Nanosecond laser gating imaging slice recovery method and system

Abstract

The invention provides a nanosecond laser gating imaging slice recovery method and system, and relates to the technical field of laser imaging slice recovery. The method comprises the steps that a nanosecond laser gating imaging platform is built, and nanosecond laser gating imaging slices are collected; constructing a high-quality and low-quality slice data set based on the collected slices; constructing a GatedFormer-based slice recovery model, wherein the GatedFormer-based slice recovery model comprises a wavelet mixed attention module, a multi-scale gating feedforward neural network, a global high-frequency information enhancement module and an information injection module; in the training stage, training the model through a space-frequency joint loss function by using a high-quality and low-quality slice data set, taking a low-quality slice as input and taking a high-quality slice as a target; and in a reasoning stage, inputting a to-be-recovered slice into the trained model, and outputting the recovered slice. The method provided by the invention is highly matched with the data form of the nanosecond laser gating imaging slice, and meanwhile, the recovery quality and the calculation efficiency are considered, so that the real-time application requirement is met.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of laser imaging slice recovery, in particular to a nanosecond laser gating imaging slice recovery method and system. BACKGROUND

[0002] Laser gating imaging is an active light imaging technology. By precisely synchronizing the camera gating window with the laser pulse emission time, only the laser signal reflected by the target object is allowed to pass, thereby effectively suppressing the interference of environmental light (such as sunlight, artificial light) and scattering medium (such as smoke, rain and snow), significantly improving the image contrast and clarity, and can be widely applied to high-precision ranging, depth perception, autonomous driving, security monitoring and industrial detection scenes.

[0003] With the increasing demand for high-speed and strong scattering environment imaging, nanosecond laser gating imaging technology has emerged. By controlling the shutter on the nanosecond scale, it greatly improves the scattering suppression capability and imaging time resolution, thereby obtaining better distance measurement accuracy. The "image slice" in nanosecond laser gating imaging refers to a frame of image obtained within a very short time window (such as 500ps~3ns). Due to the very short time window, the signal-to-noise ratio of a single frame of image is usually low. Although traditional methods such as image intensifier can increase the gain of the signal, they easily introduce strong noise, which affects the image quality.

[0004] The prior art has the following defects in the processing of nanosecond laser gating imaging slices:

[0005] 1) Limitations of nanosecond laser gating imaging system: To achieve high-precision measurement, the system needs to use nanosecond short shutter time, resulting in insufficient photon collection. Therefore, image intensifiers are needed to increase the gain. However, the increase in gain will inevitably amplify the electronic noise, causing slice blurring and detail loss, which seriously affects the imaging quality.

[0006] 2) Inefficiency of multi-slice accumulation method: Traditional multi-slice accumulation generates high-quality images by averaging 16 consecutive noisy slices. However, this process requires the collection of a large number of slices, which is time-consuming and occupies a large amount of computing resources, and cannot adapt to real-time scenarios (such as dynamic environment perception in autonomous driving).

[0007] 3) Inapplicability of existing image restoration methods: Existing image restoration methods (such as methods based on convolutional neural networks or Transformers) are optimized for visible light images and do not consider the special nature of nanosecond laser gating slices: lack of color information, sparse high-frequency components, and strong coupling between noise and edges. Texture and edge details are crucial to image quality, but existing methods are difficult to enhance these features specifically, and it is difficult to effectively separate and restore high-frequency details, resulting in poor restoration results. SUMMARY

[0008] To solve the above problems, the purpose of the present application is to provide a nanosecond laser gating imaging slice recovery method and system. On the one hand, a slice recovery dataset composed of a single frame slice and a multi-frame slice cumulative one-to-one corresponding image pair is constructed, and the cumulative slice is selected as the true target for model training, while only the single frame slice is used as the model input, so as to achieve the recovery effect of multi-frame slice accumulation from single frame slice in the inference stage, and overcome the problems of poor single frame slice quality and difficult acquisition of cumulative slice. On the other hand, based on the GatedFormer network architecture, the wavelet mixed attention module is used to realize the separation and enhancement of high and low frequency information, the global high frequency information enhancement module and the information injection module are used to improve the global consistency of high frequency details, and the spatial and frequency domain optimization is balanced through the spatial and frequency domain loss. The method is highly adaptable to the data form of nanosecond laser gating imaging slice, and takes into account the recovery quality and computing efficiency, and meets the real-time application requirements.

[0009] To solve the above problems, the present application provides the following technical solutions:

[0010] On the one hand, a nanosecond laser gating imaging slice recovery method is provided, which comprises the following steps:

[0011] S1, a nanosecond laser gating imaging platform is built, and nanosecond laser gating imaging slices are collected;

[0012] S2, a high and low quality slice dataset is constructed based on the collected slices;

[0013] S3, a slice recovery model based on GatedFormer is constructed, including: a wavelet mixed attention module for enhancing slice high frequency features and low frequency features to realize texture recovery; a multi-scale gated feedforward neural network for multi-scale feature extraction and information flow control; a global high frequency information enhancement module for further enhancing high frequency features; and an information injection module for multi-scale feature fusion;

[0014] S4, in the training stage, the high and low quality slice dataset is used, the low quality slice is used as the input, and the high quality slice is used as the target, and the slice recovery model is trained through the spatial and frequency domain joint loss function;

[0015] S5, in the inference stage, the slice to be recovered is input into the trained slice recovery model, and the recovered slice is output.

[0016] Optionally, the nanosecond laser gating imaging platform in step S1 comprises a nanosecond laser gating camera, a multi-channel synchronous timing controller, a high-power laser emitter and a matching optical assembly.

[0017] Optionally, the high and low quality slice dataset in step S2 includes a data pair composed of a low quality slice and a high quality slice, wherein the low quality slice is a single frame slice data, and the high quality slice is 16-frame cumulative average slice data corresponding to the scene of the low quality slice, both of which realize pixel-level correspondence; and the high and low quality slice dataset is divided into a training set, a validation set and a test set according to a ratio of 7:1.5:1.5.

[0018] Optionally, the wavelet hybrid attention module in step S3 specifically includes:

[0019] The slice is decomposed into low frequency features and high frequency features by wavelet transform, the high frequency features extract horizontal / vertical textures through 1x3 and 3x1 depth separable convolution, and the low frequency features model global dependence through decoupling linear attention; then the enhanced high frequency features and low frequency features are fused through inverse wavelet transform to realize collaborative recovery of high frequency details and low frequency structures.

[0020] Optionally, the multi-scale gated feedforward neural network in step S3 specifically includes:

[0021] A double-branch structure is adopted to extract multi-scale local features through 1x3+3x1 and 1x5+5x1 depth separable convolution respectively, and the information flow is dynamically adjusted through the gating mechanism of element multiplication after GELU activation, realizing more efficient feature representation and information fusion.

[0022] Optionally, the global high frequency information enhancement module in step S3 specifically includes:

[0023] Fourier transform is performed on the different scale local features of the encoder, high frequency features are extracted and enhanced through convolution blocks, nonlinear transformation and sigmoid activation, and then inverse Fourier transform is performed to obtain enhanced global high frequency features, which make up for the problem of insufficient cross-layer information interaction.

[0024] Optionally, the information injection module in step S3 specifically includes:

[0025] The global high frequency features output by the global high frequency information enhancement module are adjusted to a size matching the decoder features, and are injected into the decoder through element multiplication to enhance the global consistency of high frequency details.

[0026] Optionally, the slice restoration model based on GatedFormer in step S3 further includes a gating block, which combines the outputs of the wavelet hybrid attention module and the multi-scale gated feedforward neural network through feature splitting to improve information utilization.

[0027] Optionally, the space-frequency joint loss function in step S4 combines spatial domain constraints and frequency domain constraints.

[0028] In another aspect, a nanosecond laser gating imaging slice recovery system is provided for implementing the method of any one of the above, the system comprising:

[0029] a data acquisition module for building a nanosecond laser gating imaging platform and acquiring nanosecond laser gating imaging slices;

[0030] a data set construction module for constructing a high-low quality slice data set based on the acquired slices;

[0031] a model construction module for constructing a slice recovery model based on GatedFormer, including: a wavelet mixed attention module for enhancing slice high-frequency features and low-frequency features to achieve texture recovery; a multi-scale gated feedforward neural network for multi-scale feature extraction and information flow control; a global high-frequency information enhancement module for further enhancing high-frequency features; and an information injection module for multi-scale feature fusion;

[0032] a model training module for training the slice recovery model using the high-low quality slice data set, with low-quality slices as input and high-quality slices as target, through a space-frequency joint loss function;

[0033] an inference module for inputting a slice to be recovered into the trained slice recovery model and outputting the recovered slice.

[0034] In another aspect, an electronic device is provided, the electronic device comprising:

[0035] a processor;

[0036] a memory having computer readable instructions stored thereon, the computer readable instructions being loaded and executed by the processor to implement the steps of the nanosecond laser gating imaging slice recovery method described above.

[0037] In another aspect, a computer readable storage medium is provided, the computer readable storage medium having program code stored therein, the program code being executable by a processor to perform the steps of the nanosecond laser gating imaging slice recovery method described above.

[0038] The technical solution provided by the present application has at least the following beneficial effects:

[0039] (1) The present application constructs a high-low quality slice data set specifically for nanosecond laser gating slice recovery, and combines hardware platforms, data sets and algorithm models to provide a complete solution for the practical application of nanosecond laser gating imaging technology.

[0040] (2) The present application can achieve high-quality slice recovery with only a single slice input, without the need for multiple slice cumulative input, and has short inference time, meeting the application requirements of efficiency and real-time.

[0041] (3) According to the data characteristics of the laser gated imaging slice, the application solves the problems of less high-frequency components and strong noise coupling, so that it can be applied to imaging scenes in complex environments, a slice recovery model based on GatedFormer is constructed, through the synergistic effect of multiple modules, the high and low frequency details are effectively separated and enhanced, and the recovered slice has higher quality, the peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) are significantly better than existing methods. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0043] Figure 1 is a flowchart of a nanosecond laser gated imaging slice recovery method provided by the embodiment of the application;

[0044] Figure 2 (a) and (b) are respectively a synchronization control principle schematic diagram of the nanosecond laser gated imaging technology provided by the embodiment of the application and a comparison schematic diagram with the existing cumulative imaging technology;

[0045] Figure 3 is an architecture schematic diagram of the GatedFormer slice recovery model provided by the embodiment of the application;

[0046] Figure 4 is a structural schematic diagram of a nanosecond laser gated imaging slice recovery system provided by the embodiment of the application;

[0047] Figure 5 is a structural schematic diagram of an electronic device provided by the embodiment of the application. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical scheme and advantages of the embodiments of the application more clear, the technical scheme of the embodiments of the application will be described clearly and completely below in combination with the drawings of the embodiments of the application. Obviously, the described embodiments are part of the embodiments of the application, not all the embodiments. Based on the described embodiments of the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0049] In the embodiments of the present application, the words such as "example", "for example" are used to represent an example, illustration or description. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the use of the word example is intended to present the concept in a specific manner. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be one of the two. In the embodiments of the present application, "image" and "picture" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent.

[0050] The embodiments of the present application provide a nanosecond laser gating imaging slice recovery method, which solves the problems in the prior art through a technical solution of "hardware platform + data set + algorithm model". Figure 1 As shown in the figure, the processing flow of the method includes the following steps:

[0051] S1, a nanosecond laser gating imaging platform is built, and nanosecond laser gating imaging slices are collected.

[0052] As an optional embodiment of the present application, the nanosecond laser gating imaging platform includes: a nanosecond laser gating camera (shortest shutter width 500 picoseconds, frame rate 98 frames per second), a multi-channel synchronous timing controller (precision 10 picoseconds), a high-power laser emitter (wavelength 860 nm, peak power 4000 W) and a supporting optical assembly (18-400mm zoom lens, beam expander, etc.), which are used to collect nanosecond laser gating imaging slices.

[0053] S2, a high and low quality slice data set is constructed based on the collected slices.

[0054] In the embodiments of the present application, based on the above-mentioned nanosecond laser gating imaging platform, 4275 slices containing scenes such as buildings, offices and squares are collected. The high and low quality slice data set is used to construct a high and low quality slice data set. The high and low quality slice data set includes a data pair composed of low quality slices and high quality slices, wherein the low quality slices are single frame slice data (for example, single frame slices after original collection and preprocessing), and the high quality slices are 16-frame cumulative average slice data corresponding to the scene of the low quality slices (i.e. the slice obtained by accumulating and averaging 16 frames of slices collected in the scene corresponding to the low quality slice), both of which realize pixel-level correspondence. Then, the high and low quality slice data set is divided into a training set, a validation set and a test set according to 7:1.5:1.5, providing data support for model training.

[0055] Figure 2Figures (a) and (b) illustrate the synchronization control principle of the nanosecond laser-gated imaging technology provided by an embodiment of the present invention, and a comparison diagram with existing cumulative imaging techniques. The present invention uses a single low-quality slice as input, with high-quality slices as the target, to train the model. During the inference phase, high-quality restoration can be achieved by simply passing a single slice through the model, eliminating the need for cumulative input from multiple slices. This improves efficiency and convenience.

[0056] S3. Build a slice recovery model based on GatedFormer. Figure 3 As shown in the figure, the slice restoration model based on GatedFormer includes: a wavelet-mixed attention module (WMA) for enhancing the high-frequency and low-frequency features of the slice to achieve texture restoration; a multi-scale gated feed-forward neural network (MSGFFN) for multi-scale feature extraction and information flow control; a global high-frequency information enhancement module (GHFIEM) for further enhancing high-frequency features; and an information injection module (IM) for multi-scale feature fusion.

[0057] Specifically, the wavelet hybrid attention module specifically includes:

[0058] Set input , apply Haar wavelet transform to obtain low-frequency characteristic components and high-frequency characteristic components , the high-frequency characteristic component H is divided into three sub-bands , formed by splicing along the channel dimension , the low-frequency characteristic component L is .

[0059] After wavelet transform is used to decompose the slice into low-frequency features (LF) and high-frequency features (HF), the high-frequency features are extracted through 1×3 and 3×1 depthwise separable convolution to extract horizontal / vertical textures, and then the features are activated through GELU to produce enhanced high-frequency features. ,In the low-frequency feature branch, the query, key and value matrices , sequence length is N, and feature dimension is C. H and W represent height and width, that is, the number of pixels of the feature map in the vertical and horizontal directions. The complexity of the standard Softmax attention calculation of pair similarity is O (N²), which is costly for high-resolution restoration. To solve this problem, the application proposes a decoupled linear attention (DLA) to reduce the complexity to O (N) by normalizing the sequence with Softmax. To offset the low-rank problem that may exist in DLA, a depth separable convolution is applied to V. Projection matrix The input LF is mapped to Q, K, V to obtain the enhanced low-frequency feature component LFR. To reconstruct the enhanced low-frequency feature component LFR and the high-frequency feature component HFR, first, HFR is divided into directional components along the channel, reshaped and spliced into . The inverse wavelet transform produces the fusion feature . The high-frequency enhancement branch reduces the dimension of to by 1x1 convolution, applies a depth separable convolution with 1x5 and 5x1 kernels, and obtains after sigmoid activation. After bilinear interpolation, is aligned with and multiplied element by element to produce the final output , realizing the collaborative recovery of high-frequency details and low-frequency structures.

[0060] The multi-scale gated feedforward neural network specifically includes:

[0061] A double-branch structure is adopted to extract multi-scale local features by 1x3+3x1 and 1x5+5x1 depth separable convolutions, respectively. After GELU activation, the information flow is dynamically adjusted by the gating mechanism of element multiplication, realizing more efficient feature representation and information fusion.

[0062] The global high-frequency information enhancement module specifically includes:

[0063] Given the encoder features , and and their fusion counterparts C1, C2, C3, the GHFIEM applies Fourier transform to extract multi-scale high-frequency feature information. After aligning the size by global average pooling and bilinear interpolation, the high-frequency features are spliced and reduced in dimension by 3x3 convolution to obtain the fusion feature , which is then enhanced by a convolution block and a nonlinear transformation and sigmoid activation, and then inverse Fourier transformed to obtain the enhanced global high-frequency feature , which compensates for the lack of cross-layer information interaction.

[0064] The information injection module specifically comprises:

[0065] The size is adjusted by average pooling or bilinear interpolation to match the fusion features. After sigmoid activation, element-wise multiplication to obtain the feature representation , and then the output is obtained by processing the feature representation through a convolutional block to enhance the feature representation. The global high-frequency feature output by the global high-frequency information enhancement module is adjusted to a size matching the decoder feature, and is injected into the decoder by element-wise multiplication to enhance the global consistency of high-frequency details.

[0066] Further, the slice recovery model based on GatedFormer further comprises a gated block (Gated Block) that combines the outputs of the wavelet hybrid attention module and the multi-scale gated feedforward neural network through feature splitting to improve information utilization.

[0067] S4, in the training phase, the high and low quality slice dataset is used, the low quality slice is used as input, and the high quality slice is used as target, and the slice recovery model is trained through a space-frequency joint loss function.

[0068] In the embodiment of the application, the model is trained by constructing a special high and low quality slice dataset, and the accumulated high quality slice is used as the real target for model training, while only a single frame of low quality slice is used as the model input, so that the recovery effect of multiple frames of slices accumulated from a single frame of slice can be obtained in the inference stage, and the problems of poor quality of single frame of slice and difficulty in collecting accumulated slices are overcome.

[0069] L1 / L2 distance and SSIM loss are widely used in image restoration, but mainly operate in the spatial domain, which is difficult to effectively separate high-frequency and low-frequency information, thereby hindering the simultaneous optimization of global structure and local details. In addition, they are sensitive to noise and outliers. Nanosecond laser gated slice has a longer wavelength, lower sensitivity to atmospheric scattering, and fewer high-frequency components, and is more susceptible to noise and outliers.

[0070] To solve these problems, a space-frequency joint loss function is proposed, which separates high-frequency and low-frequency features in the spatial and frequency domains, while preserving global structure and local details, and improves the robustness and generalization ability of the model. Given M training pairs (i=1,...,M), the high-frequency feature component of the slice is extracted by Fourier transform, which represents edges and details, and the low-frequency feature component representing smooth regions and structures. Here, F(Y) is the Fourier transform of slice Y, and |F(Y)| is its magnitude spectrum.

[0071]

[0072]

[0073]

[0074]

[0075]

[0076] where, Optimizing high-frequency components to reduce differences in details, Constraining the frequency spectrum to preserve the overall frequency distribution and minimize artifacts. Targeting low-frequency components to reduce differences in smooth regions and structures. Minimizing pixel-level differences, epsilon = 10 -3 To ensure stability.

[0077] The present application can balance high and low frequency component optimization by combining spatial domain constraints and frequency domain constraints, and improve the model's ability to recover details and structures.

[0078] S5, in the inference stage, input the slice to be recovered into the trained slice recovery model, and output the recovered slice.

[0079] The specific workflow of the method is as follows:

[0080] Training stage: using the constructed high and low quality slice dataset, taking the low quality slice as the input and the 16 frame cumulative high quality slice true value as the target, training the slice recovery model based on GatedFormer through the space-frequency joint loss function;

[0081] Inference stage: input the single low quality slice to be recovered into the trained model, output the recovered slice, and realize efficient single slice recovery.

[0082] Compared with the prior art, the present application constructs a high and low quality slice dataset specially used for nanosecond laser gating slice recovery, and combines hardware platform, dataset and algorithm model, to provide a complete solution for the practical application of nanosecond laser gating imaging technology. In the embodiment of the present application, only single slice input model is needed to realize high quality slice recovery, and the inference time is short (the inference time of the basic model of GatedFormer is only about 0.411 seconds), which can meet the application requirements of efficiency and real-time.

[0083] Meanwhile, based on the data characteristics of the laser gated imaging slice, the application constructs a slice recovery model based on GatedFormer to solve the problems of less high-frequency components and strong noise coupling, effectively separates and enhances high and low frequency details through multi-module cooperation, and the recovered slice has higher quality, and the peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) are significantly better than existing methods. Table 1 is a performance comparison of the method of the application and existing image enhancement methods on the nanosecond laser gated imaging slice recovery data set.

[0084] Table 1

[0085]

[0086] Correspondingly, the embodiment of the application also provides a nanosecond laser gated imaging slice recovery system, Figure 4 is a structural block diagram of a nanosecond laser gated imaging slice recovery system according to an exemplary embodiment. As Figure 4 indicated, the system comprises:

[0087] a data acquisition module for building a nanosecond laser gated imaging platform and acquiring nanosecond laser gated imaging slices;

[0088] a data set construction module for constructing high and low quality slice data sets based on the acquired slices;

[0089] a model construction module for constructing a slice recovery model based on GatedFormer, comprising: a wavelet mixed attention module for enhancing slice high frequency features and low frequency features to realize texture recovery; a multi-scale gated feedforward neural network for multi-scale feature extraction and information flow control; a global high frequency information enhancement module for further enhancing high frequency features; and an information injection module for multi-scale feature fusion;

[0090] a model training module for training the slice recovery model by using the high and low quality slice data sets, taking low quality slices as input and high quality slices as target, and using a space-frequency joint loss function;

[0091] an inference module for inputting the slice to be recovered into the trained slice recovery model and outputting the recovered slice.

[0092] For ease of illustration, Figure 4 only the main components of the system are shown. The system of the embodiment can be used to execute Figure 1 the technical solutions of the method embodiment, and the implementation principles and technical effects are similar, which will not be repeated here.

[0093] In exemplary embodiments, the application also provides an electronic device, which comprises:

[0094] a processor;

[0095] a memory, in which computer readable instructions are stored, and the computer readable instructions are loaded and executed by the processor to implement the steps of the nanosecond laser gating imaging slice recovery method.

[0096] Figure 5 is a structural schematic diagram of an electronic device provided by an embodiment of the present application, as shown in the figure, the electronic device 300 can include a first processor 3001 and a memory 3002. Optionally, the electronic device 300 can also include a transceiver 3003. Wherein, the first processor 3001 is connected with the memory 3002 and the transceiver 3003, such as can be connected through a communication bus. The memory 3002 stores computer readable instructions, and the computer readable instructions are executed by the first processor 3001 to implement the steps of the nanosecond laser gating imaging slice recovery method. Figure 5

[0097] In a specific implementation, as an embodiment, the first processor 3001 can include one or more CPUs, such as the CPU0 and CPU1 shown in the figure. Figure 5

[0098] In a specific implementation, as an embodiment, the electronic device 300 can also include multiple processors, such as the first processor 3001 and the second processor 3004 shown in the figure. Each of these processors can be a single-CPU or a multi-CPU. The processor here can refer to one or more devices, circuits, and / or processing cores for processing data (such as computer program instructions). Figure 5

[0099] Wherein, the memory 3002 is used to store software programs for executing the scheme of the present application, and is controlled by the first processor 3001 to execute, and the specific implementation manner can refer to the above-mentioned method embodiments, which will not be repeated here.

[0100] The transceiver 3003 is used to communicate with a network device or a terminal device.

[0101] Optionally, the transceiver 3003 can include a receiver and a transmitter. Wherein, the receiver is used to implement the receiving function, and the transmitter is used to implement the sending function.

[0102] Optionally, the transceiver 3003 can be integrated with the first processor 3001, or can exist independently and be coupled with the first processor 3001 through the interface circuit of the electronic device 300, and the embodiments of the present application do not make specific limitation here.

[0103] ​​​It should be noted that Figure 5 The structure of the electronic device 300 shown in the figures is not intended to limit the electronic device, and an actual electronic device can include more or fewer components than shown, or combine some components, or arrange different components. In addition, the technical effects of the electronic device 300 can refer to the technical effects of the above method embodiments, which will not be described here.

[0104] In an exemplary embodiment, the present application also provides a computer readable storage medium, the computer readable storage medium stores at least one instruction, the at least one instruction is loaded and executed by a processor to implement the steps of the above nanosecond laser gating imaging slice recovery method. For example, the computer readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0105] It should be noted that in this paper, the term "include", "contain" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or terminal device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or includes elements inherent to such process, method, article or terminal device. Without more limitations, the element defined by the statement "including a" does not exclude the presence of another identical element in the process, method, article or terminal device including the element.

[0106] In the specification, "one embodiment", "embodiment", "exemplary embodiment", "some embodiments" and the like indicate that the described embodiment can include a particular feature, structure or characteristic, but not necessarily every embodiment includes the particular feature, structure or characteristic. In addition, when a particular feature, structure or characteristic is described in connection with an embodiment, it should be within the knowledge of those skilled in the art to implement such feature, structure or characteristic in connection with other embodiments (whether or not explicitly described).

[0107] It should be understood that the term "and / or" herein is only a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent the three cases of A alone, A and B together, and B alone, where A and B can be singular or plural. In addition, the character " / " in this paper generally represents that the front and rear associated objects are a "or" relationship, but can also represent a "and / or" relationship, which can be understood according to the context before and after.

[0108] In the present application, "at least one" means one or more, and "multiple" means two or more. "At least one of the following" or the like means any combination of the items, including a single item or any combination of multiple items. For example, at least one of a, b, or c can mean a, b, c, a-b, a-c, b-c, or a-b-c, where a, b, and c can be single or multiple.

[0109] It should be understood that the size of the sequence number of each process described above in various embodiments of the present application does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0110] In several embodiments provided by the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0111] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the present embodiment.

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

[0113] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts of the present application that essentially contribute to the prior art 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 causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0114] The present application encompasses any substitutions, modifications, equivalent methods and solutions made on the essence and scope of the present application. In order to make the public have a thorough understanding of the present application, specific details are described in the following preferred embodiments of the present application, and the present application can be fully understood without the description of these details by those skilled in the art. In addition, in order to avoid unnecessary confusion to the essence of the present application, well-known methods, processes, flows, elements and circuits, etc. are not described in detail.

[0115] The above is only the preferred embodiment of the present application, and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A nanosecond laser-gated imaging slice recovery method, characterized in that, The method comprises the following steps: S1, a nanosecond laser gated imaging platform is built, and nanosecond laser gated imaging slices are collected; S2, a high-low quality slice dataset is constructed based on the collected slices; S3, a slice restoration model based on GatedFormer is constructed, including: a wavelet mixed attention module for enhancing slice high-frequency features and low-frequency features to realize texture restoration; a multi-scale gated feedforward neural network for multi-scale feature extraction and information flow control; a global high-frequency information enhancement module for further enhancing high-frequency features; and an information injection module for multi-scale feature fusion; S4, in the training stage, the high-low quality slice dataset is used, with low-quality slices as input and high-quality slices as target, and the slice restoration model is trained through a space-frequency joint loss function; S5, in the inference stage, the slice to be restored is input into the trained slice restoration model, and the restored slice is output.

2. The nanosecond laser-gated imaging slice recovery method of claim 1, wherein, The nanosecond laser gated imaging platform in step S1 comprises a nanosecond laser gated camera, a multi-channel synchronous timing controller, a high-power laser emitter and a matching optical assembly.

3. The nanosecond laser-gated imaging slice recovery method of claim 1, wherein, The high-low quality slice dataset in step S2 comprises a data pair composed of a low-quality slice and a high-quality slice, wherein the low-quality slice is a single-frame slice data, and the high-quality slice is a 16-frame cumulative average slice data corresponding to the scene of the low-quality slice, both realizing pixel-level correspondence; the high-low quality slice dataset is divided into a training set, a validation set and a test set according to a ratio of 7:1.5:1.

5.

4. The nanosecond laser-gated imaging slice recovery method of claim 1, wherein, The wavelet mixed attention module in step S3 specifically comprises: Wavelet transform is used to decompose the slice into low-frequency features and high-frequency features, the high-frequency features are extracted through 1x3 and 3x1 depth separable convolution to obtain horizontal / vertical texture, and the low-frequency features are modeled through decoupling linear attention to obtain global dependence; then, the enhanced high-frequency features and low-frequency features are fused through inverse wavelet transform to realize collaborative recovery of high-frequency details and low-frequency structures.

5. The nanosecond laser-gated imaging slice recovery method of claim 1, wherein, The multi-scale gated feedforward neural network in step S3 specifically comprises: A double-branch structure is adopted, and 1x3+3x1 and 1x5+5x1 depth separable convolution are used to extract multi-scale local features, respectively; after GELU activation, the information flow is dynamically adjusted through the gating mechanism of element multiplication to realize more efficient feature representation and information fusion.

6. The nanosecond laser-gated imaging slice recovery method of claim 1, wherein, The global high-frequency information enhancement module in step S3 specifically comprises: Fourier transform is performed on the local features of different scales of the encoder to extract high-frequency features, which are enhanced through a convolution block, a nonlinear transformation and a sigmoid activation, and then inverse Fourier transform is performed to obtain enhanced global high-frequency features, thereby making up for the insufficient cross-layer information interaction.

7. The nanosecond laser-gated imaging slice recovery method of claim 1, wherein, The information injection module in step S3 specifically comprises: The global high-frequency features output by the global high-frequency information enhancement module are adjusted to a size matching the features of the decoder, and are injected into the decoder through element multiplication to enhance the global consistency of high-frequency details.

8. The nanosecond laser-gated imaging slice recovery method of claim 1, wherein, The slice restoration model based on GatedFormer in step S3 further comprises a gating block, which combines the outputs of the wavelet mixed attention module and the multi-scale gated feedforward neural network through feature splitting to improve information utilization.

9. The nanosecond laser-gated imaging slice recovery method of claim 1, wherein, The space-frequency joint loss function in step S4 combines a spatial domain constraint and a frequency domain constraint.

10. A nanosecond laser-gated imaging slice recovery system for implementing the method of any one of claims 1 to 9, characterized in that, The system comprises: A data acquisition module is configured to build a nanosecond laser gating imaging platform and acquire nanosecond laser gating imaging slices. A data set construction module is configured to construct a high-low quality slice data set based on the acquired slices. A model construction module is configured to construct a slice restoration model based on GatedFormer, including: a wavelet hybrid attention module configured to enhance high-frequency features and low-frequency features of the slices to achieve texture restoration; a multi-scale gated feedforward neural network configured to perform multi-scale feature extraction and information flow control; a global high-frequency information enhancement module configured to further enhance high-frequency features; and an information injection module configured to perform multi-scale feature fusion. A model training module is configured to train the slice restoration model by using the high-low quality slice data set, taking low-quality slices as input and high-quality slices as target, and through a space-frequency joint loss function. An inference module is configured to input a slice to be restored into the trained slice restoration model and output a restored slice.

Images