Shielding target detection method, neural network model and calculation program product

Abstract

The invention relates to an occlusion target detection method, a neural network model and a calculation program product. The method comprises the following steps: acquiring a feature matrix corresponding to a to-be-recognized image; on the basis of the feature matrix and the frontal lobe output matrix, updating the activation state of temporal lobe neurons to obtain a temporal lobe output matrix; on the basis of the temporal lobe output matrix, updating an activation state of a frontal lobe neuron to obtain a new frontal lobe output matrix; first iteration operation is executed until a first ending condition is met, the first iteration operation is stopped, a first steady-state temporal lobe neural network model is obtained, and the first iteration operation comprises updating the activation state of the temporal lobe neurons and updating the activation state of the frontal lobe neurons; and processing the to-be-recognized image based on the first steady-state temporal lobe neural network model to obtain a recognition result. According to the method provided by the embodiment of the invention, the recognition precision can be adaptively maintained when different types of shielding are faced, and the recognition accuracy and stability are enhanced.

Description

Occluded target detection methods, neural network models, and computational programs Technical Field

[0001] This disclosure relates to the field of computer vision technology, and in particular to an occluded target detection method, neural network model, and computing program product. Background Technology

[0002] In the field of artificial intelligence, occluded object detection is a key challenge in intelligent vision applications. Existing technologies mainly rely on deep learning models, most of which are designed based on a "big data-driven" paradigm. Mainstream AI models experience performance degradation in occluded scenarios, while humans can achieve stable recognition through small-sample learning, highlighting the superiority of biological intelligence.

[0003] From a technological development perspective: First, early convolutional neural networks and recurrent neural networks had low accuracy in occlusion detection. Later, more complex deep learning models outperformed the previous stage in occlusion detection, but performed poorly in generalization and accuracy in identifying different types of occlusions. Overall, the precision, accuracy, and stability of target occlusion detection are not high. Summary of the Invention

[0004] In view of this, this disclosure proposes an occlusion target detection scheme.

[0005] According to one aspect of this disclosure, an occluded target detection method is provided, comprising: acquiring a feature matrix corresponding to an image to be identified, wherein the elements in the feature matrix correspond to the spatial positions of temporal lobe neurons in a temporal lobe neural network model, and the elements represent the mapping vectors of image features of the image to be identified on the temporal lobe neurons; updating the activation state of the temporal lobe neurons based on the feature matrix and a frontal lobe output matrix to obtain a temporal lobe output matrix, wherein the temporal lobe output matrix contains the updated activation state information of the temporal lobe neurons, and the frontal lobe output matrix contains the updated activation state information of the frontal lobe neurons in the frontal lobe neural network model; updating the activation state of the frontal lobe neurons based on the temporal lobe output matrix to obtain a new frontal lobe output matrix; performing a first iteration operation until a first termination condition is met, stopping the first iteration operation to obtain a first steady-state temporal lobe neural network model, wherein the first iteration operation includes: updating the activation state of the frontal lobe neurons and updating the activation state of the temporal lobe neurons; and processing the image to be identified based on the first steady-state temporal lobe neural network model to obtain a recognition result.

[0006] In one possible implementation, before updating the activation state of the temporal lobe neurons based on the feature matrix and the frontal lobe output matrix to obtain the temporal lobe output matrix, the method further includes: initializing the activation state of each temporal lobe neuron in the initial temporal lobe neural network model based on the feature matrix to obtain an initial activation state matrix; and controlling the initial temporal lobe neural network model to perform multiple activation state self-updates based on the feature matrix and the initial activation state matrix until a second termination condition is met to obtain the temporal lobe neural network model.

[0007] In one possible implementation, the step of processing the image to be identified based on the first steady-state temporal lobe neural network model to obtain the identification result includes: controlling the first steady-state temporal lobe neural network model to perform multiple activation state self-updates based on the activation state of each temporal lobe neuron in the first steady-state temporal lobe neural network model until a third termination condition is met to obtain a second steady-state neural network model; and processing the image to be identified based on the second steady-state neural network model to obtain the identification result.

[0008] In one possible implementation, performing the first iteration operation includes: determining the target frontal lobe neuron corresponding to a first time step, and determining the target temporal lobe neuron corresponding to the first time step; updating the activation state of the target temporal lobe neuron based on the feature matrix and the frontal lobe output matrix corresponding to a second time step, obtaining the activation state value of the target temporal lobe neuron corresponding to the second time step, and determining the temporal lobe output matrix corresponding to the second time step based on the activation state value of the target temporal lobe neuron corresponding to the second time step, wherein the second time step is the next time step after the first time step; updating the activation state of the target frontal lobe neuron based on the temporal lobe output matrix corresponding to the second time step, obtaining the activation state value of the target frontal lobe neuron corresponding to a third time step, and based on the feature matrix and the frontal lobe output matrix corresponding to the second time step, determining the temporal lobe output matrix corresponding to the second time step, wherein the second time step is the next time step after the first time step; and updating the activation state of the target frontal lobe neuron based on the temporal lobe output matrix corresponding to the second time step, obtaining the activation state value of the target frontal lobe neuron corresponding to a third time step, and based on the feature matrix and the frontal lobe output matrix corresponding to the second time step, determining the temporal lobe output matrix corresponding to the third time step, and determining the temporal lobe output matrix corresponding to the third time step. Based on the activation state value of the target frontal lobe neuron at the third time point, determine the frontal lobe output matrix corresponding to the third time point, where the third time point is the next time point after the second time point; take the second time point as the new first time point and the third time point as the new second time point, and perform a second iteration operation until the first termination condition is met, then stop the second iteration operation. The second iteration operation includes: determining the target frontal lobe neuron corresponding to the first time point, determining the target temporal lobe neuron corresponding to the first time point, updating the activation state of the target frontal lobe neuron, determining the temporal lobe output matrix corresponding to the second time point, updating the activation state of the target temporal lobe neuron, and determining the frontal lobe output matrix corresponding to the third time point.

[0009] In one possible implementation, the frontal lobe activation state matrix represents the activation state of each frontal lobe neuron in the frontal lobe neural network model, and the temporal lobe activation state matrix represents the activation state of each temporal lobe neuron in the temporal lobe neural network model; the step of updating the activation state of the target temporal lobe neuron based on the feature matrix and the frontal output matrix corresponding to the second time step to obtain the activation state value of the target temporal lobe neuron at the second time step, and determining the temporal lobe output matrix corresponding to the second time step based on the activation state value of the target temporal lobe neuron at the second time step, includes: determining the activation state value of the target temporal lobe neuron at the second time step based on the frontal output matrix corresponding to the second time step, the temporal lobe activation state matrix corresponding to the first time step, the feature matrix, and the temporal intratemporal connection weight matrix of the target temporal lobe neuron and connected temporal lobe neurons; updating the temporal lobe activation state matrix corresponding to the first time step based on the activation state value of the target temporal lobe neuron at the second time step to obtain the temporal lobe activation state matrix corresponding to the second time step; and determining the temporal lobe output matrix corresponding to the second time step based on the activation state value of the target temporal lobe neuron at the second time step. The process involves: determining the temporal lobe output matrix at the second time step using the corresponding temporal lobe activation state matrix and the temporofrontal connection weight matrix; updating the activation state of the target frontal lobe neuron based on the temporal lobe output matrix at the second time step to obtain the activation state value of the target frontal lobe neuron at the third time step; and determining the frontal lobe output matrix at the third time step based on the activation state value of the target frontal lobe neuron at the third time step, including: determining the activation state value of the target frontal lobe neuron at the third time step based on the temporal lobe output matrix at the second time step, the frontal lobe activation state matrix at the second time step, and the intrafrontal connection weight matrix between the target temporal lobe neuron and the connected frontal lobe neurons; updating the frontal lobe activation state matrix at the second time step based on the activation state value of the target frontal lobe neuron at the third time step to obtain the frontal lobe activation state matrix at the third time step; and determining the frontal lobe output matrix at the third time step based on the temporal lobe activation state matrix at the third time step and the frontotemporal connection weight matrix.

[0010] In one possible implementation, obtaining the feature matrix corresponding to the image to be identified includes: extracting features from the image to be identified to obtain a feature vector, the feature vector including multiple feature values; mapping the multiple feature values ​​to positions corresponding to the temporal lobe neurons in a two-dimensional matrix based on a first weight vector corresponding to the temporal lobe neurons, determining the elements in the two-dimensional matrix, and obtaining the feature matrix, wherein the first weight vector contains a first weight corresponding to the feature values.

[0011] In one possible implementation, the method for training the temporofrontal connection weight matrix includes: obtaining a sample feature vector corresponding to a sample image; determining a winning frontal lobe neuron based on the sample feature vector and the temporofrontal connection weights corresponding to each frontal lobe neuron; updating the temporofrontal connection weights corresponding to other frontal lobe neurons based on the position of the winning frontal lobe neuron, the positions of other frontal lobe neurons, the temporofrontal connection weights corresponding to other frontal lobe neurons, and the frontal lobe activation state matrix; performing a third iteration operation until a fourth termination condition is met, and stopping the third iteration operation, wherein the third iteration operation includes: obtaining the sample feature vector corresponding to the sample image, determining the winning frontal lobe neuron, and updating the temporofrontal connection weights corresponding to other frontal lobe neurons.

[0012] In one possible implementation, the method for training the intrafrontal connection weight matrix includes: determining activation regions in the frontal neural network model based on sample images, the activation regions including: a living activation region composed of frontal lobe neurons that recognize living images, and a non-living activation region composed of frontal lobe neurons that recognize non-living images; for each temporal lobe neuron, determining a first activation value for each frontal lobe neuron relative to living and non-living patterns according to the activation region where the frontal lobe neuron is located; and determining the intrafrontal connection weights corresponding to each frontal lobe neuron based on the first activation values ​​corresponding to each frontal lobe neuron, thereby obtaining the intrafrontal connection weight matrix.

[0013] According to another aspect of this disclosure, an occluded target detection neural network model is provided, comprising: the occluded target detection neural network model including: a deep convolutional neural network model, a frontal lobe neural network model, and a temporal lobe neural network model; the deep convolutional neural network model and the temporal lobe neural network model are data-connected, and the temporal lobe neural network model and the frontal lobe neural network model are data-connected; the deep convolutional neural network model is used to extract features from the image to be recognized to obtain a feature vector, the feature vector including multiple feature values; the temporal lobe neural network model is used to map the multiple feature values ​​to positions corresponding to the temporal lobe neurons in a two-dimensional matrix based on a first weight vector corresponding to the temporal lobe neurons, to determine the elements in the two-dimensional matrix, to obtain a feature matrix; the frontal lobe neural network model is used to update the activation state of the temporal lobe neurons by interacting with the temporal lobe neural network model regarding the activation state of their respective neurons, so that the temporal lobe neural network model reaches a steady state; when the temporal lobe neural network model is in a steady state, it is further used to determine a recognition result based on the image to be recognized.

[0014] According to another aspect of this disclosure, a computer program product is provided, including a computer program or a non-volatile computer-readable storage medium carrying the computer program, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0015] In this embodiment, image features in the image to be identified can be mapped to a feature matrix, and then the mapping vector on the temporal lobe neurons can be determined through the feature matrix. This allows image features representing different objects in the image to be mapped to temporal lobe neurons in corresponding regions, improving the recognition accuracy of the temporal lobe neural network model. The temporal lobe neural network model and the frontal lobe neural network model undergo multiple updates and interactions of neuron activation states, ensuring that the temporal lobe neurons most suitable for the image to be identified are activated and reach a stable state. Using the first stable-state temporal lobe neural network to process the image to be identified improves the accuracy of the recognition result. Furthermore, the method in this embodiment possesses adaptive adjustment capabilities, enabling dynamic adjustment between the frontal lobe neural network model and the temporal lobe neural network model to maintain recognition accuracy when faced with different types of occlusion, thus enhancing the accuracy and stability of the recognition.

[0016] Other features and aspects of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0017] The accompanying drawings, which are included in and form part of this specification, illustrate exemplary embodiments, features, and aspects of this disclosure together with the specification and serve to explain the principles of this disclosure.

[0018] Figure 1 is a schematic flowchart of the occlusion target detection method provided in the embodiments of this disclosure. Detailed Implementation

[0019] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0020] As used herein, the terms “comprising,” “including,” “having,” or variations thereof are open-ended and include one or more of the stated features, integrals, elements, steps, components, or functions, but do not exclude the presence or addition of one or more other features, integrals, elements, steps, components, functions, or groups thereof.

[0021] When an element is referred to as “connected,” “coupled,” “responding,” or a variation thereof relative to another element, it may be directly connected, coupled, or responding to another element, or there may be an intermediate element present.

[0022] Although the terms first, second, third, etc., may be used herein to describe various elements / operations, these elements / operations should not be limited by these terms. These terms are only used to distinguish one element / operation from another. Therefore, without departing from the teachings of the inventive concept, a first element / operation in some embodiments may be referred to as a second element / operation in other embodiments.

[0023] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0024] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0025] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant regions.

[0026] Figure 1 is a schematic flowchart of an occlusion target detection method provided in an embodiment of this disclosure. The method includes:

[0027] S11, obtain the feature matrix corresponding to the image to be identified. The elements in the feature matrix correspond to the spatial positions of the temporal lobe neurons in the temporal lobe neural network model. The elements represent the mapping vector of the image features of the image to be identified on the temporal lobe neurons.

[0028] The object represented in the image to be recognized can be a person, a living object, an inanimate object, a face, etc. The object may be partially occluded or unoccluded. The temporal lobe neural network model functions similarly to the ventral anterior temporal lobe cortex in the human brain, both used for object recognition and detection. The temporal lobe neural network model can contain multiple neurons. For ease of description, the neurons in the temporal lobe neural network model are named temporal lobe neurons.

[0029] In temporal lobe neural network models, the dominant temporal lobe neurons may differ depending on the type of object being recognized. Different regions of the temporal lobe neurons can recognize different types of objects. For example, in recognizing faces, the neurons in the middle region of the temporal lobe may be dominant. Similarly, in recognizing bodies, the neurons in the right side of the temporal lobe may be dominant.

[0030] Furthermore, the temporal lobe neural network model and the ventral anterior temporal lobe cortex in the human brain play the dominant roles in recognizing different types of objects. For example, in face recognition, neurons in the central region of the ventral anterior temporal lobe cortex in the human brain play a dominant role; similarly, temporal lobe neurons in the central region of the temporal lobe neural network model play a dominant role.

[0031] Image features of an image to be identified can characterize different types of objects within that image. These image features can be mapped onto the corresponding temporal lobe neurons in a temporal lobe neural network model. This gives the temporal lobe neurons a mapping vector, which the model can then use to identify and detect objects. The elements in the feature matrix represent these mapping vectors, and the spatial distribution of these elements corresponds to the spatial distribution of the temporal lobe neurons. The elements in the feature matrix can be directly assigned to the temporal lobe neurons corresponding to their spatial locations.

[0032] S12, based on the feature matrix and the frontal output matrix, update the activation state of the temporal lobe neurons to obtain the temporal output matrix. The temporal output matrix contains the updated activation state information of the temporal lobe neurons, and the frontal output matrix contains the updated activation state information of the frontal lobe neurons in the frontal neural network model.

[0033] In object recognition, the ventral frontal cortex in the human brain can enhance the stability of recognition by the ventral temporal cortex. The frontal lobe neural network model improves the accuracy, stability, and precision of object recognition by activating corresponding regions (neurons within the activated regions) in the temporal lobe neural network model by passing the frontal output matrix to the frontal lobe neural network model. A frontal lobe neural network model can contain multiple neurons. For ease of description, the neurons in the frontal lobe neural network model are named frontal lobe neurons.

[0034] In this embodiment, the feature matrix and the frontal lobe output matrix can be input into the temporal lobe neural network model to update the activation state of the temporal lobe neurons, thus obtaining the temporal lobe output matrix. The temporal lobe output matrix contains the updated activation state of the temporal lobe neurons. The temporal lobe output matrix may contain activation state information representing the updated activation state of each temporal lobe neuron.

[0035] S13. Based on the temporal lobe output matrix, update the activation state of the frontal lobe neurons to obtain a new frontal lobe output matrix.

[0036] In this embodiment, the temporal lobe output matrix can be input into the frontal lobe neural network model to update the activation state of the frontal lobe neurons, thus obtaining the frontal lobe output matrix. The frontal lobe output matrix contains the updated activation state of the frontal lobe neurons. The frontal lobe output matrix may contain activation state information representing the updated activation state of each frontal lobe neuron.

[0037] S14, perform the first iteration operation until the first termination condition is met, stop the first iteration operation, and obtain the first steady-state temporal lobe neural network model. The first iteration operation includes: updating the activation state of the frontal lobe neurons and updating the activation state of the temporal lobe neurons.

[0038] In the disclosed embodiments, the first iteration operation may include two steps, S12 and S13. The first termination condition may be a condition for stopping the first iteration operation. For example, the first termination condition may be: the difference in activation state of temporal lobe neurons in the temporal lobe network model after two consecutive first iteration operations is less than a temporal lobe activation difference threshold. Another example: the first termination condition may be: the number of executions of the first iteration operation reaches a first-time threshold. In one example, the first-time threshold may be 30,000 times. For ease of description, the temporal lobe neural network model that stops the first iteration operation is named the first steady-state temporal lobe neural network model.

[0039] S15, Based on the first steady-state temporal lobe neural network model, process the image to be identified to obtain the recognition result.

[0040] In this embodiment, the image to be recognized can be input into a first steady-state temporal lobe neural network model to obtain a recognition result. Alternatively, the first steady-state temporal lobe neural network model can be subjected to multiple rounds of activation state self-updates to obtain a new temporal lobe neural network model. The image to be recognized is then input into the new temporal lobe neural network model to obtain a recognition result.

[0041] In this embodiment, image features in the image to be identified can be mapped to a feature matrix, and then the mapping vector on the temporal lobe neurons can be determined through the feature matrix. This allows image features representing different objects in the image to be mapped to temporal lobe neurons in corresponding regions, improving the recognition accuracy of the temporal lobe neural network model. The temporal lobe neural network model and the frontal lobe neural network model undergo multiple updates and interactions of neuron activation states, ensuring that the temporal lobe neurons most suitable for the image to be identified are activated and reach a stable state. Using the first stable-state temporal lobe neural network to process the image to be identified improves the accuracy of the recognition result. Furthermore, the method in this embodiment possesses adaptive adjustment capabilities, enabling dynamic adjustment between the frontal lobe neural network model and the temporal lobe neural network model to maintain recognition accuracy when faced with different types of occlusion, thus enhancing the accuracy and stability of the recognition.

[0042] In one possible implementation, before updating the activation state of the temporal lobe neurons based on the feature matrix and the frontal lobe output matrix to obtain the temporal lobe output matrix, the method further includes: initializing the activation state of each temporal lobe neuron in the initial temporal lobe neural network model based on the feature matrix to obtain an initial activation state matrix; and controlling the initial temporal lobe neural network model to perform multiple activation state self-updates based on the feature matrix and the initial activation state matrix until a second termination condition is met to obtain the temporal lobe neural network model.

[0043] In this embodiment, multiple rounds of activation state self-updating can be performed. Activation state self-updating can be achieved by using the activation state and feature matrix of each temporal lobe neuron in the current round to determine the activation state of each temporal lobe neuron in the next round. For ease of understanding, formula (1) is used to represent activation state self-updating.

[0044] (1)

[0045] in, Represents the characteristic matrix, This represents the temporal lobe activation state matrix corresponding to the self-updating of activation states in round t. This temporal lobe activation state matrix contains the activation states of each temporal lobe neuron. This represents the temporal lobe activation state matrix corresponding to the self-updating of the activation state in round t+1. () represents an activation function, for example, () can be a non-linear activation function. When the independent variable of the activation function is not less than 0, the dependent variable is 1; when the independent variable of the activation function is less than 0, the dependent variable is -1.

[0046] For ease of understanding, we can let t=0, and then... Formula (1) is used to determine the temporal lobe activation state matrix corresponding to the first round. Thus, formula (1) exemplarily demonstrates the process of initializing the activation state of each temporal lobe neuron based on the feature matrix.

[0047] The second termination condition can be a condition that stops the self-updating of activation states. For example, the second termination condition could be: the difference in activation states of temporal lobe neurons in the initial temporal lobe network model after two consecutive self-updating of activation states is less than a temporal lobe activation difference threshold. Another example: the second termination condition could be: the number of self-updating of activation states reaches a second threshold. In one example, the second threshold could be 50,000 times.

[0048] In this embodiment, the initial temporal lobe neural network model is first initialized, and the activation state is updated multiple times to obtain the temporal lobe neural network model. In this way, before interacting with the frontal lobe neural network model, the temporal lobe neural network itself has already performed preliminary activation of the temporal lobe neurons in the corresponding region of the image to be recognized, providing the most accurate information possible for the first round of interaction with the frontal lobe neural network model, which can improve the probability of accurately activating the temporal lobe neurons in the end.

[0049] In one possible implementation, the step of processing the image to be identified based on the first steady-state temporal lobe neural network model to obtain the identification result includes: controlling the first steady-state temporal lobe neural network model to perform multiple activation state self-updates based on the activation state of each temporal lobe neuron in the first steady-state temporal lobe neural network model until a third termination condition is met to obtain a second steady-state neural network model; and processing the image to be identified based on the second steady-state neural network model to obtain the identification result.

[0050] In this embodiment of the disclosure, the termination condition for the activation state of each temporal lobe neuron in the first steady-state temporal lobe neural network model can be a condition for stopping the self-updating of the activation state. For example, the third termination condition can be: the difference in activation state of the temporal lobe neurons in the first steady-state temporal lobe network model after two consecutive self-updating of activation states is less than a temporal lobe activation difference threshold. Another example: the third termination condition can be: the number of times the activation state self-updating is executed reaches a third threshold. In one example, the third threshold can be 80,000 times.

[0051] In this embodiment, after the temporal lobe neural network model and the frontal lobe neural network model reach a steady state through activation state interaction, the activation state is self-updated based on the activation state of the temporal lobe neurons at the steady state. Thus, after the frontal lobe neural network model helps the temporal lobe neural network model activate the region corresponding to the image to be recognized, the temporal lobe neural network model continues to self-update its activation state, further improving the stability and accuracy of the activated temporal lobe neurons.

[0052] In one possible implementation, performing the first iteration operation includes: determining the target frontal lobe neuron corresponding to a first time step, and determining the target temporal lobe neuron corresponding to the first time step; updating the activation state of the target temporal lobe neuron based on the feature matrix and the frontal lobe output matrix corresponding to a second time step, obtaining the activation state value of the target temporal lobe neuron corresponding to the second time step, and determining the temporal lobe output matrix corresponding to the second time step based on the activation state value of the target temporal lobe neuron corresponding to the second time step, wherein the second time step is the next time step after the first time step; updating the activation state of the target frontal lobe neuron based on the temporal lobe output matrix corresponding to the second time step, obtaining the activation state value of the target frontal lobe neuron corresponding to a third time step, and based on the feature matrix and the frontal lobe output matrix corresponding to the second time step, determining the temporal lobe output matrix corresponding to the second time step, wherein the second time step is the next time step after the first time step; and updating the activation state of the target frontal lobe neuron based on the temporal lobe output matrix corresponding to the second time step, obtaining the activation state value of the target frontal lobe neuron corresponding to a third time step, and based on the feature matrix and the frontal lobe output matrix corresponding to the second time step, determining the temporal lobe output matrix corresponding to the third time step, and determining the temporal lobe output matrix corresponding to the third time step. Based on the activation state value of the target frontal lobe neuron at the third time point, determine the frontal lobe output matrix corresponding to the third time point, where the third time point is the next time point after the second time point; take the second time point as the new first time point and the third time point as the new second time point, and perform a second iteration operation until the first termination condition is met, then stop the second iteration operation. The second iteration operation includes: determining the target frontal lobe neuron corresponding to the first time point, determining the target temporal lobe neuron corresponding to the first time point, updating the activation state of the target frontal lobe neuron, determining the temporal lobe output matrix corresponding to the second time point, updating the activation state of the target temporal lobe neuron, and determining the frontal lobe output matrix corresponding to the third time point.

[0053] In this embodiment of the disclosure, for each second iteration, one or more target temporal lobe neurons can be randomly selected. Alternatively, one or more target temporal lobe neurons can be selected in a preset order. This embodiment of the disclosure does not limit this. The frontal lobe neuron that corresponds spatially to the target temporal lobe neuron in the frontal lobe neural network can be used as the target frontal lobe neuron.

[0054] For ease of understanding, each second iteration includes data corresponding to three time points (time 1, time 2, and time 3). After each second iteration, the data corresponding to time 3 is used as the new data corresponding to time 2, and the data corresponding to time 2 is used as the new data corresponding to time 1. Then, the second iteration continues.

[0055] The temporal lobe neural network model can use the frontal lobe output matrix corresponding to the second time step to update the activation state of the target temporal lobe neuron. The updated activation state of the temporal lobe neuron corresponds to the first time step. This yields the activation state value of the target temporal lobe neuron at the second time step. After updating the activation state value, the temporal lobe neural network model can output the temporal lobe output matrix corresponding to the second time step. This output matrix contains information representing the activation state value of the target temporal lobe neuron at that time step. Before the first execution of the second iteration, the temporal lobe output matrix can be 0.

[0056] The frontal lobe neural network model can use the temporal lobe output matrix corresponding to the second time step to update the activation state of the target frontal lobe neuron. The updated activation state of the frontal lobe neuron corresponds to the third time step. This allows the model to obtain the activation state value of the target frontal lobe neuron at the third time step. After updating the activation state value, the frontal lobe neural network model can output the frontal lobe output matrix corresponding to the third time step. This output matrix contains information representing the activation state value of the target frontal lobe neuron at that time step.

[0057] In this embodiment of the disclosure, each time the second iteration operation is performed, the frontal lobe neural network model and the temporal lobe neural network model update the activation state of their respective neurons and interact with each other's activation states.

[0058] The frontal lobe neural network model can adjust the frontal lobe output matrix based on the temporal lobe output matrix to more accurately activate the region in the temporal lobe neural network model corresponding to the image to be recognized. The temporal lobe neural network model can adjust the activation state of temporal lobe neurons based on the frontal lobe output matrix and feed back the temporal lobe output matrix containing the updated activation state information to the frontal lobe neural network model to obtain instructions for the next adjustment of the activation state from the frontal lobe neural network. After multiple rounds of this second iteration, the temporal lobe neurons corresponding to the image to be recognized in the temporal lobe neural network can be accurately activated, improving the accuracy and precision of recognition.

[0059] In one possible implementation, the frontal lobe activation state matrix represents the activation state of each frontal lobe neuron in the frontal lobe neural network model, and the temporal lobe activation state matrix represents the activation state of each temporal lobe neuron in the temporal lobe neural network model; the step of updating the activation state of the target temporal lobe neuron based on the feature matrix and the frontal output matrix corresponding to the second time step to obtain the activation state value of the target temporal lobe neuron at the second time step, and determining the temporal lobe output matrix corresponding to the second time step based on the activation state value of the target temporal lobe neuron at the second time step, includes: determining the activation state value of the target temporal lobe neuron at the second time step based on the frontal output matrix corresponding to the second time step, the temporal lobe activation state matrix corresponding to the first time step, the feature matrix, and the temporal intratemporal connection weight matrix of the target temporal lobe neuron and connected temporal lobe neurons; updating the temporal lobe activation state matrix corresponding to the first time step based on the activation state value of the target temporal lobe neuron at the second time step to obtain the temporal lobe activation state matrix corresponding to the second time step; and determining the temporal lobe output matrix corresponding to the second time step based on the activation state value of the target temporal lobe neuron at the second time step. The process involves: determining the temporal lobe output matrix at the second time step using the corresponding temporal lobe activation state matrix and the temporofrontal connection weight matrix; updating the activation state of the target frontal lobe neuron based on the temporal lobe output matrix at the second time step to obtain the activation state value of the target frontal lobe neuron at the third time step; and determining the frontal lobe output matrix at the third time step based on the activation state value of the target frontal lobe neuron at the third time step, including: determining the activation state value of the target frontal lobe neuron at the third time step based on the temporal lobe output matrix at the second time step, the frontal lobe activation state matrix at the second time step, and the intrafrontal connection weight matrix between the target temporal lobe neuron and the connected frontal lobe neurons; updating the frontal lobe activation state matrix at the second time step based on the activation state value of the target frontal lobe neuron at the third time step to obtain the frontal lobe activation state matrix at the third time step; and determining the frontal lobe output matrix at the third time step based on the temporal lobe activation state matrix at the third time step and the frontotemporal connection weight matrix.

[0060] The first and second moments can be two adjacent moments, with the first moment preceding the second moment. For ease of understanding, formula (2) is used to represent the process of determining the activation state value of the target temporal lobe neuron corresponding to the second moment.

[0061] (2)

[0062] Here, the target temporal lobe neuron is temporal lobe neuron number i, t represents time t, let time t be the first time, and let time t+1 be the second time. This represents the activation state value of the target temporal lobe neuron at the second time step. This represents the temporal lobe activation state matrix corresponding to the first time step. This represents the intrafrontal connectivity weight matrix of temporal lobe neurons connected to the target temporal lobe neuron. The intrafrontal connectivity weight matrix contains multiple intrafrontal connectivity weights, and each intrafrontal connectivity weight corresponds to a single temporal lobe neuron connected to the target temporal lobe neuron. Represents the characteristic matrix, This represents the temporal lobe output matrix corresponding to the second time step.

[0063] In one example, the temporal lobe activation state matrix can contain the activation state of each temporal lobe neuron. For example, 1 can be used to represent activation, and -1 to represent inactivation. The activation state value of the target temporal lobe neuron at the second time step can be used to replace the corresponding position in the temporal lobe activation state matrix at the first time step. This yields the temporal lobe activation state matrix for the second time step.

[0064] For ease of understanding, formula (3) is used to represent the process of the temporal lobe output matrix at the second time step.

[0065] (3)

[0066] in, This represents the temporal lobe output matrix. This represents the temporofrontal connection weight matrix.

[0067] For ease of understanding, Equation (4) is used to represent the activation state value of the target frontal lobe neuron corresponding to the third time step.

[0068] (4)

[0069] Let time t+2 be the third time. This represents the activation state value of the target frontal lobe neuron at the third time step. This represents the frontal lobe activation state matrix corresponding to the second time step. This represents the temporal lobe output matrix corresponding to the second time step. This represents the intrafrontal connection weight matrix between the target frontal lobe neuron and its connected frontal lobe neurons. The intrafrontal connection weight matrix contains multiple intrafrontal connection weights, and each intrafrontal connection weight corresponds to a frontal lobe neuron connected to the target frontal lobe neuron.

[0070] In one example, the frontal lobe activation state matrix can contain the activation state of each frontal lobe neuron. For example, 1 can represent activation, and -1 can represent inactivation. The activation state value of the target frontal lobe neuron at time 3 can be used to replace the corresponding position in the frontal lobe activation state matrix at time 2. This yields the frontal lobe activation state matrix at time 3.

[0071] For ease of understanding, formula (5) is used to represent the process of determining the frontal output matrix corresponding to the third time step.

[0072] (5)

[0073] in, This represents the frontal lobe output matrix at the third time step. This represents the frontotemporal connection weight matrix, where the frontotemporal connection weight matrix and the temporofrontal connection weight matrix are transposes of each other. This represents the frontal lobe activation state matrix corresponding to the third time step. This represents the magnification factor, for example: a=4.

[0074] Using the method of this disclosure, frontal lobe neurons and temporal lobe neurons can be updated asynchronously and in a targeted manner in each round of the second iteration. Each time, one or more target frontal lobe neurons and their corresponding target temporal lobe neurons are focused, making the activation state more targeted. Furthermore, the signal corresponding to the frontal lobe output matrix is ​​stronger, allowing the frontal lobe neural network model to play a dominant role in the second iteration and more accurately guide the temporal lobe neural network model to activate the temporal lobe neurons corresponding to the image to be identified.

[0075] In one possible implementation, obtaining the feature matrix corresponding to the image to be identified includes: extracting features from the image to be identified to obtain a feature vector, the feature vector including multiple feature values; mapping the multiple feature values ​​to positions in a two-dimensional matrix corresponding to the temporal lobe neurons based on a first weight vector corresponding to the temporal lobe neurons, determining the elements in the two-dimensional matrix, and obtaining the feature matrix, wherein the first weight vector contains a first weight corresponding to the feature values.

[0076] A two-dimensional matrix can serve as the initial state of the feature matrix. The spatial distribution of elements in the two-dimensional matrix corresponds to the spatial distribution of temporal lobe neurons. The values ​​of the elements in the two-dimensional matrix can be empty or preset values. Each element in the two-dimensional matrix (or the temporal lobe neuron corresponding to an element) can correspond to a first weight vector. Each weight in the first weight vector is multiplied by its corresponding eigenvalue to obtain a product, and the sum of these products is used as an element in the feature matrix. In this way, the feature matrix can be filled with elements that convey the feature meaning of the image to be recognized, allowing the feature matrix to both contain image features and accurately map those features onto each temporal lobe neuron.

[0077] In one possible implementation, the method for training the temporofrontal connection weight matrix includes: obtaining a sample feature vector corresponding to a sample image; determining a winning frontal lobe neuron based on the sample feature vector and the temporofrontal connection weights corresponding to each frontal lobe neuron; updating the temporofrontal connection weights corresponding to other frontal lobe neurons based on the position of the winning frontal lobe neuron, the positions of other frontal lobe neurons, the temporofrontal connection weights corresponding to other frontal lobe neurons, and the frontal lobe activation state matrix; performing a third iteration operation until a fourth termination condition is met, and stopping the third iteration operation, wherein the third iteration operation includes: obtaining the sample feature vector corresponding to the sample image, determining the winning frontal lobe neuron, and updating the temporofrontal connection weights corresponding to other frontal lobe neurons.

[0078] In this embodiment of the disclosure, feature vectors of sample images can be obtained. For ease of differentiation, the feature vectors of sample images are named sample feature vectors. The temporofrontal connection weights corresponding to each frontal lobe neuron can be compared with the sample feature vectors to determine the differences, such as the difference values, between the temporofrontal connection weights corresponding to each frontal lobe neuron and the sample feature vectors. The frontal lobe neuron with the smallest difference is selected as the winning frontal lobe neuron.

[0079] The temporofrontal connection weights corresponding to other frontal lobe neurons are adjusted in a direction that reduces the difference from the frontal lobe activation state matrix. The distances between the winning frontal lobe neuron and the winning frontal lobe neuron can be determined using the positions of the winning and other frontal lobe neurons. These distances are then used to control the adjustment magnitude.

[0080] For ease of understanding, Equation (6) is used to represent the process of updating the temporofrontal connection weights of each frontal lobe neuron except the winning frontal lobe neuron. For ease of description, each frontal lobe neuron except the winning frontal lobe neuron is named the frontal lobe neuron to be updated.

[0081] (6)

[0082] in, This represents the temporofrontal connection weights of the frontal lobe neurons to be updated at time t+1. This represents the temporofrontal connection weights of the frontal lobe neurons to be updated at time t. This represents the frontal lobe activation state matrix. This represents the neighborhood function between the frontal lobe neuron to be updated and the winning frontal lobe neuron. r represents the location coordinates of the frontal lobe neuron to be updated, s represents the location coordinates of the winning frontal lobe neuron, d represents the diameter of the region containing the frontal lobe neuron to be updated and the winning frontal lobe neuron, and e is the natural base.

[0083] The temporofrontal connection weights of the frontal lobe neurons to be updated at time t+1 are used as the updated temporofrontal connection weights of the frontal lobe neurons. Conversely, the temporofrontal connection weights of the frontal lobe neurons to be updated at time t are used as the original temporofrontal connection weights of the frontal lobe neurons. This updates the temporofrontal connection weight matrix.

[0084] The fourth termination condition can be a condition for stopping the training of the temporal frontal connection weight matrix. For example, the fourth termination condition can be: the difference between the temporal frontal connection weight matrices obtained from two consecutive third iterations is less than the first difference threshold. Alternatively, the fourth termination condition can be: the number of executions of the third iteration reaches the fourth number threshold.

[0085] In this embodiment of the disclosure, the temporal frontal connectivity weight matrix can be accurately determined through the third iteration operation.

[0086] In one possible implementation, the method for training the intrafrontal connection weight matrix includes: determining activation regions in the frontal neural network model based on sample images, the activation regions including: a living activation region composed of frontal lobe neurons that recognize living images, and a non-living activation region composed of frontal lobe neurons that recognize non-living images; for each temporal lobe neuron, determining a first activation value for each frontal lobe neuron relative to living and non-living patterns according to the activation region where the frontal lobe neuron is located; and determining the intrafrontal connection weights corresponding to each frontal lobe neuron based on the first activation values ​​corresponding to each frontal lobe neuron, thereby obtaining the intrafrontal connection weight matrix.

[0087] Sample images can include images representing living organisms as well as images representing non-living organisms. Frontal lobe neurons in living activation areas are more easily activated when recognizing images representing living organisms. Frontal lobe neurons in non-living activation areas are more easily activated when recognizing images representing non-living organisms.

[0088] The living pattern represents the images that frontal lobe neurons process, representing living organisms. The non-living pattern represents the images that frontal lobe neurons process, representing non-living organisms.

[0089] The first activation value can characterize the activation state of frontal lobe neurons in different activation regions in response to the sample image.

[0090] For example, when frontal lobe neurons in the life activation area recognize sample images representing living organisms, the first activation value can be 1; when frontal lobe neurons in the life activation area recognize sample images representing non-living organisms, the first activation value can be -1.

[0091] For example, when frontal lobe neurons in non-living activation areas recognize sample images representing living organisms, the first activation value can be -1; when frontal lobe neurons in non-living activation areas recognize sample images representing living organisms, the first activation value can be 1.

[0092] The intrafrontal connection weights between frontal neurons are established based on the Hebbian learning rule and the anti-Hebbian learning rule, which allow the frontal neural network model to strengthen or weaken the connection strength according to the relevant activity of the neurons.

[0093] For ease of understanding, Equation (7) is used to represent the process of updating the intra-connection weights.

[0094] (7)

[0095] Where μ=1 represents the living mode and μ=2 represents the non-living mode. This indicates the activation state of neuron i in the frontal lobe under living and non-living modes. This indicates the activation state of neuron j in the frontal lobe under both living and non-living modes. The weight of the intrafrontal connection between frontal lobe neuron i and frontal lobe neuron j is represented by N, where N represents the number of frontal lobe neurons.

[0096] In this way, the intrafrontal connection weights can be accurately determined, and thus the intrafrontal connection weight matrix can be determined.

[0097] This disclosure provides an occluded target detection neural network model, comprising: a deep convolutional neural network model, a frontal lobe neural network model, and a temporal lobe neural network model; the deep convolutional neural network model and the temporal lobe neural network model are data-connected, and the temporal lobe neural network model and the frontal lobe neural network model are data-connected; the deep convolutional neural network model is used to extract features from the image to be recognized, obtaining a feature vector, the feature vector including multiple feature values; the temporal lobe neural network model is used to map the multiple feature values ​​to positions corresponding to the temporal lobe neurons in a two-dimensional matrix based on a first weight vector corresponding to the temporal lobe neurons, determining the elements in the two-dimensional matrix to obtain a feature matrix; the frontal lobe neural network model is used to update the activation state of the temporal lobe neurons by interacting with the temporal lobe neural network model regarding the activation states of their respective neurons, so that the temporal lobe neural network model reaches a steady state; when the temporal lobe neural network model is in a steady state, it is further used to determine the recognition result based on the image to be recognized.

[0098] The frontal lobe neural network model and the temporal lobe neural network model can be the Hopfield neural network model.

[0099] In this embodiment, multiple updates and interactions of neuron activation states are performed using a frontal lobe neural network model and a temporal lobe neural network model, ensuring that the frontal lobe neurons most suitable for the image to be recognized are activated in the frontal lobe neural network model. Furthermore, this occluded target detection neural network model possesses adaptive adjustment capabilities, dynamically adjusting between the temporal lobe neural network model and the frontal lobe neural network model when faced with different types of occlusion, maintaining recognition accuracy and enhancing the accuracy and stability of recognition. Moreover, unsupervised training can be completed with only a small number of samples, effectively achieving the desired training effect.

[0100] This disclosure also provides a computer program product, including a computer program or a non-volatile computer-readable storage medium carrying the computer program, wherein the computer program, when executed by a processor, implements the steps of the above method.

[0101] Computer-readable storage media can be tangible devices capable of holding and storing programs / instructions used by instruction execution devices. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0102] The computer program (or computer-readable program instructions) described herein can be downloaded from a computer-readable storage medium to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage medium in the respective computing / processing device.

[0103] The computer program (or computer program instructions) used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions to implement various aspects of this disclosure.

[0104] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0105] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0106] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0107] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0108] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for detecting occluded targets, characterized in that, include: Obtain the feature matrix corresponding to the image to be identified. The elements in the feature matrix correspond to the spatial positions of the temporal lobe neurons in the temporal lobe neural network model. The elements represent the mapping vectors of the image features of the image to be identified on the temporal lobe neurons. Based on the feature matrix and the frontal lobe output matrix, update the activation state of the temporal lobe neurons to obtain the temporal lobe output matrix. The temporal lobe output matrix contains the updated activation state information of the temporal lobe neurons. The frontal lobe output matrix contains the updated activation state information of the frontal lobe neurons in the frontal lobe neural network model. Based on the temporal lobe output matrix, update the activation state of the frontal lobe neurons to obtain a new frontal lobe output matrix. Perform a first iteration operation until a first termination condition is met, then stop the first iteration operation to obtain a first steady-state temporal lobe neural network model. The first iteration operation includes: updating the activation state of the frontal lobe neurons and updating the activation state of the temporal lobe neurons. Based on the first steady-state temporal lobe neural network model, the image to be identified is processed to obtain the recognition result.

2. The method according to claim 1, characterized in that, Before updating the activation state of the temporal lobe neurons based on the feature matrix and the frontal lobe output matrix to obtain the temporal lobe output matrix, the method further includes: initializing the activation state of each temporal lobe neuron in the initial temporal lobe neural network model based on the feature matrix to obtain an initial activation state matrix; and controlling the initial temporal lobe neural network model to perform multiple activation state self-updates based on the feature matrix and the initial activation state matrix until the second termination condition is met to obtain the temporal lobe neural network model.

3. The method according to claim 1, characterized in that, The step of processing the image to be identified based on the first steady-state temporal lobe neural network model to obtain the recognition result includes: controlling the first steady-state temporal lobe neural network model to perform multiple activation state self-updates based on the activation state of each temporal lobe neuron in the first steady-state temporal lobe neural network model until a third termination condition is met to obtain a second steady-state neural network model; and processing the image to be identified based on the second steady-state neural network model to obtain the recognition result.

4. The method according to claim 1, characterized in that, The execution of the first iteration operation includes: determining the target frontal lobe neuron corresponding to the first time step, and determining the target temporal lobe neuron corresponding to the first time step; updating the activation state of the target temporal lobe neuron based on the feature matrix and the frontal lobe output matrix corresponding to the second time step, obtaining the activation state value of the target temporal lobe neuron corresponding to the second time step, and determining the temporal lobe output matrix corresponding to the second time step based on the activation state value of the target temporal lobe neuron corresponding to the second time step, wherein the second time step is the next time step after the first time step; updating the activation state of the target frontal lobe neuron based on the temporal lobe output matrix corresponding to the second time step, obtaining the activation state value of the target frontal lobe neuron corresponding to the third time step, and determining the temporal lobe output matrix corresponding to the third time step based on the activation state value of the target temporal lobe neuron corresponding to the third time step. The activation state value of the target frontal lobe neuron at the corresponding moment is used to determine the frontal lobe output matrix at the third moment, where the third moment is the next moment after the second moment. The second moment is taken as the new first moment, and the third moment is taken as the new second moment. The second iteration operation is performed until the first termination condition is met, and then the second iteration operation is stopped. The second iteration operation includes: determining the target frontal lobe neuron corresponding to the first moment, determining the target temporal lobe neuron corresponding to the first moment, updating the activation state of the target frontal lobe neuron, determining the temporal lobe output matrix corresponding to the second moment, updating the activation state of the target temporal lobe neuron, and determining the frontal lobe output matrix corresponding to the third moment.

5. The method according to claim 1, characterized in that, The frontal lobe activation state matrix represents the activation state of each frontal lobe neuron in the frontal lobe neural network model, and the temporal lobe activation state matrix represents the activation state of each temporal lobe neuron in the temporal lobe neural network model. The step of updating the activation state of the target temporal lobe neuron based on the feature matrix and the frontal output matrix corresponding to the second time step to obtain the activation state value of the target temporal lobe neuron at the second time step, and determining the temporal lobe output matrix corresponding to the second time step based on the activation state value of the target temporal lobe neuron at the second time step, includes: determining the activation state value of the target temporal lobe neuron at the second time step based on the frontal output matrix corresponding to the second time step, the temporal lobe activation state matrix corresponding to the first time step, the feature matrix, and the temporal intratemporal connection weight matrix between the target temporal lobe neuron and connected temporal lobe neurons; updating the temporal lobe activation state matrix corresponding to the first time step based on the activation state value of the target temporal lobe neuron at the second time step to obtain the temporal lobe activation state matrix corresponding to the second time step; and determining the temporal lobe output matrix corresponding to the second time step based on the activation state value of the target temporal lobe neuron at the second time step. The activation state matrix and the temporofrontal connection weight matrix are used to determine the temporal lobe output matrix corresponding to the second time step. The activation state of the target frontal lobe neuron is updated based on the temporal lobe output matrix corresponding to the second time step to obtain the activation state value of the target frontal lobe neuron corresponding to the third time step. Based on the activation state value of the target frontal lobe neuron corresponding to the third time step, the frontal lobe output matrix corresponding to the third time step is determined. This includes: determining the activation state value of the target frontal lobe neuron corresponding to the third time step based on the temporal lobe output matrix corresponding to the second time step, the frontal lobe activation state matrix corresponding to the second time step, and the intrafrontal connection weight matrix between the target temporal lobe neuron and its connected frontal lobe neurons; updating the frontal lobe activation state matrix corresponding to the second time step based on the activation state value of the target frontal lobe neuron corresponding to the third time step to obtain the frontal lobe activation state matrix corresponding to the third time step; and determining the frontal lobe output matrix corresponding to the third time step based on the temporal lobe activation state matrix corresponding to the third time step and the frontotemporal connection weight matrix.

6. The method according to claim 1, characterized in that, The step of obtaining the feature matrix corresponding to the image to be identified includes: extracting features from the image to be identified to obtain a feature vector, the feature vector including multiple feature values; mapping the multiple feature values ​​to the positions corresponding to the temporal lobe neurons in a two-dimensional matrix based on the first weight vector corresponding to the temporal lobe neurons, determining the elements in the two-dimensional matrix, and obtaining the feature matrix, wherein the first weight vector contains a first weight corresponding to the feature values.

7. The method according to claim 5, characterized in that, The method for training the temporofrontal connection weight matrix includes: obtaining sample feature vectors corresponding to sample images; determining the winning frontal lobe neuron based on the sample feature vectors and the temporofrontal connection weights corresponding to each frontal lobe neuron; updating the temporofrontal connection weights corresponding to other frontal lobe neurons based on the position of the winning frontal lobe neuron, the positions of other frontal lobe neurons, the temporofrontal connection weights corresponding to other frontal lobe neurons, and the frontal lobe activation state matrix; performing a third iteration operation until a fourth termination condition is met, and stopping the third iteration operation, wherein the third iteration operation includes: obtaining the sample feature vectors corresponding to the sample images, determining the winning frontal lobe neuron, and updating the temporofrontal connection weights corresponding to other frontal lobe neurons.

8. The method according to claim 5, characterized in that, The method for training the intrafrontal connection weight matrix includes: determining activation regions in the frontal lobe neural network model based on sample images, wherein the activation regions include: a living activation region composed of frontal lobe neurons that recognize living images, and a non-living activation region composed of frontal lobe neurons that recognize non-living images; for each temporal lobe neuron, determining a first activation value for each frontal lobe neuron relative to living and non-living patterns according to the activation region where the frontal lobe neuron is located; and determining the intrafrontal connection weights corresponding to each frontal lobe neuron based on the first activation values ​​corresponding to each frontal lobe neuron, thereby obtaining the intrafrontal connection weight matrix.

9. A neural network model for detecting occluded targets, characterized in that, The occlusion target detection neural network model includes: a deep convolutional neural network model, a frontal lobe neural network model, and a temporal lobe neural network model; the deep convolutional neural network model and the temporal lobe neural network model are data-connected, and the temporal lobe neural network model and the frontal lobe neural network model are data-connected; the deep convolutional neural network model is used to extract features from the image to be recognized to obtain a feature vector, the feature vector including multiple feature values; the temporal lobe neural network model is used to map the multiple feature values ​​to the positions corresponding to the temporal lobe neurons in a two-dimensional matrix based on the first weight vector corresponding to the temporal lobe neurons, to determine the elements in the two-dimensional matrix, and obtain a feature matrix; the frontal lobe neural network model is used to update the activation state of the temporal lobe neurons by interacting with the temporal lobe neural network model regarding the activation state of their respective neurons, so that the temporal lobe neural network model reaches a steady state; when the temporal lobe neural network model is in a steady state, it is further used to determine the recognition result based on the image to be recognized.

10. A computer program product comprising a computer program, or a non-volatile computer-readable storage medium carrying a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

Images