the root buffer pool.

[0050] Step 4, old leaf node arrangement processing, using OpenMP multi-core parallel processing to realize the old leaf node arrangement function, including deletion, retention and update three kinds, traversing all the old leaf nodes, connecting the old leaf nodes to the root. Calculate the pulse number difference, that is, the current pulse number minus the pulse number of the old leaf node, and process according to the pulse number difference. According to the pulse number difference, it is divided into three categories:

[0051] If the pulse number difference is equal to 0, it means that it has been updated, and the number of leaf nodes corresponding to the root is reduced by 1, because the number of updated leaf nodes is no longer a leaf node in the next pulse.

[0052] If the pulse number difference is greater than the deletion threshold, the leaf node deletion condition is met, the root node number is reduced by 1, and if the current root leaf node number is 0, it means that all leaf nodes of the current root have been deleted, and need to be output, and the root node state is set to output 2.

[0053] If the pulse node difference is greater than 0 and less than the deletion threshold, the current old leaf node has no new leaf node associated with it, and does not meet the deletion condition, the old node is added to the free place in the new buffer pool, and in the next pulse processing process, the new buffer becomes old buffer, and can still be retained in the buffer pool.

[0054] Step 5, update operation feature processing, through OpenMP multi-core parallel processing mode to realize traversal of all new nodes and root associated pairs, assign valid nodes to corresponding root node positions, record angular resolution forced splitting record maximum power vector, and land segment processing operation.

[0055] Step 6, find segment boundary processing. Through OpenMP multi-core parallel processing to find the boundary between segments, update all valid states to the buffer pool for processing; find all distance gate number start and end that meet the conditions; find the minimum value of the start distance gate and the maximum value of the end distance gate, and store the number and distance gate number start and end in the array sequence.

[0056] Step 7, segment boundary merging processing, single-core serial loop traverses all segments to get the current segment end node and the next segment start node. If the two nodes meet the association condition and one of them is in the output state, update the output node in the other node, and clean up the output node;

[0057] Step 8, target output and cleaning process, using OpenMP multi-core parallel processing to traverse all root buffers, if the index number in the root buffer pool is not equal to its own position, cleaning is performed, if the state in the root buffer pool is output, indicating that the output operation needs to be performed, the target number of the output buffer pool is added by 1, the centroid of the range and azimuth is calculated, and the false alarm and target of the output target are judged by azimuth expansion, and finally the effective target information attribute is output.

[0058] The target condensation method for completing the dynamic connection processing through the above steps is realized, and the output target condensation result is sent to a subsequent module for track processing and target tracking processing.

[0059] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for target cohesion implementation of dynamic connectivity processing, characterized in that, It comprises the following steps: In combination with the processing cycle and space usage achieved by engineering, a distance length coding algorithm is selected for real-time implementation of two-dimensional matrix connection, wherein the distance length coding is obtained by first merging the threshold crossing points in a single pulse according to distance, and then encoding the merging of the single pulse through the start distance and the end distance; Weighted fast merging, each time a root node of a certain node is calculated, the nodes checked along the path are also pointed to the root node, the tree is maximized to be flattened, and through a recursive method, the index number of the root node is pointed to the root node while the root node is found; In the inter-pulse merging process, the OpenMP parallel processing method is used to allocate several distance segments to each core, and the multi-core parallel processing is used to simultaneously perform inter-pulse merging processing, and then the segmented boundaries are separately processed; In the engineering implementation, the path compression weighted merging method is used to establish and delete the tree in a real-time dynamic process, and the root buffer pool and the leaf buffer pool are created for dynamic management; the leaf buffer pool uses the space of the new and old buffer pools to alternately perform dynamic management, the root buffer pool is managed through the number of leaf nodes, the management of the root buffer pool includes the initialization process, the increase of the number of leaf nodes, the decrease of the number of leaf nodes and the cleaning, the root buffer pool is dynamically managed through the state, and in the pulse merging process, the idle position in the root buffer pool is dynamically searched according to the state mark to realize dynamic management; The storage space of the multi-core processor includes the LL2 memory space unique to each core and the shared SL2 memory space, through the decomposition of the algorithm steps, the processing data of the root buffer pool and the leaf buffer pool are placed in the LL2 memory space independent of each core in the parallel process, and finally the output data of the target condensation are uniformly synchronized to the shared space; Step 1, data initialization, the measurement information of the threshold crossing detection points is input and arranged into a new leaf buffer pool through OpenMP multi-core parallel processing; Step 2, inter-pulse association processing, realized through OpenMP multi-core parallel processing, the binary search method is used to associate and judge the new and old leaf buffers, if the association condition is met, it is judged whether it has been associated before, if it has been associated, the root node association pair needs to be further obtained, finally the pulse number of the old node is updated, the position index number of the old node is assigned to the position index number of the new node, and the distance association pair and the root node association pair are output; Step 3, new leaf node association to root processing, all new leaf nodes are traversed using OpenMP multi-core parallel processing, if the new leaf node has been associated, the current leaf node is connected to the root, and the number of leaf nodes of the root is increased by 1, if the new leaf node has not been associated, the position with the smallest number and the state vector of 0 in the root buffer pool is selected, the root node at the current position is initialized, and the new node is associated to the root buffer pool; Step 4, old leaf node arrangement processing, the old leaf node arrangement function is realized using OpenMP multi-core parallel processing, including deletion, retention and update, all old leaf nodes are connected to the root, the pulse number difference is calculated, and different processing is performed according to the pulse number difference; Step 5, the update operation seeks feature processing, through OpenMP multi-core parallel processing mode to realize traversal of all new nodes and the association between the root, the effective node is assigned to the corresponding root node position, record the maximum power vector based on the angle resolution forced split, and the operation in the land section processing; Step 6, find the segment boundary processing, through OpenMP multi-core parallel processing to find the boundary between the segments, update all the state to the buffer pool for processing; find out all the starting and ending of the distance gate number that meet the conditions; find the minimum value of the starting distance gate and the maximum value of the ending distance gate, store the start and end of the distance gate number in the array sequence; Step 7, segment boundary merging processing, single-core serial loop traverses all segments, gets the node at the end of the current segment and the node at the beginning of the next segment, if the two nodes meet the association condition and one of them is an output state, update the output node to the other node, and clean up the output node; Step 8, target output and cleaning processing, using OpenMP multi-core parallel processing to traverse all root buffers, if the index number in the root buffer pool is not equal to its own position, clean up, if the state in the root buffer pool is output, it means that the output operation is needed, add 1 to the target number of the output buffer pool, judge the false alarm and target of the output target by calculating the centroid of distance and azimuth and by azimuth expansion, and finally output the effective target information attribute.

2. The method of claim 1, wherein the target coalescence is achieved by dynamic connectivity processing. The weighted fast merging uses the halving path compression algorithm, which moves the nodes up until all nodes are connected to the root node.

3. The method of claim 1, wherein the target coalescence is achieved by dynamic connectivity processing. In step 4, if the pulse number difference is equal to 0, the leaf node number of the corresponding root is reduced by 1; If the pulse number difference is greater than the deletion threshold, the leaf node deletion condition is met, the root node number is reduced by 1, and if the leaf node number of the current root is 0, it means that all leaf nodes of the current root have been deleted, and need to be output, the root node state is set to output 2; If the pulse node difference is greater than 0 and less than the deletion threshold, the current old leaf node has no new leaf node association, and also does not meet the deletion condition, the old node is added to the free place in the new buffer pool, and in the next pulse processing process, the new buffer becomes the old buffer, which can still be retained in the buffer pool.

Images