Spatial index establishment method and system based on adaptive spiking neural network
By using an adaptive spiking neural network to group and index geospatial data, the problem of traditional methods destroying spatial relationships is solved, and efficient data management and querying are achieved.
Patent Information
- Application Number
- CN202511475549.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-15
- Publication Date
- 2025-12-30
AI Technical Summary
Traditional GIST algorithms may disrupt the spatial relationships between adjacent and intersecting features when building geospatial indexes, resulting in index accuracy that fails to meet the requirements of specific scenarios.
An adaptive spiking neural network is used to group and index grid features, grouping intersecting and adjacent features into the same index area. The trained spiking neural network is used for group management to create an index layer.
It effectively preserves the spatial relationships of various elements within the source layer, improves data processing efficiency, and supports hardware parallel computing.
Smart Images
Figure CN121233601A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of spatial indexing technology, and more specifically to a spatial indexing method and system based on an adaptive spiking neural network. Background Technology
[0002] A geospatial index is an index structure built for geospatial data to improve the efficiency of spatial queries by limiting the search to the corresponding index range, rather than performing a linear traversal of the entire dataset. Therefore, when dealing with large volumes of spatial data, geospatial indexes can significantly speed up queries. In certain scenarios, it is necessary to create geospatial indexes without disrupting the spatial relationships between adjacent or intersecting elements.
[0003] Currently, the traditional GIST algorithm indexes spatial data based on a simple bounding rectangular region. This may assign originally adjacent features to different index regions, thereby disrupting spatial relationships and causing its accuracy to fail to meet the needs of specific scenarios.
[0004] Therefore, in view of the shortcomings of the existing technology, how to provide a spatial indexing method and system based on adaptive spiking neural networks, which groups elements into indexes and groups intersecting, adjacent, or other elements with certain spatial relationships into the same index area to facilitate subsequent data retrieval and management, is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0005] In view of this, the present invention provides a spatial indexing method and system based on an adaptive spiking neural network, which groups grid features into indexes and groups intersecting, adjacent, or other spatially related features into the same index area to facilitate subsequent data retrieval and management.
[0006] To achieve the above objectives, the present invention adopts the following technical solution: a spatial indexing method based on an adaptive spiking neural network, comprising: Create a grid layer based on the source data; A spiking neural network is established using the aforementioned grid layer, and the spiking neural network is trained to obtain the trained spiking neural network. Based on the trained spiking neural network, the source layers are grouped and indexed, and an index layer is created. The process of establishing a spiking neural network using the grid layer includes: Traverse the grid elements of the grid layer and divide the grid into intersecting and non-intersecting areas; The center point of each grid element is taken as a neuron. The self-charging current of the neuron corresponding to the intersecting grid is set to m, and the self-charging current of the neuron corresponding to the non-intersecting grid is set to n. For each neuron, the neurons connecting its four neighboring neurons (up, down, left, and right) serve as its synapses, and these synapses connect adjacent neurons.
[0007] Preferably, if the grid features in the grid layer intersect with the features in the source layer, then the grid is an intersecting area grid; otherwise, it is a non-intersecting area grid.
[0008] Preferably, the neuron possesses two properties: membrane potential and self-charging current, with its membrane potential initialized to 0. Preferably, training the spiking neural network includes: After the spiking neural network is built and initialized, training begins. At each time step, the membrane potential of each neuron is recorded. Based on the pulse signals of the synapses to which they are connected and self-charging current Update when the membrane potential is greater than the potential threshold. When the neuron emits a pulse signal to its corresponding synapse, its membrane potential changes. Return to 0.
[0009] Preferably, the formula for calculating membrane potential update is as follows: .
[0010] Preferably, at each moment, when the connecting synapse receives a pulse signal, its trace value is updated. and weight The calculation formula is as follows: ; ; ; ; in, This represents the weight of the synapse connecting the i-th neuron and the j-th neuron. The change in weight. and It is a non-negative constant. The time when the neuron last fired a pulse. is the time constant.
[0011] Preferably, after the spiking neural network is trained, the neurons connected to each synapse are selected according to their weights, and then the corresponding grids are retrieved based on these neurons. These grids are then merged one by one to form the final result index layer. Group and manage source layer features based on the index layer.
[0012] Preferably, a spatial indexing system based on an adaptive spiking neural network includes: The grid layer creation module is used to create grid layers based on source data; A spiking neural network training module is used to establish a spiking neural network using the grid layer, train the spiking neural network, and obtain a trained spiking neural network. The index layer module is used to group and index the source layers based on the trained spiking neural network, and to create index layers.
[0013] As can be seen from the above technical solution, compared with the prior art, this invention discloses a spatial indexing method and system based on an adaptive spiking neural network. The spiking neural network-based method for indexing geographic information vector data aims to improve data processing efficiency. This invention first establishes a grid layer based on the source data, then uses the grid layer to build a spiking neural network. Through training the spiking neural network, the source layer is grouped and indexed, and an index layer is established. The index established by this method retains the spatial relationships of various elements within the source layer and is easily parallelized using hardware. Attached Figure Description
[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0015] Figure 1 This invention provides a schematic flowchart of a spatial indexing method based on an adaptive spiking neural network.
[0016] Figure 2 This is a schematic diagram illustrating the establishment of a spiking neural network as provided in an embodiment of the present invention.
[0017] Figure 3 This is a schematic diagram of an index layer provided in an embodiment of the present invention. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] This invention discloses a spatial indexing method based on an adaptive spiking neural network, such as... Figure 1 As shown, it includes: Create a grid layer based on the source data; A spiking neural network is established using the aforementioned grid layer, and the spiking neural network is trained to obtain the trained spiking neural network. Based on the trained spiking neural network, the source layers are grouped and indexed, and an index layer is created. The process of establishing a spiking neural network using the grid layer includes: Traverse the grid elements of the grid layer and divide the grid into intersecting and non-intersecting areas; The center point of each grid element is taken as a neuron. The self-charging current of the neuron corresponding to the intersecting grid is set to m, and the self-charging current of the neuron corresponding to the non-intersecting grid is set to n. For each neuron, the neurons connecting its four neighboring neurons (up, down, left, and right) serve as its synapses, and these synapses connect adjacent neurons.
[0020] Where m and n are both constants; m is a positive value and n is a negative value.
[0021] Specifically, if a grid feature in a grid layer intersects with a feature in the source layer, then the grid is an intersecting area grid; otherwise, it is a non-intersecting area grid.
[0022] Specifically, neurons have two properties: membrane potential and self-charging current, with the initial value of the membrane potential being 0.
[0023] In one specific embodiment of the present invention, the process of establishing and initializing the spiking neural network includes: First, such as Figure 2 As shown, the system divides the area into intersecting and non-intersecting regions. A grid layer is created, and its features are traversed. If a grid feature intersects with a feature in the source layer, that grid is considered an intersecting region; otherwise, it is considered a non-intersecting region. Figure 2 (a) in the image represents the source layer, such as... Figure 2 As shown in (b), the purple area represents the intersecting area layer, and the green area represents the non-intersecting area layer.
[0024] Then, the center point of each grid element is extracted as a neuron, and the neuron has a membrane potential. and self-charging current Two properties, its membrane potential The initial value is 0, and the self-charging current of the neurons corresponding to the intersecting grids is... Set to m, m=1, the self-charging current of neurons corresponding to non-intersecting grids. Let n be the number of elements, and n = -5. Figure 2 As shown in (c), the red neurons are neurons in the intersecting region, and the rest are neurons in the non-intersecting region.
[0025] Finally, for each neuron, the neurons connecting its four neighboring neurons (up, down, left, and right) serve as its synapses. These synapses are used to connect adjacent neurons, such as... Figure 2 As shown in (c) above. Synapses have pulse signals. ,trace and weight Three attributes, including pulse signal Heji The value is initialized to 0, and the weight is... Its initial value is 1.
[0026] Specifically, training a spiking neural network includes: After the spiking neural network is built and initialized, training begins. At each time step, the membrane potential of each neuron is recorded. Based on the pulse signals of the synapses to which they are connected and self-charging current Update when the membrane potential is greater than the potential threshold. When the neuron emits a pulse signal to its corresponding synapse, its membrane potential changes. Return to 0.
[0027] Specifically, the formula for calculating membrane potential update is as follows: .
[0028] Specifically, at each moment, when the connecting synapse receives a pulse signal, its trace value is updated. and weight The calculation formula is as follows: ; ; ; ; in, This represents the weight of the synapse connecting the i-th neuron and the j-th neuron. The change in weight. and It is a non-negative constant. The time when the neuron last fired a pulse. is the time constant.
[0029] In one specific embodiment of the present invention, the spiking neural network begins training after being established and initialized. At each time step, the membrane potential of each neuron... Based on the pulse signals of the synapses to which they are connected and self-charging current Update when the membrane potential is greater than the potential threshold. When the neuron emits a pulse signal to its corresponding synapse, its membrane potential changes. Reset to 0. When the neural network training ends, iterate through the synaptic weights, and merge and group the grids corresponding to the neurons at both ends of the synapse with higher weight values, ultimately forming an index range layer, such as... Figure 3 As shown, (a) is the source layer and (b) is the index layer.
[0030] Specifically, the training process of a spiking neural network includes: (1) Initialize the spiking neural network and the membrane potential of all neurons. and synaptic weight The initial value is 0, and the self-charging current of the neurons corresponding to the intersecting grids is... Setting it to 1 indicates the self-charging current of neurons corresponding to non-intersecting grids. The synapse has a pulse signal at a current of -5A. ,trace and weight Three attributes, including pulse signal Heji The value is initialized to 0, and the weight is... Its initial value is 1.
[0031] (2) At each moment, when the synapse receives a pulse signal And self-charging current When positive, update the neuron membrane potential. The value is when the membrane potential is greater than the potential threshold. At that time, the neuron sends a pulse signal to its adjacent synapses. This affects the membrane potential of the next neuron. The formula for calculating membrane potential updates is shown in formula (1). Otherwise, the neuron does not generate pulse signals and does not affect the updates of the next neuron.
[0032] (1)
[0033] (3) At each moment, when the connecting synapse receives a pulse signal, its trace value is updated. and weight The calculation formula is as follows: (2) (3) (4) (5) in, This represents the weight of the synapse connecting the i-th neuron and the j-th neuron. The change in weight. and It is a non-negative constant. The time when the neuron last fired a pulse. is the time constant.
[0034] In this embodiment of the invention, the adjustment range of the synapse depends not only on the current state of the spiking neural network, but also on the update state of the synapse in the past. If the synapse weight has been continuously updated in the same direction in the past, then its update range will continue to increase. However, if the update of the synapse weight in a certain direction has been constantly changing in the past, it indicates that the synapse is not stable, which will weaken the update range of the synapse weight.
[0035] (4) As can be seen from formula (2), when a neuron emits a pulse signal, the weight of the synapse... The weight will increase, and if this increasing trend continues, it will be further strengthened. When a neuron does not emit pulse signals, its weight... According to each moment The value is attenuated. If this attenuation trend continues, it will be further strengthened. To ensure the transmission of the pulse signal, the weight is limited. It will not be lower than -1. If a synapse swings between two different trends of increase and decrease, it indicates that the synapse is in an unstable state, and the magnitude of the weight update for that synapse will be reduced.
[0036] Specifically, after the spiking neural network is trained, the neurons connected to each synapse are selected according to their weights. Then, based on these neurons, their corresponding grids are retrieved, and these grids are merged one by one to form the final result index layer. Group and manage source layer features based on the index layer.
[0037] In this embodiment of the invention, when the spiking neural network has been trained, the weights of each synapse are used to... The process involves filtering out neurons connected to synapses (selecting neurons connected to synapses with weights greater than a certain threshold). Based on these neurons, their corresponding grids are retrieved, and these grids are merged one by one to form the final index layer. Users can then group and manage source layer features using this index layer.
[0038] In this embodiment of the invention, (1) the map is divided into intersecting feature regions and non-intersecting feature regions, and the map is modeled as a spiking neural network. Each feature corresponds to a neuron in the spiking neural network, and each neuron is connected by a bidirectional synapse. (2) The parameters of the corresponding neuron and synapse are initialized. (3) At each time, the membrane potential of each neuron is updated according to the pulse situation of the incoming synapse and the self-charging current. When the membrane potential is greater than the potential threshold, a pulse signal is emitted. The pulse signal is transmitted to the adjacent neuron through the synapse. After the neuron emits the pulse signal, the membrane potential returns to zero. (4) At each time, the trace value of each synapse is updated according to whether a pulse signal is transmitted, and the weight of the synapse is updated according to the trace value. (5) After the pulse signal transmission ends, the range of the index area is planned according to the weight value of the synapse, and finally the spatial index is obtained.
[0039] In one specific embodiment of the present invention, a spatial indexing system based on an adaptive spiking neural network includes: The grid layer creation module is used to create grid layers based on source data; A spiking neural network training module is used to establish a spiking neural network using the grid layer, train the spiking neural network, and obtain a trained spiking neural network. The index layer module is used to group and index the source layers based on the trained spiking neural network, and to create index layers.
[0040] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.
[0041] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for spatial index establishment based on adaptive pulse neural network, characterized in that, The method comprises the following steps: establishing a grid layer based on source data; training a spiking neural network by using the grid layer, to obtain a trained spiking neural network; grouping and indexing the source layer based on the trained spiking neural network, and establishing an index layer; wherein the step of training the spiking neural network by using the grid layer comprises the following steps: traversing grid elements of the grid layer, and dividing intersecting area grids and non-intersecting area grids; taking a center point of each grid element as a neuron, and setting a self-charging current of a neuron corresponding to an intersecting area grid as m and a self-charging current of a neuron corresponding to a non-intersecting area grid as n; for each neuron, connecting neurons of four adjacent areas thereof as synapses of the neuron, and connecting two adjacent neurons by using the synapses.
2. The method of claim 1, wherein the method is based on an adaptive pulse neural network. If a grid element in the grid layer intersects with a source layer element, the grid is an intersecting area grid; otherwise, the grid is a non-intersecting area grid.
3. The method of claim 1, wherein the method further comprises: The neuron has two attributes of membrane potential and self-charging current, and an initialized value of the membrane potential is 0.
4. The method of claim 3, wherein the method further comprises: The step of training the spiking neural network comprises the following steps: After the spiking neural network is built and initialized, training begins. At each time step, the membrane potential of each neuron is recorded. Based on the pulse signals of the synapses to which they are connected and self-charging current Update when the membrane potential is greater than the potential threshold. When the neuron emits a pulse signal to its corresponding synapse, its membrane potential changes. Return to 0.
5. The method of claim 4, wherein the method further comprises: a membrane potential updating calculation formula is as follows: 。 6. The method of claim 4, wherein the method further comprises: At each time step, when a connected synapse receives a spike, its trace is updated and weights are computed according to the following formula: ； ； ； ； wherein, represents the weight of the synapse connecting the i-th neuron with the j-th neuron, is the amount of change in the weight, and is a non-negative constant, is the time of the last spike emitted by the neuron, is a time constant.
7. The method of claim 1, wherein the method further comprises: after the training of the spiking neural network is completed, according to weights of synapses, neurons connected with the synapses are screened out, and according to the neurons, corresponding grids are searched, the grids are merged one by one, and a final result index layer is formed; source layer elements are grouped and managed according to the index layer.
8. A system for spatial index establishment based on an adaptive spiking neural network, characterized in that, The method comprises the following steps: a grid layer establishing module is configured to establish a grid layer based on source data; a spiking neural network training module is configured to train a spiking neural network by using the grid layer, to obtain a trained spiking neural network; an index layer module is configured to group and index the source layer based on the trained spiking neural network, and establish an index layer.