Laser interference scene recognition method based on deep learning

Abstract

The invention discloses a laser interference scene recognition method based on deep learning, and relates to the field of laser interference recognition, and the method comprises the steps: inputting a laser interference scene image collected by a photoelectric imaging system into a preset laser interference scene recognition model, so as to obtain a corresponding laser interference state; the laser interference state is that the photoelectric imaging system is interfered by laser or not interfered by laser; wherein the preset laser interference scene recognition model is an optimal model obtained by training and comparing different deep learning networks by adopting a laser interference scene sample set. According to the invention, the laser interference scene can be intelligently, effectively and accurately identified.

Description

Technical Field

[0001] The present application relates to the field of laser interference identification, and in particular to a laser interference scene identification method based on deep learning. Background Art

[0002] As the core equipment of modern detection systems, optoelectronic imaging systems are widely used in medical imaging, image media, security management, high-resolution target reconnaissance and identification, optoelectronic precision guidance, fire control and targeting, flight assistance, and autonomous driving. However, direct light from natural light sources such as the sun, artificial light sources such as neon lights and street lamps, and reflected light from objects can cause serious optoelectronic interference to the image information generated by optoelectronic imaging systems.

[0003] Researchers have conducted research on identifying and eliminating incoherent light (sunlight or lighting) interference scenes, primarily using traditional image processing methods such as Gaussian filtering and histogram equalization. However, these methods rely heavily on optical hardware parameters, limiting their ability to express features based on prior knowledge. This makes it difficult to capture the characteristics of laser interference in complex scenarios without prior knowledge. In particular, these methods suffer from systemic flaws such as weak model generalization and poor scene adaptability when processing high-dimensional interference features. Summary of the Invention

[0004] The purpose of this application is to provide a laser interference scene recognition method based on deep learning, which can intelligently, effectively and accurately identify laser interference scenes.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] This application provides a laser jamming scene recognition method based on deep learning, including:

[0007] Obtain images of the laser interference scene collected by the optoelectronic imaging system;

[0008] Inputting the laser interference scene image into a preset laser interference scene recognition model to obtain a corresponding laser interference state; the laser interference state is that the photoelectric imaging system is interfered with by the laser or the photoelectric imaging system is not interfered with by the laser;

[0009] Among them, the preset laser interference scene recognition model is the optimal model obtained by training and comparing different deep learning networks using a laser interference scene sample set.

[0010] According to the specific embodiments provided in this application, the present application has the following technical effects: This application uses a preset laser interference scene recognition model, which is obtained by training and comparing different deep learning networks using a laser interference scene sample set. By applying deep learning networks to laser interference scene recognition, it fills a technical gap in this field and can effectively address the reliance on optical hardware parameters in existing technologies. In addition, the intelligent recognition method based on deep learning adopted by this application in laser interference scene recognition provides a precise decision-making basis for the subsequent laser interference suppression algorithm, thereby avoiding invalid suppression operations and ensuring the optimal allocation of system resources in critical scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0012] Figure 1 A flowchart of a deep learning-based laser interference scene recognition method provided in one embodiment of the present application.

[0013] Figure 2 Schematic diagram of the process of determining the preset laser interference scene recognition model.

[0014] Figure 3 Schematic diagram of laser interference with target objects at different powers.

[0015] Figure 4 This is a structural diagram of Resnet50.

[0016] Figure 5 A structural diagram of the laser jamming scene recognition model.

[0017] Figure 6 Another structural diagram of the laser interference scene recognition model.

[0018] Figure 7 Schematic diagram of the DCNv2 convolutional subunit.

[0019] Figure 8 Schematic diagram of the CBAM attention subunit.

[0020] Figure 9 Schematic diagram of the external attention subunit.

[0021] Figure 10 It is the loss accuracy curve of the deep learning model on the training set and validation set. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0023] This application can identify laser interference scenes and effectively determine whether the scene content of the optical imaging system is subject to laser interference, so as to facilitate the system to implement different suppression measures for laser interference scenes, and has wide application adaptability.

[0024] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0025] In an exemplary embodiment, Figure 1 As shown, a laser interference scene recognition method based on deep learning is provided, including the following steps 201 to 202.

[0026] Step 201: Acquire a laser interference scene image captured by an optoelectronic imaging system.

[0027] Step 202: Input the laser interference scene image into a preset laser interference scene recognition model to obtain a corresponding laser interference state; the laser interference state is that the photoelectric imaging system is interfered with by the laser or the photoelectric imaging system is not interfered with by the laser; wherein, the preset laser interference scene recognition model is an optimal model obtained by training and comparing different deep learning networks using a laser interference scene sample set.

[0028] Before executing the above steps 201 and 202, it is necessary to first determine the preset laser interference scene recognition model, such as Figure 2 As shown, in an application example, the process of determining the preset laser interference scene recognition model includes the following three parts.

[0029] (1) Create a laser interference image dataset.

[0030] The laser interference image dataset is a sample set of laser interference scenes. The data in this dataset mainly include scenes where the target object (such as a car) is interfered with by laser. The steps include the following (11)-(14).

[0031] (11) A photoelectric imaging system is used to collect multiple laser interference sample images; different laser interference sample images are subjected to laser interference of different powers.

[0032] In a specific practical application, an optoelectronic imaging system, laser, computer, target cart, and Sorebo beam expander can be used to collect sample images. The specific parameters of these instruments are shown in Table 1. The optoelectronic imaging system includes two beam splitters, four 25mm focal length lenses, three apertures, and three image plane detectors.

[0033] Based on the above-mentioned instruments and equipment, the sample image acquisition process includes: aiming the optical imaging system at the target object (i.e., the target car), with the distance between the target car and the image plane detector of the photoelectric imaging system being about 3m; the laser beam is sent to the target car via the Sorebo beam expander; fixing the position of the laser to the image plane detector, and by adjusting the laser power, collecting images of laser interference of different powers, the target scene light spot can be continuously expanded, such as Figure 3 As shown, by using computer control to collect pictures in real time, 4,800 laser interference pictures (corresponding to the laser interference state, the photoelectric imaging system is interfered with by the laser) and 4,800 non-laser interference pictures (corresponding to the laser interference state, the photoelectric imaging system is not interfered with by the laser) can be collected.

[0034] Table 1

[0035] Instruments and Equipment Specific parameters Photoelectric imaging system Detector resolution: 1440*1080 laser Wavelength: 532nm computer Model: LegionY7000PIRH8 Target car Size: 30cm*20cm*20cm Sorebo beam expander GBE10-A

[0036] (12) Determine the laser interference state corresponding to each of the laser interference sample images.

[0037] (13) Preprocessing operations are performed on all the laser interference sample images; the preprocessing operations include standardization processing, normalization processing and data enhancement processing.

[0038] The image resolution of the laser interference sample images obtained through the above step (11) is 1440*1080. In order to simplify the model training and evaluation in the subsequent steps, the collected images are standardized, that is, all images are uniformly adjusted to 224×224 pixels and normalized to eliminate the impact of different resolutions and ensure data consistency. In order to improve the generalization ability of the model and reduce overfitting, the data enhancement processing used includes random inversion, random cropping, center cropping, etc.

[0039] (14) Any laser interference sample image after preprocessing and the corresponding laser interference state constitutes a sample; multiple samples constitute a laser interference scene sample set.

[0040] (2) Construction of laser interference scene recognition model.

[0041] Before model training, this application built the following deep learning networks: improved ResNet50, ResNet50, VGG16 and AlexNet, etc.

[0042] Among them, ResNet50 effectively overcomes the vanishing gradient problem in deep learning classification models. The ResNet50 architecture is divided into four main parts, each called a stage, and each performs different functions. In the first stage, ResNet50 uses a 7×7 convolution kernel combined with a 3×3 maximum pooling operation to effectively reduce the input image size and provide a more compact data representation for subsequent processing. In the second stage, through a series of innovative residual structure designs, namely Conv2, Conv3, Conv4, and Conv5 residual blocks, the model can deeply explore the high-level features of the image. These highly abstract features are then fed into the fully connected layer of the third stage to complete the final classification task.

[0043] ResNet is mainly a residual network that solves the gradient problem through cross-layer connections. This structure allows the input to skip certain layers directly and be added to the processed output to form the concept of "residual". Assuming that the input image is x and the output is H(x), the output after convolution is a nonlinear function of F(x). The final output is H(x)=F(x)+x, which can be converted into a residual function F(x)=H(x)-x. This residual function is easier to optimize than F(x)=H(x). The structure of Resnet50 is as follows Figure 4 As shown in Figure 2, the ResNet residual network introduces a residual structure. By stacking multiple residual units, networks of varying depths and complexities are constructed to meet task requirements, achieving significant performance improvements in various image-related tasks.

[0044] VGG16 and AlexNet all use existing structures and will not be described in detail here.

[0045] In an exemplary application, the preset laser interference scene recognition model is an improved ResNet50, such as Figure 5 and Figure 6 As shown, it includes an input module, a convolutional pooling module, a ResNet residual module and a pooling classification output module that are set in sequence.

[0046] (21) The input module is used to receive the laser interference scene image; specifically, a 224×224 pixel laser interference image is used as the input of the network.

[0047] (22) The convolutional pooling module includes a first convolutional layer (conv), a batch normalization layer, a ReLU activation function layer, and a maximum pooling layer (Max Pool) arranged in sequence; specifically, the Batch Normalization layer is used to accelerate the convergence speed, and the ReLU activation function is used to increase nonlinearity, and sampling is performed through the maximum pooling layer.

[0048] (23) The ResNet residual module includes multiple residual sub-modules (ResNet50 Bottleneck), and the number of residual units contained in different residual sub-modules is different; the residual unit includes convolution processing, DCNv2 (Deformable Convolutional Networks version 2) convolution processing, CBAM attention mechanism and external attention mechanism; specifically, four residual sub-modules are set in this application, containing 3, 4, 6 and 3 residual units respectively.

[0049] The residual unit includes a second convolutional layer, a DCNv2 convolutional subunit, a third convolutional layer, a CBAM attention subunit and an external attention subunit arranged in sequence; the input end of the second convolutional layer serves as the input end of the residual unit; the input end of the second convolutional layer is connected to the output end of the external attention subunit through an addition operation, and the connection end serves as the output end of the residual unit.

[0050] Specifically, for each residual unit, the feature map input to the residual unit first passes through a 1×1 second convolutional layer to adjust the number of channels in the feature map; then passes through a 3×3 DCNv2 convolutional sub-unit for multi-scale feature extraction and spatial transformation; then passes through a 1×1 third convolutional layer to further adjust the number of channels or fuse features; then the CBAM attention sub-unit performs enhanced attention in the channel and spatial dimensions; and the external attention mechanism of the external attention sub-unit captures potential correlations between different samples. Finally, all processed features are fused through an addition operation to form the final output feature map. The residual unit structure within the second, third, and fourth residual sub-modules is the same as above.

[0051] In an exemplary embodiment, the DCNv2 convolution subunit includes an offset generation unit and a sampling convolution unit; the offset generation unit is used to: for a received feature map, calculate the sampling point offset of the convolution kernel in the horizontal and vertical directions of the feature map through a preset convolution operation; the sampling convolution unit is used to: based on the sampling point offset and the feature map, perform a bilinear interpolation operation to determine the sampling points of the convolution kernel on the feature map; and perform a convolution operation on the feature map according to the sampling points of the convolution kernel.

[0052] Specifically, if Figure 7As shown in the figure, DCNv2 is an improved version of the Deformable Convolutional Network (DCNv2). It introduces a deformable convolution layer to the traditional convolutional neural network to improve the network's adaptability to changes in the shape of objects in the image. The offset generation unit uses a specific convolution operation to accurately calculate the offset of the convolution kernel sampling points in the horizontal direction (x-axis) and the vertical direction (y-axis) of the received feature map. The sampling convolution unit uses this offset information (i.e., the sampling point offset) and the input feature map to perform bilinear interpolation to accurately locate the sampling points of the convolution kernel, thereby completing efficient and flexible convolution processing. Finally, the convolution operation is performed using these sampling points.

[0053] In another exemplary embodiment, the CBAM attention sub-unit includes a channel attention unit and a spatial attention unit; the feature map received by the channel attention unit and the feature map output by the channel attention unit are multiplied and input into the spatial attention unit; the feature map received by the spatial attention unit and the feature map output by the spatial attention unit are multiplied and input into the external attention sub-unit.

[0054] like Figure 8 As shown, the channel attention unit includes a first global maximum pooling layer, a second global average pooling layer, a shared fully connected layer, a second global maximum pooling layer, a third global average pooling layer and a first Sigmoid function layer; the input end of the first global maximum pooling layer and the input end of the second global average pooling layer are both connected to the output end of the third convolutional layer; the output end of the first global maximum pooling layer and the output end of the second global average pooling layer are both connected to the input end of the shared fully connected layer; the output end of the shared fully connected layer is respectively connected to the input end of the second global maximum pooling layer and the input end of the third global average pooling layer; the output end of the second global maximum pooling layer and the output end of the third global average pooling layer are connected to the first Sigmoid function layer after performing an addition operation; the first Sigmoid function layer is used to activate and output a feature map. The spatial attention unit includes average pooling and maximum pooling layers, a fourth convolutional layer and a second Sigmoid function layer arranged in sequence.

[0055] Specifically, in the channel attention unit, for each channel of the received feature map as input, global max pooling and global average pooling are first performed on the feature map. The maximum and average eigenvalues on each channel are calculated, resulting in two vectors containing the channel numbers, representing the global maximum feature and average feature of each channel, respectively. The global maximum feature and average feature are input into a shared fully connected layer. The fully connected layer is used to learn the attention weights for each channel in the feature map. The output is then subjected to global max pooling and global average pooling again. The global maximum feature vector and the average feature vector are then intersected to obtain the final attention weight vector. Finally, the feature map is activated using a sigmoid function.

[0056] In the spatial attention unit, the received feature map is subjected to average pooling and maximum pooling along the channel dimension. The features obtained after the average pooling and maximum pooling operations are concatenated along the channel dimension. After passing through a convolutional layer, a feature map with contextual information at different scales is obtained. Finally, the sigmoid function is used to activate the output feature map.

[0057] In another exemplary embodiment, Figure 9 As shown, the external attention mechanism ExternalAttention in the external attention sub-unit is a method of enhancing network capabilities by using two external memory matrices MemoryUnit that are independent of the feature map as input as key and value. Specifically, the feature map received by the external attention sub-unit is mapped to the Query matrix Q, the key matrix k (corresponding to an external memory matrix) and the value matrix v (corresponding to another external memory matrix) to model the similarity between the i-th pixel and the j-th row of the feature map, and the external memory matrix MemoryUnit is learnable and variable in size. At the same time, the external memory matrix MemoryUnit can also model the connection between different samples in the entire dataset as the training process progresses, and greatly reduces the computational and memory costs. Among them, Linear refers to linear function, and Matrix multiplication refers to matrix multiplication.

[0058] (24) The pooling classification output module is used to output the laser interference status; the pooling classification output module includes a first global average pooling layer (GlobalAveragepool), a fully connected layer (FC), a Softmax classification layer and an output layer arranged in sequence; specifically, the feature map is reduced to 1×1 through global average pooling, classified through a fully connected layer, and the final classification result is output through Softmax.

[0059] (3) Model training and model indicator evaluation.

[0060] In an exemplary application example, the process of determining the preset laser interference scene recognition model includes the following steps (31)-(36).

[0061] (31) Obtain a laser interference scene sample set; each sample in the laser interference scene sample set includes a laser interference sample image and a corresponding laser interference state; this step can directly call the data results obtained in the above (1). That is, obtain the laser interference sample image and the sample image not interfered with by the laser obtained by the image plane detector (such as a camera) in the photoelectric imaging system, and label them with the corresponding laser interference state.

[0062] (32) The laser interference scene sample set is divided into a training set, a validation set, and a test set; specifically, the 9,600 collected image data are divided into a training set, a validation set, and a test set according to categories, with a division ratio of 8:1:1.

[0063] (33) Using the training set to train different deep learning networks to obtain corresponding multiple optimized deep learning models.

[0064] Specifically, the hardware environment used in this application is Intel Core i5, 32GB of memory, and an NVIDIA GeForce RTX 3090 24GB graphics card. The programming language used is Python 3.9, Pytorch version 2.2.0, and CUDA version 12.1. During the model training process, the parameters of each model are adjusted to determine the optimal learning rate. All models use the SGD optimizer, momentum is selected as 0.9, and the learning rate is set to 0.01. The environment parameters used for model training and testing are shown in Table 2. The number of training rounds for all network models is 100 to ensure that the model can fully learn and optimize its performance.

[0065] Table 2

[0066] Experimental system Windows 10 CPU 3.00GHzIntel(R)Xeon(R)Gold6248R GPU NVIDIA GeForce RTX 3090 Memory 32GB Development Environment Python 3.9 Deep Learning Framework Pytorch 2.2.0

[0067] (34) Using the validation set, each of the optimized deep learning models is validated and adjusted to obtain corresponding multiple final deep learning models; Figure 10 As shown in the figure, the accuracy of the trained model on the validation set stabilized by epoch 45 and on the training set by epoch 50. The model also converged rapidly on the loss curve. These results demonstrate that this method effectively accelerates the model's convergence process, improving its stability and generalization capabilities on the validation set, laying a solid foundation for subsequent model applications.

[0068] (35) Using the test set, each of the final deep learning models is tested to obtain corresponding multiple test results. Specifically, in order to measure the effectiveness of the model, indicators such as accuracy, recall rate, and precision rate are used for evaluation, and the calculation formula is as follows:

[0069]

[0070] Among them, TP represents the number of samples correctly predicted as positive, TN represents the number of samples correctly predicted as negative, FP represents the number of samples incorrectly predicted as positive, and FN represents the number of samples incorrectly predicted as negative.

[0071] (36) Compare multiple test results and mark the final deep learning model with the best performance as the preset laser interference scene recognition model.

[0072] As shown in Table 3 below, the training results of the three models of improved ResNet50 (this application), ResNet50, VGG16 and AlexNet under the same parameter settings are listed, including the three indicators of accuracy, recall rate and precision rate. It can be seen that the improved ResNet50 model performed best in terms of accuracy, recall rate and precision rate, which are 0.9913, 0.9889 and 0.9761 respectively. This shows that the improved ResNet50 model has higher accuracy and stability in identifying laser interference scenes. In comparison, the performance of the ResNet50, VGG16 and AlexNet models is relatively poor, but it also verifies to a certain extent the application potential of the improved deep learning model in the field of laser interference scene recognition.

[0073] Table 3

[0074]

[0075]

[0076] In order to verify the effectiveness and generalization of the improved model, under the same experimental conditions and parameter settings, the proposed improvement strategies were respectively integrated into the ResNet50 model, and three groups were conducted to compare the accuracy, recall and precision of the models, as shown in Table 4, where √ represents the model with the addition of the module, and × represents the model without the addition of the module. From the experimental results, it can be seen that the addition of each module has improved the indicators of the model, proving the effectiveness of the improved model. In comparison, the improvement of the external attention mechanism EA has the greatest impact on the performance of the model. Therefore, the improved ResNet50 in this application fully combines the powerful feature extraction capabilities of the CBAM attention mechanism and the external attention mechanism EA, and uses DCNv2 for efficient convolution operations to achieve high-precision laser interference scene classification and recognition.

[0077] Table 4

[0078] Model CBAM EA DCNNv2 Accuracy Recall Precision 1 × × × 0.9408 0.9321 0.9414 2 √ × × 0.9515 0.9456 0.9448 3 √ √ × 0.9748 0.9669 0.9694 4 √ √ √ 0.9913 0.9889 0.9761

[0079] When applied in actual scenarios, in order to identify inaccurate images of laser interference scenes, we can increase the diversity of the data set by collecting and annotating more images of the scene, and then use data augmentation techniques (such as image cropping, image rotation, etc.) to simulate interference in different situations. Then, we integrate these new data into the training set and retrain the model to enhance the model's adaptability and generalization capabilities to complex scenarios. Finally, through iterative optimization and continuous monitoring, we ensure the robustness and accuracy of the model when facing new problems.

[0080] In summary, the laser interference scene recognition based on deep learning implemented in this application is an improved recognition model that uses the ResNet network model as the baseline network and combines the CBAM attention mechanism, external attention mechanism, etc. First, different laser interference scene pictures are collected, and at the same time, pictures of scenes without laser interference are collected; the collected pictures are used to train models of different classification and recognition algorithms, and the optimal classification and recognition algorithms are compared. The trained algorithm model is used to verify the actual effect, and relevant data sets are supplemented for scenes with classification errors. The model is retrained and optimized to increase the accuracy and generalization of the model classification and recognition. A residual learning architecture is introduced into the above-mentioned algorithm model to construct a multi-scale feature fusion mechanism while maintaining the stability of deep network training, effectively improving the representation ability of complex interference patterns.

[0081] This application proposes the feasibility of deep learning classification and recognition algorithms being applied to the field of laser interference. Secondly, it uses deep learning classification algorithms to identify whether the optoelectronic imaging system is interfered with by lasers. It compares several common classification and recognition algorithms, and improves the algorithms with better effects. Finally, it proposes a laser interference suppression scene recognition method based on a deep learning model, which can effectively identify whether the optoelectronic imaging system is interfered with by lasers, thereby preparing for subsequent laser interference suppression.

[0082] In addition, although there are other alternatives (such as improvements to traditional image processing methods, other deep learning networks such as other variants of convolutional neural networks (CNN) such as VGG, GoogLeNet, etc., and combinations of traditional image processing methods and deep learning methods) that can be tried for the identification of laser interference scenes, the laser interference scene recognition method based on the improved ResNet50 proposed in this application has significant advantages in feature extraction capabilities, deep network training stability, and adaptability to complex scenes. It can more effectively achieve high-precision laser interference scene classification and recognition, thereby better achieving the purpose of the invention.

[0083] Based on the same inventive concept, the embodiment of the present application also provides a laser interference scene recognition system based on deep learning, including:

[0084] The data acquisition subsystem is used to obtain images of the laser interference scene collected by the optoelectronic imaging system.

[0085] The laser interference recognition subsystem is used to: input the laser interference scene image into a preset laser interference scene recognition model to obtain a corresponding laser interference state; the laser interference state is that the photoelectric imaging system is interfered with by the laser or the photoelectric imaging system is not interfered with by the laser; wherein, the preset laser interference scene recognition model is an optimal model obtained by training and comparing different deep learning networks using a laser interference scene sample set.

[0086] Compared with the prior art, this application also has the following advantages:

[0087] (1) This application does not rely on prior knowledge but directly utilizes deep learning theory to extract features from images output by an optoelectronic imaging system and identify laser interference scenarios, filling a technological gap in this field and possessing significant innovation and practicality. Furthermore, the deep learning training network includes, but is not limited to, a residual network.

[0088] (2) This application uses the improved residual network (ResNet) to achieve high-precision classification and identification of laser interference scenes. This application innovatively adopts the residual network as the core architecture and improves it to identify laser interference scenes. The residual network combined with CABM and the external attention mechanism's powerful feature extraction capabilities can effectively learn the complex features in laser interference scenes. At the same time, combined with DCNv2 for efficient convolution operations, the stability of its deep network training also ensures the model's good performance in complex data environments. By using the improved ResNet50, key information in laser interference scenes can be automatically extracted without over-reliance on manual feature extraction, thereby better adapting to complex and changing interference scenes.

[0089] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0090] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A laser jamming scene recognition method based on deep learning, characterized in that: The method comprises: Obtain images of the laser interference scene collected by the optoelectronic imaging system; Inputting the laser interference scene image into a preset laser interference scene recognition model to obtain a corresponding laser interference state; the laser interference state is that the photoelectric imaging system is interfered with by the laser or the photoelectric imaging system is not interfered with by the laser; Among them, the preset laser interference scene recognition model is the optimal model obtained by training and comparing different deep learning networks using a laser interference scene sample set.

2. The laser jamming scene recognition method based on deep learning according to claim 1 is characterized in that: The preset laser interference scene recognition model includes an input module, a convolution pooling module, a ResNet residual module and a pooling classification output module which are arranged in sequence; The input module is used to receive the laser interference scene picture; The convolutional pooling module includes a first convolutional layer, a batch normalization layer, a ReLU activation function layer and a maximum pooling layer arranged in sequence; The ResNet residual module includes multiple residual submodules, and the number of residual units contained in different residual submodules is different; the residual unit includes convolution processing, DCNv2 convolution processing, CBAM attention mechanism and external attention mechanism; The pooling classification output module is used to output the laser interference status; the pooling classification output module includes a first global average pooling layer, a fully connected layer, a Softmax classification layer and an output layer arranged in sequence.

3. The laser jamming scene recognition method based on deep learning according to claim 2 is characterized in that: The residual unit includes a second convolutional layer, a DCNv2 convolutional subunit, a third convolutional layer, a CBAM attention subunit and an external attention subunit arranged in sequence; The input end of the second convolutional layer serves as the input end of the residual unit; The input end of the second convolutional layer is connected to the output end of the external attention sub-unit through an addition operation, and the connection end serves as the output end of the residual unit.

4. The laser jamming scene recognition method based on deep learning according to claim 3 is characterized in that: The DCNv2 convolution subunit includes an offset generation unit and a sampling convolution unit; The offset generating unit is used to: calculate the sampling point offsets of the convolution kernel in the horizontal direction and the vertical direction of the feature map through a preset convolution operation for the received feature map; The sampling convolution unit is used to: perform a bilinear interpolation operation based on the sampling point offset and the feature map to determine the sampling points of the convolution kernel on the feature map; and perform a convolution operation on the feature map according to the sampling points of the convolution kernel.

5. The laser jamming scene recognition method based on deep learning according to claim 3 is characterized in that: The CBAM attention subunit includes a channel attention part and a spatial attention part; The feature map received by the channel attention unit and the feature map output by the channel attention unit are multiplied and input into the spatial attention unit; The feature map received by the spatial attention unit and the feature map output by the spatial attention unit are multiplied and input into the external attention sub-unit.

6. The laser jamming scene recognition method based on deep learning according to claim 5 is characterized in that: The channel attention unit includes a first global maximum pooling layer, a second global average pooling layer, a shared fully connected layer, a second global maximum pooling layer, a third global average pooling layer and a first Sigmoid function layer; The input end of the first global maximum pooling layer and the input end of the second global average pooling layer are both connected to the output end of the third convolutional layer; The output end of the first global maximum pooling layer and the output end of the second global average pooling layer are both connected to the input end of the shared fully connected layer; The output end of the shared fully connected layer is connected to the input end of the second global maximum pooling layer and the input end of the third global average pooling layer respectively; The output end of the second global maximum pooling layer and the output end of the third global average pooling layer are added and then connected to the first Sigmoid function layer; the first Sigmoid function layer is used to activate and output a feature map; The spatial attention unit includes an average pooling layer and a maximum pooling layer, a fourth convolutional layer and a second Sigmoid function layer arranged in sequence.

7. The laser jamming scene recognition method based on deep learning according to claim 1, characterized in that: The process of determining the preset laser interference scene recognition model includes: Obtaining a laser interference scene sample set; each sample in the laser interference scene sample set includes a laser interference sample image and a corresponding laser interference state; Dividing the laser jamming scene sample set into a training set, a validation set, and a test set; Using the training set to train different deep learning networks to obtain corresponding multiple optimized deep learning models; Using the validation set, each of the optimized deep learning models is validated and adjusted to obtain a corresponding plurality of final deep learning models; Using the test set, testing each of the final deep learning models respectively to obtain corresponding multiple test results; By comparing multiple test results, the final deep learning model with the best performance is marked as the preset laser interference scene recognition model.

8. The laser jamming scene recognition method based on deep learning according to claim 1 is characterized in that: The process of acquiring the laser interference scene sample set includes: Use the photoelectric imaging system to collect multiple laser interference sample images; different laser interference sample images are subjected to laser interference of different powers; Determining a laser interference state corresponding to each of the laser interference sample images; Performing preprocessing operations on all the laser interference sample images; the preprocessing operations include standardization, normalization and data enhancement; Any laser interference sample image after preprocessing and the corresponding laser interference state constitutes a sample; multiple samples constitute a laser interference scene sample set.

Images