Method and system for target detection based on optoelectronic hybrid computing

Abstract

The application provides a target detection method and system based on photoelectric hybrid computing. The method comprises the following steps: acquiring an image, and pre-processing the image; inputting the image into a light computing unit to perform linear change calculation, and outputting optical characteristics of the image; the light computing unit comprises a Mach-Zehnder interferometer array; inputting the optical characteristics into an electronic computing unit to perform nonlinear operation target detection, and outputting a target detection result of the image; the electronic computing unit comprises an electrical target detection model constructed based on a single multi-frame detector. In this way, efficient target detection can be realized through deep fusion of optical computing and electronic computing, the computing efficiency is optimized, the energy efficiency ratio is improved, the delay is reduced, and the system flexibility is enhanced.

Description

Technical Field

[0001] This invention relates to the field of neural network technology, and in particular to a target detection method and system based on optoelectronic hybrid computing. Background Technology

[0002] Object detection, as a core task of computer vision, has significant application value in fields such as autonomous driving, security monitoring, and industrial quality inspection. Deep learning-based object detection technology has evolved from two-stage detectors (R-CNN (Region-based Convolutional Neural Network) series) to single-stage detectors (such as YOLO (You Only Look Once) and SSD (Single Shot MultiBox Detector)), continuously improving detection accuracy and speed.

[0003] However, these electronically based convolutional neural networks (CNNs) have inherent drawbacks such as high computational complexity and low energy efficiency, posing significant challenges in applications with high real-time requirements. This is particularly pronounced in mobile devices and edge computing scenarios, where the power consumption and latency issues of traditional electronic chips become even more acute. As Moore's Law gradually becomes obsolete, the development bottlenecks of traditional electronic chips in terms of computing speed and energy consumption are becoming increasingly apparent, urgently requiring breakthroughs and innovations in new computing paradigms. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a target detection method and system based on optoelectronic hybrid computing, which uses photonic computing to accelerate convolution operations and combines GPU (Graphics Processing Unit) electronic computing to process nonlinear operations, so as to achieve efficient and low-power target detection.

[0005] In a first aspect, embodiments of the present invention provide a target detection method based on optoelectronic hybrid computing. The method includes: acquiring an image and preprocessing the image; inputting the image into an optical computing unit for linear transformation calculation and outputting the optical features of the image; wherein the optical computing unit includes a Mach-Zehnder interferometer array; inputting the optical features into an electronic computing unit for nonlinear target detection and outputting the target detection result of the image; wherein the electronic computing unit includes an electrical target detection model constructed based on a single-shot multi-frame detector.

[0006] In optional embodiments of this application, the above-described image preprocessing steps include: normalizing, resizing, convolution calculation, and reshaping the image.

[0007] In an optional embodiment of this application, the step of inputting the image into the optical computing unit for linear transformation calculation and outputting the optical features of the image includes: converting the image from an electrical signal to an optical signal; inputting the image into a Mach-Zehnder interferometer array for light field propagation and interference, and outputting the optical features of the image.

[0008] In an optional embodiment of this application, the steps of inputting an image into a Mach-Zehnder interferometer array for light field propagation and interference, and outputting the optical features of the image, include: determining the Mach-Zehnder interferometer matrix of the Mach-Zehnder interferometer array; cascading two Mach-Zehnder interferometer matrices to obtain a target matrix; and performing matrix multiplication between the image matrix and the target matrix to obtain a matrix of the optical features of the image.

[0009] In an optional embodiment of this application, the step of inputting optical features into an electronic computing unit for nonlinear operation of target detection and outputting the target detection result of the image includes: converting the optical features from light signals to electrical signals; inputting the optical features into an electrical target detection model and outputting multi-scale features in the electrical domain; and determining the bounding box and category of the image based on the multi-scale features as the target detection result.

[0010] In an optional embodiment of this application, the step of inputting an image into an electrical target detection model and outputting multi-scale features of the electrical domain includes: inputting optical features into the electrical target detection model and generating a feature map corresponding to the optical features as multi-scale features of the electrical domain through a convolutional layer; the step of determining the bounding box and category of the image based on the multi-scale features as the target detection result includes: setting anchor boxes with different length ratios at different positions of the feature map, and determining the offset and category probability of each anchor box as the bounding box and category of the image, respectively.

[0011] In optional embodiments of this application, the above method further includes: adjusting the parameters of the electrical target detection model through backpropagation based on the localization loss and classification loss.

[0012] In an optional embodiment of this application, the step of adjusting the parameters of the electric target detection model through backpropagation based on localization loss and classification loss includes: determining localization loss and classification loss based on the target detection results and ground truth labels of the image; wherein, the localization loss is calculated based on the positive sample anchor boxes of the image; the total loss is obtained by weighted calculation based on the localization loss and classification loss; and the parameters of the electric target detection model are adjusted based on the total loss.

[0013] Secondly, embodiments of the present invention also provide a target detection system based on photoelectric hybrid computing, which is used to execute the above-described target detection method based on photoelectric hybrid computing.

[0014] In an optional embodiment of this application, the above-mentioned target detection system based on optoelectronic hybrid computing includes: a graphics processing unit and an optical computing chip; the graphics processing unit is used to acquire an image and preprocess the image; the optical computing chip is used to input the image into the optical computing unit for linear transformation calculation and output the optical features of the image; the graphics processing unit is also used to input the optical features into the optical computing unit for nonlinear target detection and output the target detection result of the image.

[0015] The embodiments of the present invention bring the following beneficial effects: This invention provides a target detection method and system based on optoelectronic hybrid computing. The method involves acquiring an image, preprocessing the image, inputting the image into an optical computing unit for linear transformation calculations, and outputting the optical features of the image. The optical features are then input into an electronic computing unit for nonlinear target detection, and the target detection result is output. This approach achieves high-efficiency target detection through the deep integration of optical and electronic computing, optimizing computational efficiency, improving energy efficiency, reducing latency, and enhancing system flexibility.

[0016] Other features and advantages of this disclosure will be set forth in the following description, or some features and advantages may be inferred from the description or determined without doubt, or may be learned by practicing the techniques described above.

[0017] To make the above-mentioned objects, features and advantages of this disclosure more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 A flowchart of a target detection method based on photoelectric hybrid computing provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a system architecture for a target detection method based on photoelectric hybrid computing, provided in an embodiment of the present invention. Figure 3 A flowchart of another target detection method based on photoelectric hybrid computing provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an MZI provided in an embodiment of the present invention; Figure 5This is a schematic diagram of a target detection system based on photoelectric hybrid computing, provided as an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Currently, as Moore's Law gradually becomes obsolete, the development bottlenecks of traditional electronic chips in terms of computing speed and energy consumption are becoming increasingly prominent, and there is an urgent need to break through and innovate new computing paradigms.

[0022] Optical computing technology offers a revolutionary solution to this dilemma. This technology utilizes photons instead of electrons for information processing, achieving computational tasks by manipulating the unique properties of light, such as intensity, phase, and wavelength. The Mach-Zehnder interferometer (MZI) demonstrates the unique advantages of optical computing in matrix operations by implementing unitary matrix operations in the optical domain. First, optical computing possesses inherent parallelism, enabling high-speed matrix multiplication. Second, light signal propagation generates almost no heat, achieving an energy efficiency more than 100 times that of electronic computing. Third, optical computing is unaffected by electromagnetic interference, exhibiting higher stability. However, due to the relatively weak nonlinearity of light signals, pure optical computing faces significant limitations in implementing nonlinear activation functions such as ReLU (Rectified Linear Unit), severely restricting its application in deep neural networks.

[0023] Optoelectronic hybrid computing technology has emerged, combining the advantages of photons and electrons to form a complementary computing architecture. The main advantages of this technology are reflected in the following aspects: 1. Optimized computational efficiency: Linear operations (such as convolution and matrix multiplication) are handled by optical computing units, leveraging the parallelism of light to achieve ultra-high-speed computation; while nonlinear operations are handled by electronic computing units, fully utilizing their advantages in nonlinear processing. This division of labor significantly improves overall computational efficiency.

[0024] 2. Breakthrough in energy efficiency: The energy efficiency of the optoelectronic hybrid architecture far exceeds that of pure electronic computing, making it particularly suitable for power-sensitive applications.

[0025] 3. Reduced latency: The speed of light propagation is close to the physical limit, which can significantly reduce data transmission latency. In scenarios with high real-time requirements, such as target detection, optoelectronic hybrid systems can achieve microsecond-level response.

[0026] 4. Enhanced system flexibility: By dynamically configuring optoelectronic computing resources, the system can flexibly adapt to detection tasks of varying complexity, enabling on-demand allocation of computing resources.

[0027] Based on this, in response to the bottlenecks of existing target detection technologies in terms of computational efficiency, energy efficiency ratio, and real-time performance, this invention provides a target detection method and system based on optoelectronic hybrid computing, which mainly involves the fields of optical computing chips, target detection, and optoelectronic fusion computing. Specifically, it provides a method for implementing a target detection network GPU based on an optoelectronic hybrid computing architecture, which can combine the advantages of high parallelism and low power consumption of optical computing with the flexible nonlinear processing capabilities of electronic computing.

[0028] To facilitate understanding of this embodiment, a target detection method based on photoelectric hybrid computing disclosed in this embodiment of the invention will first be described in detail.

[0029] Example 1: This invention provides a target detection method based on photoelectric hybrid computing, see [link to relevant documentation]. Figure 1 The flowchart shown illustrates a target detection method based on photoelectric hybrid computing, which includes the following steps: Step S102: Acquire the image and preprocess it.

[0030] See also Figure 2 The diagram shows a system architecture of a target detection method based on photoelectric hybrid computing. In this embodiment, an image can be input first, and the image can be preprocessed.

[0031] Step S104: Input the image into the optical computing unit for linear transformation calculation and output the optical features of the image; wherein, the optical computing unit includes: a Mach-Zehnder interferometer array.

[0032] like Figure 2 As shown, in this embodiment, an optical computing unit including an MZI array can be set up to achieve high-speed parallel computing and output the optical features of the image.

[0033] This embodiment can perform optical computing through an optical computing unit. A programmable photonic computing unit can be constructed using an MZI array. By utilizing the parallel propagation characteristics of light, high-speed and low-power convolution operations can be achieved, significantly improving feature extraction efficiency.

[0034] Step S106: Input the optical features into the electronic computing unit to perform nonlinear operation on target detection, and output the target detection result of the image; wherein, the electronic computing unit includes: an electrical target detection model constructed based on a single-shot multi-frame detector.

[0035] like Figure 2 As shown, this embodiment may include an electronic computing unit comprising an electrical target detection model built on SSD. Optical features can be transmitted to the electronic computing unit via a photoelectric conversion interface, and the electronic computing unit performs target detection and outputs the target detection results of the image.

[0036] This embodiment can perform electronic computing through an electronic computing unit. Based on the electrical target detection model built by the single-shot multi-frame detector, it is responsible for handling nonlinear activation, bounding box regression and classification tasks to ensure detection accuracy.

[0037] This embodiment can achieve optoelectronic synergy optimization through optical computing units and electronic computing units: by using the GPU heterogeneous computing architecture, it can realize efficient synergy between optical and electronic computing, optimize data flow scheduling, reduce optoelectronic conversion overhead, thereby significantly improving inference speed and reducing energy consumption while maintaining high accuracy.

[0038] This invention provides a target detection method based on optoelectronic hybrid computing. The method involves acquiring an image, preprocessing the image, inputting the image into an optical computing unit for linear transformation calculations, and outputting the optical features of the image. The optical features are then input into an electronic computing unit for nonlinear target detection, and the target detection result is output. This approach achieves high-efficiency target detection through the deep integration of optical and electronic computing, optimizing computational efficiency, improving energy efficiency, reducing latency, and enhancing system flexibility.

[0039] Example 2: This invention provides another target detection method based on optoelectronic hybrid computing, implemented on the basis of the aforementioned embodiments, focusing on the specific methods of preprocessing, optical computing, and electronic computing. See also... Figure 3 The flowchart shown represents another target detection method based on photoelectric hybrid computing, which includes the following steps: Step S302: Acquire the image and perform preprocessing operations such as normalization, resizing, convolution calculation, and reshaping.

[0040] This embodiment can input an image and perform preprocessing, which may include operations such as normalization, resizing, convolution calculation, and reshaping.

[0041] In this embodiment, images and labels from the training dataset can be input, representing... ,in For the input image, For labels (category and bounding box). Represents a matrix. For any image... Preprocessing (including normalization, resizing, convolution calculation, and reshaping) is performed to obtain the processed image. First, the size was adjusted... Adjusted to Secondly, after convolution calculation, become The size of the convolution kernel is The process is as shown in formula (1): (1) Finally Perform reshaping operation This enables it to meet the input requirements for optical MZI interferometry calculations. The process is shown in formula (2): (2) Step S304: Input the image into the optical computing unit for linear transformation calculation and output the optical features of the image; wherein, the optical computing unit includes: a Mach-Zehnder interferometer array.

[0042] In some embodiments, the image can be converted from an electrical signal to an optical signal; the image is input into a Mach-Zehnder interferometer array for light field propagation and interference, and the optical characteristics of the image are output.

[0043] This embodiment can perform optical MZI interferometry calculations: by simulating light field propagation and interference using an MZI interferometer array, corresponding modulated optical features are generated. Specifically, the pre-processed image can first be converted from an electrical signal to an optical signal using a high-speed photoelectric conversion module.

[0044] Mach-Zehnder interference is an optical phenomenon based on the principle of two-beam interference, which can be found in [reference needed]. Figure 4 The diagram shows a schematic of an MZI structure. After the laser beam is emitted from the laser, it is split into two optical fibers of approximately the same length by a beam splitter formed by coupler 1. and After being modulated by different paths or phases, the light is recombined through coupler 2 to generate interference light, thus producing interference fringes. The structure of the MZI allows adjustment of the optical path difference to achieve interference of different phases. When the input optical signal passes through the MZI, it is split into two beams and then recombined in space to form a new optical signal. This new optical signal is considered to be a linear combination of the input optical signals, so it can realize optical matrix operations.

[0045] In some embodiments, the Mach-Zehnder interferometer matrix of the Mach-Zehnder interferometer array can be determined; the two Mach-Zehnder interferometer matrices can be cascaded to obtain the target matrix; and the matrix of the image can be multiplied by the target matrix to obtain the matrix of the optical features of the image.

[0046] In this embodiment, a single MZI transmission matrix (i.e., a Mach-Zehnder interferometer matrix) can be set. for: ;in, For beam splitter parameters, The phase difference is calculated by cascading individual MZI matrices into a series of matrices of size [missing information]. Target matrix Then... and Perform matrix multiplication to extract optical features The process is shown in formula (3): (3) Step S306: Input the optical features into the electronic computing unit to perform nonlinear operation on target detection, and output the target detection result of the image; wherein, the electronic computing unit includes: an electrical target detection model constructed based on a single-shot multi-frame detector.

[0047] In some embodiments, optical features can be converted from optical signals to electrical signals; the optical features are input into an electrical target detection model, and multi-scale features in the electrical domain are output; the bounding box and category of the image are determined based on the multi-scale features as the target detection result.

[0048] This embodiment can perform electrical SSD network processing: target detection is performed through the SSD network, multi-scale features in the electrical domain are extracted, and bounding boxes and categories are predicted. Specifically, optical features can first be converted from optical signals to electrical signals using a high-speed photoelectric conversion module.

[0049] This embodiment can transform the optical features after interference processing. The input is fed into the SSD network for training, including (multi-scale feature extraction, generating default boxes, and predicting output).

[0050] In some embodiments, optical features can be input into an electrical target detection model, and a feature map corresponding to the optical features can be generated through a convolutional layer as a multi-scale feature of the electrical domain. The step of determining the bounding box and category of the image based on the multi-scale features as the target detection result includes: setting anchor boxes with different length ratios at different positions of the feature map, and determining the offset and category probability of each anchor box as the bounding box and category of the image, respectively.

[0051] In this embodiment, multi-scale feature extraction can be performed: feature maps are generated through convolutional layers (VGG16). The process is shown in formula (5): (4) Secondly, at each feature map location Preset anchor frames with different length ratios (where m is the anchor frame index), and finally predict the offset for each anchor frame. and category probability The process is shown in formula (5). express The probe part in the network: (5) In some embodiments, the parameters of the electrical target detection model can also be adjusted through backpropagation based on the localization loss and classification loss.

[0052] In this embodiment, loss calculation and training can also be performed: combining localization loss (Smooth L1) and classification loss (cross-entropy loss), the parameters of the electric target detection model are optimized through backpropagation.

[0053] In some embodiments, localization loss and classification loss can be determined based on the target detection results and ground truth labels of the image; wherein, the localization loss is calculated based on the positive sample anchor boxes of the image; the total loss is obtained by weighted calculation based on the localization loss and classification loss; and the parameters of the electrical target detection model are adjusted based on the total loss.

[0054] This embodiment can predict the results. and real labels Loss calculation and backpropagation are performed for optimization. The total loss is the localization loss. and classification loss The weighted sum can be expressed as: ; Among them, the localization loss (Smooth L1) For: Only anchor boxes for positive samples (and ground truth boxes) The calculation process is shown in formula (6): (6) in .

[0055] Classification loss As shown in formula (7), For anchor box index, The total number of categories, For the true value, For predicted values: (7) In summary, this invention proposes a GPU-assisted acceleration method for a target detection system based on optoelectronic hybrid computing, achieving high-performance target detection through the deep integration of optical and electronic computing. The system employs a Mach-Zehnder interferometer (MZI) array to construct a programmable optical computing unit responsible for linear transformation calculations in the neural network's forward inference; simultaneously, it utilizes the GPU platform to implement nonlinear calculations, bounding box regression, and classification tasks for the SSD target detection network.

[0056] Furthermore, this embodiment provides a method that, while maintaining the aforementioned precision, also enables efficient conversion between optical and electrical signals via a high-speed photoelectric conversion module, and establishes a unified task scheduler to achieve pipelined parallelism of optical and electronic computing. This scheme maintains detection accuracy (mAP) comparable to pure electronic computing on datasets while significantly reducing inference time. It can be widely applied in fields such as intelligent driving, industrial inspection, and intelligent security, meeting the low latency and high energy efficiency requirements of real-time target detection.

[0057] Example 3: Corresponding to the above method embodiments, this invention provides a target detection system based on photoelectric hybrid computing, which is implemented on the basis of the foregoing embodiments. The target detection system based on photoelectric hybrid computing is used to execute the target detection method based on photoelectric hybrid computing provided in the foregoing embodiments.

[0058] See Figure 5 The diagram shows a target detection system based on optoelectronic hybrid computing. The system includes a graphics processing unit and an optical computing chip. The graphics processing unit is used to acquire images and preprocess them. The optical computing chip is used to input the images into the optical computing unit for linear transformation calculations and output the optical features of the images. The graphics processing unit is also used to input the optical features into the optical computing unit for nonlinear target detection and output the target detection results of the images.

[0059] like Figure 5 As shown, the target detection system based on optoelectronic hybrid computing may further include: a high-speed optoelectronic conversion module, used to achieve efficient conversion between the electrical signals of the GPU and the optical signals of the optical computing chip.

[0060] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the target detection system based on photoelectric hybrid computing described above can be referred to the corresponding process in the aforementioned embodiments of the target detection method based on photoelectric hybrid computing, and will not be repeated here.

[0061] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0062] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the referred element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0063] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A target detection method based on photoelectric hybrid computing, characterized in that, The method includes: Acquire an image and preprocess the image; The image is input into an optical computing unit for linear transformation calculation, and the optical features of the image are output; wherein, the optical computing unit includes: a Mach-Zehnder interferometer array; The optical features are input into an electronic computing unit for nonlinear target detection, and the target detection result of the image is output; wherein, the electronic computing unit includes: an electrical target detection model constructed based on a single-shot multi-frame detector.

2. The method according to claim 1, characterized in that, The steps for preprocessing the image include: The image undergoes preprocessing operations including normalization, resizing, convolution calculation, and reshaping.

3. The method according to claim 1, characterized in that, The step of inputting the image into a light computing unit for linear transformation calculation and outputting the optical features of the image includes: The image is converted from an electrical signal to an optical signal; The image is input into the Mach-Zehnder interferometer array for light field propagation and interference, and the optical characteristics of the image are output.

4. The method according to claim 1, characterized in that, The steps of inputting the image into the Mach-Zehnder interferometer array for light field propagation and interference, and outputting the optical features of the image, include: Determine the Mach-Zehnder interferometer matrix of the Mach-Zehnder interferometer array; The target matrix is ​​obtained by cascading the two Mach-Zehnder interferometer matrices. The matrix of the image is multiplied by the matrix of the target to obtain the matrix of the optical features of the image.

5. The method according to claim 1, characterized in that, The steps of inputting the optical features into an electronic computing unit for nonlinear target detection and outputting the target detection result of the image include: The optical features are converted from optical signals to electrical signals; The optical features are input into the electrical target detection model, and the multi-scale features in the electrical domain are output. The bounding box and category of the image are determined based on the multi-scale features as the object detection result.

6. The method according to claim 5, characterized in that, The steps of inputting the image into the electrical target detection model and outputting multi-scale features in the electrical domain include: The optical features are input into the electrical target detection model, and the feature map corresponding to the optical features is generated by the convolutional layer as a multi-scale feature of the electrical domain. The step of determining the bounding box and category of the image based on the multi-scale features as the object detection result includes: Anchor boxes with different length ratios are set at different positions in the feature map, and the offset and class probability of each anchor box are determined as the bounding box and class of the image, respectively.

7. The method according to claim 5, characterized in that, The method further includes: The parameters of the electrical target detection model are adjusted through backpropagation based on localization loss and classification loss.

8. The method according to claim 7, characterized in that, The steps of adjusting the parameters of the electrical target detection model through backpropagation based on localization loss and classification loss include: The localization loss and classification loss are determined based on the target detection results and ground truth labels of the image; wherein, the localization loss is calculated based on the positive sample anchor boxes of the image; The total loss is obtained by weighting the localization loss and classification loss. The parameters of the electrical target detection model are adjusted based on the total loss.

9. A target detection system based on photoelectric hybrid computing, characterized in that, The target detection system based on photoelectric hybrid computing is used to execute the target detection method based on photoelectric hybrid computing as described in any one of claims 1-8.

10. The target detection system based on photoelectric hybrid computing according to claim 9, characterized in that, The target detection system based on optoelectronic hybrid computing includes: a graphics processing unit and an optical computing chip; The graphics processing unit is used to acquire images and preprocess the images; The optical computing chip is used to input the image into the optical computing unit for linear transformation calculation and output the optical features of the image; The graphics processing unit is also used to input the optical features into the electronic computing unit for nonlinear target detection and output the target detection result of the image.

Images