An image orientation identification method, device, equipment and storage medium

Abstract

The application discloses an image orientation recognition method and device, equipment and a storage medium. The method comprises the following steps: using a projection neural network to preprocess a target image to obtain standard image data, the preprocessing comprising dimension reduction, noise reduction, normalization, cutting and gain control; encoding the standard image data according to APL neurons and a WTA mechanism through a preset KC layer to obtain encoded image data, the projection neural network and the preset KC layer comprising first synaptic weights subject to random distribution; inputting the encoded image data into a preset MBON layer for orientation analysis to obtain an orientation recognition result, the preset KC layer and the preset MBON layer comprising second synaptic weights updated based on a preset update algorithm. The application avoids the problems of catastrophic forgetting or low learning speed of the existing image orientation recognition model due to large distribution difference of samples with the same label, and avoids the problems of low learning and recognition efficiency of rare details in the image caused by the use of convolution kernels.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of biological neural networks, and particularly relates to an image orientation recognition method and device, equipment and a storage medium. BACKGROUND

[0002] In a natural environment, many cues can indicate direction, such as the direction of the sun, the shift of the sun, the direction of the wind blowing, and landmark buildings. Animals can use them for direction and navigation. Most of these cues are perceived visually. Even simple insects can use visual memory to remember the way home, which determines the way home by finding a familiar direction after walking along the route once. When memorizing the direction, their lightweight neural circuits are shorter than typical artificial neural networks (ANN). Therefore, the direction memory obtained from visual signals can be retrieved by studying the structure and learning rules of these circuits.

[0003] There have been many models for finding the direction of an object in an image plane; but these methods usually involve deep networks and require random ordering of many data samples during the training process, which is complex and requires a long training time, making it difficult to adapt to the real-time, fast, and low-power learning requirements in application scenarios such as robots. SUMMARY

[0004] The present application provides an image orientation recognition method, device, equipment and storage medium, which is used to solve the technical problem that the existing image orientation recognition model network is relatively complex, the training sample requirement is high, the operation is complex, a long training time is required, and the recognition efficiency is low.

[0005] Therefore, the first aspect of the present application provides an image orientation recognition method, comprising:

[0006] The target image is preprocessed using a projection neural network to obtain standard image data, and the preprocessing includes dimension reduction, noise reduction, normalization, cropping and gain control;

[0007] The standard image data is encoded by a preset KC layer according to an APL neuron and a WTA mechanism to obtain encoded image data, and the projection neural network and the preset KC layer include first synaptic weights subject to a random distribution;

[0008] The encoded image data is input into a preset MBON layer for orientation analysis to obtain an orientation recognition result, and the preset KC layer and the preset MBON layer include second synaptic weights updated based on a preset update algorithm.

[0009] Preferably, the encoding the standard image data according to the APL neuron and the WTA mechanism through the preset KC layer to obtain encoded image data comprises:

[0010] The first synaptic weight in a binary form between the projection neural network and the preset KC layer is randomly generated based on a Bernoulli distribution;

[0011] The mapping data is obtained through linear mapping according to the first synaptic weight;

[0012] The mapping data is laterally inhibited according to the APL neuron and the WTA mechanism, and is encoded into a sparse vector to obtain the encoded image data.

[0013] Preferably, the inputting the encoded image data into the preset MBON layer for orientation analysis to obtain an orientation recognition result comprises:

[0014] The synaptic weight between the preset KC layer and the preset MBON layer is updated and calculated based on a preset update algorithm to obtain a second synaptic weight;

[0015] The orientation recognition result is obtained through orientation activity analysis based on mapping of the encoded image data according to the second synaptic weight.

[0016] Preferably, the updating and calculating the synaptic weight between the preset KC layer and the preset MBON layer based on the preset update algorithm to obtain the second synaptic weight comprises:

[0017] The preset update algorithm is an update algorithm based on a Hebbian rule;

[0018] The synaptic weight between the preset KC layer and the preset MBON layer is learned and updated based on the Hebbian rule to obtain the second synaptic weight in a convergent state.

[0019] Preferably, the updating and calculating the synaptic weight between the preset KC layer and the preset MBON layer based on the preset update algorithm to obtain the second synaptic weight comprises:

[0020] The preset update algorithm is a binary-based update algorithm;

[0021] The synaptic weight between the preset KC layer and the preset MBON layer is updated in a binary form in real time based on the binary to obtain the second synaptic weight.

[0022] The second aspect of the application provides an image orientation recognition device, comprising:

[0023] a preprocessing unit configured to preprocess a target image using a projection neural network to obtain standard image data, the preprocessing including dimension reduction, noise reduction, normalization, cropping, and gain control;

[0024] an image encoding unit configured to encode the standard image data according to an APL neuron and a WTA mechanism through a preset KC layer to obtain encoded image data, the projection neural network and the preset KC layer including first synaptic weights subject to a random distribution;

[0025] a direction recognition unit configured to input the encoded image data into a preset MBON layer for direction analysis to obtain a direction recognition result, the preset KC layer and the preset MBON layer including second synaptic weights updated based on a preset update algorithm.

[0026] Preferably, the image encoding unit is specifically configured to:

[0027] the first synaptic weights between the projection neural network and the preset KC layer are randomly generated in a binary form based on a Bernoulli distribution;

[0028] linear mapping is performed according to the first synaptic weights to obtain mapping data;

[0029] the mapping data is subjected to lateral inhibition according to the APL neuron and the WTA mechanism, and is encoded into a sparse vector to obtain the encoded image data.

[0030] Preferably, the direction recognition unit is specifically configured to:

[0031] the synaptic weights between the preset KC layer and the preset MBON layer are updated and calculated based on a preset update algorithm to obtain the second synaptic weights;

[0032] the encoded image data is subjected to mapping-based direction activity analysis according to the second synaptic weights to obtain the direction recognition result.

[0033] The third aspect of the present application provides an image direction recognition device, the device comprising a processor and a memory;

[0034] the memory is configured to store program code and transmit the program code to the processor;

[0035] the processor is configured to execute the image direction recognition method of the first aspect according to instructions in the program code.

[0036] The fourth aspect of the present application provides a computer readable storage medium for storing program code, the program code being used to execute the image direction recognition method of the first aspect.

[0037] From the above technical solution, the embodiments of the application have the following advantages:

[0038] In the present application, an image orientation recognition method is provided, which comprises: pre-processing a target image using a projection neural network to obtain standard image data, the pre-processing including dimension reduction, noise reduction, normalization and gain control; encoding the standard image data according to APL neurons and WTA mechanism through a preset KC layer to obtain encoded image data, the projection neural network and the preset KC layer including first synaptic weights subject to random distribution; inputting the encoded image data into a preset MBON layer for orientation analysis to obtain an orientation recognition result, the preset KC layer and the preset MBON layer including second synaptic weights updated based on a preset update algorithm.

[0039] The image orientation recognition method provided by the present application analyzes and recognizes the orientation of a target in an image based on a mushroom body model, and incrementally learns the direction of an image object according to a biological learning rule. The model in this process only needs to be observed once to learn the direction of the target in the image and obtain the orientation recognition result. Moreover, the network structure is relatively simple, the requirements for training time and training samples are low, the orientation recognition result can be quickly and accurately obtained, the recognition efficiency is improved, and the needs of various application scenarios can be met. Therefore, the present application can solve the technical problem that the existing image orientation recognition model network is relatively complex, the requirements for training samples are high, the operation is complex, a long training time is required, and the recognition efficiency is low. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 A flowchart of an image orientation recognition method provided by an embodiment of the present application is shown;

[0041] Figure 2 A structural diagram of an image orientation recognition device provided by an embodiment of the present application is shown;

[0042] Figure 3 A schematic diagram of a fruit fly mushroom body biological structure and a corresponding simplified calculation model network structure provided by an embodiment of the present application is shown;

[0043] Figure 4 An object orientation example diagram of a COIL-100 dataset provided by an experimental application example of the present application is shown;

[0044] Figure 5 A precision and loss rate curve diagram on a COIL-100 dataset provided by an experimental application example of the present application is shown. DETAILED DESCRIPTION

[0045] In order for those skilled in the art to better understand the scheme of the present application, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0046] It is very interesting to explore how insects can remember the direction of a landmark by observing it once without caching image data and use its underlying mechanism for object direction retrieval or detection. The neural circuit of insects can perform sparse coding without learning, and the resulting sparse representation reduces the conflict in learning. The mushroom body (MB) of fruit flies is a typical example; the biological field has conducted detailed research on the nervous system of fruit flies, reconstructed the circuit in three dimensions, and analyzed the connection group data. Using these biological data can more accurately explain the working mechanism of fruit flies. Based on the working principle of fruit fly mushroom body, a more efficient and reliable image orientation recognition method can be conceived, and the present application is based on this principle.

[0047] For navigation tasks using visual information collected near the ground, landmark objects usually have a fixed angle relative to the ground, so the direction in the view plane is not so important relative to the angle of the object direction observer. The traditional object angle recognition focuses on determining which direction the object is located in the observer's view, but the model of the present application focuses on the orientation of the object itself in the image.

[0048] For ease of understanding, please refer to Figure 1 The embodiment of the image orientation recognition method provided by the present application comprises:

[0049] Step 101, using a projection neural network to preprocess the target image to obtain standard image data, the preprocessing includes dimension reduction, noise reduction, normalization, cropping and gain control.

[0050] It should be noted that the embodiment is based on the mushroom body to design the recognition model of the image target orientation, which specifically includes three layers; the first layer is a projection neural network layer composed of projection neurons (PNs), which is used for preprocessing the image; the second layer is a preset KC layer including a plurality of Kenyon cells (KCs), which is used for mapping and encoding the image; the third layer is a preset MBON layer, which is used for learning and analyzing the orientation possibility of the target. Therefore, the insect receptor input will undergo a series of preprocessing operations on the target image by the projection neurons PNs in the projection neural network before being transmitted to the Kenyon cells (KCs), which is the design based on the processing principle of the fruit fly mushroom body (MB).

[0051] Each Kenyon cell (KC) in the Drosophila mushroom body (MB) receives signals from several specific class of projection neurons (PNs), typically 3 to 5, and the KC population encodes the signals in a sparse coding manner. Only a small fraction of KCs can be activated at the same time, typically 5% to 15%, because the anterior paired-lateral neurons (APLs) receive inputs from all KCs and inhibit them. The axons of KCs pass through multiple segments, each typically with a mushroom body output neuron (MBON) for outputting the direction novelty based on visual analysis and a dopamine neuron for reward and learning regulation. Since the number of KCs is much larger than PNs and the activity of KCs is sparse, the existing connections between KCs and MBONs are unlikely to be affected by subsequent learning.

[0052] This is how the Drosophila mushroom body model in this embodiment works. For the biological structure of the mushroom body and the simplified mushroom body model in this embodiment, please refer to Figure 3 , the biological structure of the mushroom body will show the plasticity of the connection from the olfactory receptor neuron to the mushroom body output neuron, and the simplified model describes the mapping relationship between all neurons.

[0053] It can be seen that the mushroom body model of the present application includes class projection neurons PNs, a preset KC layer for encoding, and a preset MBON layer for output. The model ignores the morphology of neurons, uses a firing-rate neuron model instead of a spiking neuron model, and ignores the internal dynamics of neurons, and considers the synapses between neurons as linear mapping. In this way, the complexity of the model is maximized, and the processing efficiency of the model is improved.

[0054] Although the actual preprocessing of visual signals by insects can be complex, according to the previous model, the mushroom body structure can process and learn on the image without complex feature extraction, only down-sampling and normalization. In the first layer of the model in this embodiment, the original image is normalized so that the average value of pixel intensity is 0. Although an image has two dimensions, in order to generate this model on other modalities, each image is flattened into one dimension. Therefore, if a given dataset , each row represents a data point , where is the dimension of the data point, and the data is normalized by the mean of before being passed to the class projection neurons PNs. The standard image data obtained by preprocessing is expressed as:

[0055]

[0056] where, is the standard image data, and n represents the total amount of data.

[0057] In step 102, the standard image data is encoded by a preset KC layer according to an APL neuron and a WTA mechanism to obtain encoded image data, and the projection neural network and the preset KC layer include first synaptic weights subject to random distribution.

[0058] Further, in step 102, the method comprises:

[0059] The first synaptic weights in binary form between the projection neural network and the preset KC layer are randomly generated based on a Bernoulli distribution.

[0060] Linear mapping is performed according to the first synaptic weights to obtain mapping data.

[0061] The mapping data is laterally inhibited according to the APL neuron and the WTA mechanism, and is encoded into a sparse vector to obtain the encoded image data.

[0062] It should be noted that the standard image data is mapped and encoded between the projection neural network and the preset KC layer based on the first synaptic weights subject to random distribution. The specific random distribution can be a Bernoulli distribution or other distributions, which are only examples and are not limited herein. If the first synaptic weights are expressed as wherein, is the number of Kenyon-like cells (KCs), and the first synaptic weight is random and binary, and follows a Bernoulli distribution, that is, wherein, represents the probability of connection, and b represents the expectation of the connection of the projection-like neuron PNs to the Kenyon-like cell. Therefore, the mapping based on the first synaptic weight can be expressed as:

[0063]

[0064] wherein, is the standard image data obtained above, is the mapping data.

[0065] In the mapping process of the MB, the APL neuron will laterally inhibit the KCs, so only the KC with the strongest activity can play a role, and the output value of the first h most excited KCs is kept to approximate this WTA inhibition encoding mechanism, and the others are set to zero. The specific process is expressed as:

[0066]

[0067] where h can control the washout property of the encoding, and j is the index of the active excitatory KC. After WTA inhibition encoding, the output of the KCs is expressed as .

[0068] Since the KCs that are always active provide less useful information, threshold can be used to disable such KCs, for example, set the threshold to 0.25, if the KC remains active in more than one quarter of the images, then its output is 0.

[0069] Step 103, input the encoded image data into the preset MBON layer for orientation analysis to obtain an orientation recognition result, and the preset KC layer and the preset MBON layer include a second synaptic weight updated based on a preset update algorithm.

[0070] Further, step 103 comprises:

[0071] The preset update algorithm is a Hebbian rule-based update algorithm;

[0072] The preset update algorithm is a binary-based update algorithm;

[0073] Further, the preset update algorithm is a Hebbian rule-based update algorithm;

[0074] Further, the preset update algorithm is a binary-based update algorithm;

[0075] Further, the preset update algorithm is a Hebbian rule-based update algorithm;

[0076] Further, the preset update algorithm is a binary-based update algorithm;

[0077] Further, the preset update algorithm is a binary-based update algorithm;

[0078] Further, the preset update algorithm is a binary-based update algorithm;

[0079] It should be noted that in order to explain the associative learning of insects as supervised learning in machine learning, the weights between neural layers can be changed for learning improvement. Therefore, the preset update algorithm is used to update the synaptic weight from the preset KC layer to the preset MBON layer to obtain a more reliable MBON update weight, i.e. the second synaptic weight.

[0080] If the second synaptic weight is expressed as different preset update algorithms can be used to obtain different synaptic weights based on different ways. The output of the mapping-based orientation activity analysis of the encoded image data according to the second synaptic weight can be expressed as:

[0081]

[0082] The active excitation degree of each MBON represents the orientation angle possibility corresponding to the sample, and the most active angle is the orientation recognition result of the target.

[0083] The present embodiment uses two different methods to update and calculate the second synaptic weight. The first method is to use a method similar to the Hebbian rule to realize the plasticity rule of neurons, which is used to realize convergence and can adjust the synaptic weight multiple times. The second synaptic weight is expressed as:

[0084]

[0085] wherein, is the learning rate, and after each iteration of the sample, the learning rate is attenuated according to the rule, and the attenuation rule is expressed as: .

[0086] The second method is to update the second synaptic weight based on a binary method, which is specifically expressed as:

[0087]

[0088] This method initializes all the weights between the KCs and the MBONs to 0. During learning, when the image is sent into the network, the most active KCs are activated, and the label corresponding to the weight between the KCs and the MBONs is set to 1. Any one of the above two methods can be used to update the weight to obtain the updated weight of the MBON, i.e., the second synaptic weight.

[0089] For ease of understanding, the present application conducts experiments of the image orientation recognition method on the COIL dataset, which contains 100 objects, captured at 72 different angles, and a total of 7200 images. The size of the original image is 128x128, and for each image, there are 128x128x3=49152 value channels because the image is RGB color. The dataset is divided into two groups according to whether the object has axial symmetry and whether it has a texture pattern, which are COIL-100-Ordinary (COIL-100-O) and COIL-100-Axisymmetric (COIL-100-AS), respectively. For COIL-100-O, the object does not have axial symmetry or has a clear texture pattern. For COIL-100-AS, the object is axisymmetric and has no obvious texture pattern. In COIL-100-AS, different views of the same object are very similar that the human eye cannot even distinguish them. Please refer to Figure 4 for views of two objects, the first row is from COIL-100-O, and the second row is from COIL-100-AS.

[0090] In this study, the direction of the object is discretized into 72 directions, ranging from 0 to 355, and k is set to 72. The accuracy of the model provided by the present application in distinguishing objects in trained directions and unknown directions is evaluated.

[0091] The first experiment on the COIL dataset is the analysis of the direction of the object. This experiment does not distinguish between the training set and the test set. Instead, the model should analyze the direction of the object in the previously viewed images. This is conceptually simple, but since the same object shares the same features, it creates interference when a typical artificial neural network learns the direction. The effect of the number of KCs on the accuracy evaluates the effect of the number of active KCs on the retrieval accuracy. As the number of KCs increases, the accuracy of the method of the present application on the two datasets increases, and the accuracy tends to be stable when the number of KCs approaches 10000, please refer to Figure 5 (a). The present application selects 10240 MBONs. The results show that the algorithm model of the present application can achieve an accuracy of more than 90% on the two datasets after learning once, thereby identifying the direction of the observed object.

[0092] The baseline model needs longer training time for the same performance. The accuracy, training time, and incremental learning ability of the two weight update methods of the algorithm model of the present application are compared with deep convolutional neural networks such as AlexNet, GoogleNet, VGG16, and ResNet50. The change of the accuracy and loss of the baseline model with the increase of the number of iterations is shown in Figure 5 . The training time of the baseline model is shown in Figure 5In (b), the method of the present application shows as a horizontal line because it only needs to be learned once. Other models need 1.6 to 80.6 times as long to reach similar performance. Moreover, the method of the present application has not accelerated the model in the experiment. Therefore, the experimental results can prove that the present application can still accurately and reliably identify the orientation of the target object in the image without increasing the complexity of the model and model training, and has strong comprehensive performance.

[0093] The incremental learning ability calculates the accuracy of the previously trained objects to evaluate whether the model can correctly identify the direction of the previous objects after obtaining new objects. More accurately, after the i-th object is trained, the accuracy of the model for the direction of each object ≤i angle is evaluated. It can be found that the model of the present application can remember the old memory while learning new knowledge, even for axisymmetric objects that are difficult for humans to identify. All baseline models will forget the learned objects after 10 optimization iterations, but the model of the present application is almost not affected by the order of samples, that is, the sample sorting operation is not required at all during the model training process, which can minimize the complexity of model training and improve the efficiency of model training.

[0094] The image orientation identification method provided by the embodiment of the present application analyzes and identifies the orientation of the target in the image based on the mushroom model, and incrementally learns the direction of the image object according to the biological learning rule. The model in this process only needs to be observed once to learn the direction of the target in the image and obtain the orientation identification result. Moreover, the network structure is relatively simple, the requirements for training time and training samples are low, the orientation identification result can be quickly and accurately obtained, the identification efficiency is improved, and the needs of various application scenarios can be met. Therefore, the embodiment of the present application can solve the technical problem that the existing image orientation identification model network is complex, the training sample requirement is high, the operation is complex, a long training time is required, and the identification efficiency is low.

[0095] For ease of understanding, please refer to Figure 2 The present application provides an embodiment of an image orientation identification device, which comprises:

[0096] The preprocessing unit 201 is configured to preprocess the target image using a projection neural network to obtain standard image data, and the preprocessing includes dimension reduction, noise reduction, normalization, cropping and gain control.

[0097] The image encoding unit 202 is configured to encode the standard image data according to the APL neuron and the WTA mechanism through a preset KC layer to obtain encoded image data, and the projection neural network and the preset KC layer comprise first synaptic weights subject to a random distribution.

[0098] The orientation recognition unit 203 is configured to input the encoded image data into a preset MBON layer for orientation analysis to obtain an orientation recognition result, and the preset KC layer and the preset MBON layer include second synaptic weights updated based on a preset update algorithm.

[0099] Further, the image encoding unit 202 is specifically configured to:

[0100] The first synaptic weights in a binary form between the projection neural network and the preset KC layer are randomly generated based on a Bernoulli distribution;

[0101] Linear mapping is performed based on the first synaptic weights to obtain mapping data;

[0102] The mapping data is subjected to lateral inhibition according to an APL neuron and a WTA mechanism, and is encoded into a sparse vector to obtain encoded image data.

[0103] Further, the orientation recognition unit 203 is specifically configured to:

[0104] The synaptic weights between the preset KC layer and the preset MBON layer are updated and calculated based on a preset update algorithm to obtain second synaptic weights;

[0105] The encoded image data is subjected to mapping-based orientation activity analysis based on the second synaptic weights to obtain an orientation recognition result.

[0106] The application further provides an image orientation recognition device, which includes a processor and a memory;

[0107] The memory is configured to store program code and transmit the program code to the processor;

[0108] The processor is configured to execute an image orientation recognition method in the above method embodiments according to instructions in the program code.

[0109] The application further provides a computer readable storage medium, which is configured to store program code, and the program code is configured to execute an image orientation recognition method in the above method embodiments.

[0110] In several embodiments provided in the application, it should be understood that the disclosed apparatus and method can be implemented in other manners. For example, the above-described apparatus embodiments are merely schematic. For example, the division of the units is only a logical function division. There can be another division manner for actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, apparatuses or units, and can be electrical, mechanical or in other forms.

[0111] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, i.e. may be located in one place, or may be distributed on multiple network units. Part or all of the units may be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0112] In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0113] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the essential part of the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for executing all or part of the steps of the method described in each embodiment of the present application by a computer device (which can be a personal computer, a server, or a network device, etc.). The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (English full name: Read-Only Memory, English abbreviation: ROM), a random access memory (English full name: Random Access Memory, English abbreviation: RAM), a magnetic disk or an optical disk, and various program code storage media.

[0114] The above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An image orientation recognition method characterized by, The method comprises the steps of: using a projection neural network to preprocess a target image to obtain standard image data, wherein the preprocessing comprises dimension reduction, noise reduction, normalization, cropping and gain control; encoding the standard image data according to APL neurons and a WTA mechanism through a preset KC layer to obtain encoded image data, wherein the projection neural network and the preset KC layer comprise first synaptic weights subject to random distribution, and the specific process is as follows: randomly generating the first synaptic weights in binary form between the projection neural network and the preset KC layer based on a Bernoulli distribution; performing linear mapping according to the first synaptic weights to obtain mapping data; performing lateral inhibition on the mapping data according to APL neurons and a WTA mechanism, and encoding the mapping data into a sparse vector to obtain encoded image data; inputting the encoded image data into a preset MBON layer for orientation analysis to obtain an orientation recognition result, wherein the preset KC layer and the preset MBON layer comprise second synaptic weights updated based on a preset update algorithm, and the specific process is as follows: updating and calculating the synaptic weights between the preset KC layer and the preset MBON layer based on the preset update algorithm to obtain the second synaptic weights; performing mapping-based orientation activity analysis on the encoded image data according to the second synaptic weights to obtain the orientation recognition result.

2. The image orientation recognition method according to claim 1, characterized by, The updating and calculating of the synaptic weights between the preset KC layer and the preset MBON layer based on the preset update algorithm to obtain the second synaptic weights comprises: the preset update algorithm is an update algorithm based on Hebbian rules; learning and updating the synaptic weights between the preset KC layer and the preset MBON layer based on the Hebbian rules to obtain the second synaptic weights in a convergent state.

3. The image orientation recognition method according to claim 1, characterized by, The updating and calculating of the synaptic weights between the preset KC layer and the preset MBON layer based on the preset update algorithm to obtain the second synaptic weights comprises: the preset update algorithm is a binary-based update algorithm; performing binary instant updating of the synaptic weights between the preset KC layer and the preset MBON layer based on the binary to obtain the second synaptic weights.

4. An image orientation recognition apparatus characterized by comprising: The method comprises the steps of: a preprocessing unit is configured to use a projection neural network to preprocess a target image to obtain standard image data, wherein the preprocessing comprises dimension reduction, noise reduction, normalization, cropping and gain control; an image encoding unit is configured to encode the standard image data according to APL neurons and a WTA mechanism through a preset KC layer to obtain encoded image data, wherein the projection neural network and the preset KC layer comprise first synaptic weights subject to random distribution, and the image encoding unit is specifically configured to: randomly generate the first synaptic weights in binary form between the projection neural network and the preset KC layer based on a Bernoulli distribution; perform linear mapping according to the first synaptic weights to obtain mapping data; perform lateral inhibition on the mapping data according to APL neurons and a WTA mechanism, and encode the mapping data into a sparse vector to obtain encoded image data; The orientation recognition unit is configured to input the encoded image data into a preset MBON layer for orientation analysis to obtain an orientation recognition result, the preset KC layer and the preset MBON layer include second synaptic weights updated based on a preset update algorithm, and the orientation recognition unit is specifically configured to: perform update calculation on the synaptic weights between the preset KC layer and the preset MBON layer based on a preset update algorithm to obtain second synaptic weights; perform mapping-based orientation activity analysis on the encoded image data according to the second synaptic weights to obtain an orientation recognition result.

5. An image orientation recognition apparatus characterized by comprising: The device comprises a processor and a memory; The memory is configured to store program code and transmit the program code to the processor; The processor is configured to execute the image orientation recognition method according to the instructions in the program code.

6. A computer-readable storage medium, characterized in that, The computer readable storage medium is configured to store program code, and the program code is configured to execute the image orientation recognition method. The computer readable storage medium is configured to store program code, and the program code is configured to execute the image orientation recognition method.

Images