CN-120495842-B - Laser interference scene recognition method based on deep learning

CN120495842BCN 120495842 BCN120495842 BCN 120495842BCN-120495842-B

Abstract

The application discloses a laser interference scene recognition method based on deep learning, which relates to the field of laser interference recognition, and comprises the steps of inputting a laser interference scene picture acquired by a photoelectric imaging system 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 by laser or the photoelectric imaging system is 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. The application can intelligently, effectively and accurately identify the laser interference scene.

Inventors

YE QING
LIU AIBING
WU YUNLONG
XIN CHENG
SUN WEIBING
LIU YE

Assignees

中国人民解放军国防科技大学

Dates

Publication Date: 20260512
Application Date: 20250514

Claims (7)

1. A laser interference scene recognition method based on deep learning, the method comprising: acquiring a laser interference scene picture acquired by a photoelectric imaging system; Inputting the laser interference scene picture into a preset laser interference scene recognition model to obtain a corresponding laser interference state, wherein the laser interference state is that the photoelectric imaging system is interfered by laser or the photoelectric imaging system is not interfered by laser; 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; The preset laser interference scene identification model comprises an input module, a convolution pooling module, a ResNet residual module and a pooling classification output module which are sequentially arranged, wherein the ResNet residual module comprises a plurality of residual sub-modules, the number of residual units contained in the different residual sub-modules is different, the residual units comprise a second convolution layer, a DCNv convolution sub-unit, a third convolution layer, a CBAM attention sub-unit and an external attention sub-unit which are sequentially arranged, the input end of the second convolution layer is used as the input end of the residual unit, the input end of the second convolution layer is connected with the output end of the external attention sub-unit through addition operation, and the connection end is used as the output end of the residual unit; The DCNv < 2 > convolution subunit comprises an offset generation part and a sampling convolution part, the CBAM attention subunit comprises a channel attention part and a space attention part, and an external attention mechanism in the external attention subunit is used for enhancing the network capability by using two external Memory matrix Memory units which are independent of a feature map as input as keys and values.
2. The depth learning based laser interference scene recognition method of claim 1, wherein the input module is configured to receive the laser interference scene picture; The convolution pooling module comprises a first convolution layer, a batch normalization layer, a ReLU activation function layer and a maximum pooling layer which are sequentially arranged; The residual unit comprises convolution processing, DCNv's convolution processing, CBAM's attention mechanism and external attention mechanism; The pooling classification output module is used for outputting a laser interference state and comprises a first global average pooling layer, a full-connection layer, a Softmax classification layer and an output layer which are sequentially arranged.
3. The depth learning-based laser interference scene recognition method as set forth in claim 2, wherein the offset generating section is configured to calculate, for a received feature map, a sampling point offset of a convolution kernel in a horizontal direction and a vertical direction of the feature map by a preset convolution operation; The sampling convolution part is used for executing bilinear interpolation operation based on the sampling point offset and the feature map to determine the sampling point of a convolution kernel on the feature map, and carrying out convolution operation on the feature map according to the sampling point of the convolution kernel.
4. The method for recognizing a laser interference scene based on deep learning according to claim 2, wherein the feature map received by the channel attention portion and the feature map output by the channel attention portion are input to the spatial attention portion after multiplication processing is performed; The feature map received by the spatial attention unit and the feature map output by the spatial attention unit are input to an external attention subunit after multiplication processing is performed.
5. The deep learning-based laser interference scene recognition method of claim 4, wherein the channel attention portion comprises a first global max pooling layer, a second global average pooling layer, a shared full connection layer, a second global max 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 connected with the output end of the third convolution layer; The output end of the first global maximum pooling layer and the output end of the second global average pooling layer are connected with the input end of the shared full-connection layer; The output end of the shared full-connection layer is respectively connected with 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 with the first Sigmoid function layer after executing addition operation, and the first Sigmoid function layer is used for activating and outputting a feature map; The spatial attention part comprises an average pooling layer, a maximum pooling layer, a fourth convolution layer and a second Sigmoid function layer which are sequentially arranged.
6. The laser interference scene recognition method based on deep learning of claim 1, wherein the determining process of the preset laser interference scene recognition model comprises the following steps: acquiring a laser interference scene sample set, wherein each sample in the laser interference scene sample set comprises a laser interference sample picture and a corresponding laser interference state; Dividing the laser interference scene sample set into a training set, a verification set and a test set; training different deep learning networks by adopting the training set to obtain a plurality of corresponding optimized deep learning models; Respectively verifying and adjusting each optimized deep learning model by adopting the verification set to obtain a plurality of corresponding final deep learning models; Testing each final deep learning model by adopting the test set to obtain a plurality of corresponding test results; And comparing a plurality of test results, and marking the final deep learning model with optimal performance as a preset laser interference scene recognition model.
7. The method for identifying a laser interference scene based on deep learning according to claim 1, wherein the process of obtaining the laser interference scene sample set comprises: Acquiring a plurality of laser interference sample pictures by using a photoelectric imaging system, wherein different laser interference sample pictures are interfered by laser with different powers; Determining a laser interference state corresponding to each laser interference sample picture; Preprocessing all the laser interference sample pictures, wherein the preprocessing comprises standardization processing, normalization processing and data enhancement processing; the laser interference sample picture after any preprocessing operation and the corresponding laser interference state form one sample, and a plurality of samples form a laser interference scene sample set.

Description

Laser interference scene recognition method based on deep learning Technical Field The application relates to the field of laser interference identification, in particular to a laser interference scene identification method based on deep learning. Background The photoelectric imaging system is used as core equipment of a modern detection system and is widely applied to the fields of medical imaging, image media, security management, target high-resolution reconnaissance identification, photoelectric accurate guidance, fire control and aiming, flight assistance, automatic driving and the like. However, direct light from natural light sources such as the sun, direct light from artificial light sources such as neon lights, street lamps, and reflected light from objects can cause serious photoelectric interference with image information generated by a photoelectric imaging system. At present, students have developed research on a method for identifying and eliminating incoherent light (sunlight or lamplight) interference scenes, and the research mainly includes traditional image processing methods such as Gaussian filtering, histogram equalization and the like. However, the method excessively depends on optical hardware parameters, so that the feature expression capability is limited by priori knowledge, laser interference features under complex scenes without priori knowledge are difficult to capture, and systematic defects such as weak model generalization capability, poor scene adaptability and the like exist particularly when the high-dimensional interference features are processed. Disclosure of Invention The application aims to provide a laser interference scene recognition method based on deep learning, which can intelligently, effectively and accurately recognize a laser interference scene. In order to achieve the above object, the present application provides the following solutions: the application provides a laser interference scene recognition method based on deep learning, which comprises the following steps: acquiring a laser interference scene picture acquired by a photoelectric imaging system; Inputting the laser interference scene picture into a preset laser interference scene recognition model to obtain a corresponding laser interference state, wherein the laser interference state is that the photoelectric imaging system is interfered by laser or the photoelectric imaging system is not interfered by laser; 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 specific embodiment of the application, the method has the following technical effects that the method adopts the preset laser interference scene recognition model, the model is obtained by training and comparing different deep learning networks by adopting the laser interference scene sample set, and the problem of dependence on optical hardware parameters in the prior art can be effectively solved because the deep learning network is applied to the laser interference scene recognition, so that the technical gap in the field is filled. The intelligent recognition mode based on deep learning adopted in the aspect of laser interference scene recognition provides accurate decision basis for the subsequent laser interference suppression algorithm, so that invalid suppression operation is avoided, and optimal configuration of system resources in a key scene is ensured. Drawings In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art. Fig. 1 is a flow chart of a laser interference scene recognition method based on deep learning according to an embodiment of the application. Fig. 2 is a schematic diagram of a determining process of a preset laser interference scene recognition model. Fig. 3 is a schematic diagram of a laser interfering with a target at different powers. Fig. 4 is a schematic structural diagram of Resnet. Fig. 5 is a schematic structural diagram of a laser interference scene recognition model. Fig. 6 is another schematic structural diagram of a laser interference scene recognition model. Fig. 7 is a schematic diagram of DCNv2 convolution subunits. Fig. 8 is a schematic diagram of CBAM attention subunits. Fig. 9 is a schematic diagram of an external attention subunit. FIG. 10 is a graph of the loss accuracy of the deep learning model over the training set and validation set. Detailed Description The following description of the embodiments of the present application will be made clearly and compl