Focusing optimization method, device and equipment based on quantum deep learning

Abstract

The invention provides a focusing optimization method, device and equipment based on quantum deep learning. The method comprises the following steps: acquiring a continuous shooting image sequence of a target object; the method comprises the following steps: shooting a target object by using photographic equipment to obtain a continuous shooting image sequence; quantum image features corresponding to the continuous shooting image sequence are extracted; wherein the quantum image features comprise quantum bits, quantum dynamic trajectories and quantum depth-of-field distribution; inputting the quantum image features into a pre-trained quantum neural network to obtain a focus prediction result of the target object; and adjusting the focal length of the photographic equipment according to the focus prediction result. Through the above method, the focusing speed and precision are improved, so as to meet the shooting requirements in ultra-high-speed photography and complex scenes.

Description

Technical Field

[0001] The present invention relates to the field of photography technology, and in particular to a focusing optimization method, device and equipment based on quantum deep learning. Background Art

[0002] In the field of optical imaging, autofocus technology, as a core function to improve image clarity and shooting efficiency, has evolved from mechanical to electronic, and from single function to intelligent since its inception. With the development of electronic technology and algorithms, phase detection autofocus and contrast detection autofocus have become mainstream technical paths, significantly improving focusing speed and accuracy. At this stage, autofocus technology has been widely used in photography, videography, security monitoring, smart terminals and other fields. However, current autofocus technology still faces the following core challenges: Classical computing speed bottleneck: Existing autofocus algorithms are mostly based on traditional computing architectures. When processing ultra-high-speed continuous shooting (such as more than 20 frames per second) or complex scenes (such as low illumination, high dynamic range, and multi-target interference), they are limited by classical computing speeds and it is difficult to achieve millisecond-level focus prediction and dynamic adjustment. Summary of the Invention

[0003] The main technical problem solved by the present invention is to improve the speed and accuracy of focusing to meet the needs of ultra-high-speed photography and shooting in complex scenes.

[0004] According to the first aspect, an embodiment provides a focusing optimization method based on quantum deep learning, comprising:

[0005] Acquire a sequence of continuous images of a target object; wherein the target object is photographed using a photographic device to acquire the sequence of continuous images;

[0006] Extracting quantum image features corresponding to the continuous image sequence; wherein the quantum image features include quantum bits, quantum dynamic trajectories, and quantum depth of field distribution;

[0007] Inputting the quantum image features into a pre-trained quantum neural network to obtain a focus prediction result of the target object;

[0008] The focal length of the photographic device is adjusted according to the focus prediction result.

[0009] In some embodiments, the pre-trained quantum neural network is trained in the following manner:

[0010] Acquire a training data set; wherein the training data set includes training image sequences and corresponding focus annotations under different environmental conditions, wherein the environmental conditions include the motion speed of the training object, the scene complexity of the scene in which the training object is located, and the lighting conditions of the training object;

[0011] Extracting a training feature set corresponding to the training data set, and establishing a mapping relationship between training image features in the training feature set and their corresponding focus annotations using a preset quantum variational algorithm; wherein the training image features include training quantum bits, training quantum dynamic trajectories, and training quantum depth of field distributions;

[0012] The quantum neural network to be trained is trained based on the mapping relationship to obtain a trained quantum neural network.

[0013] In some embodiments, the step of training the quantum neural network to be trained based on the mapping relationship to obtain a trained quantum neural network includes:

[0014] Inputting the training image features in the mapping relationship into the quantum neural network to be trained to obtain a focus prediction result;

[0015] Evaluating the focus prediction result based on the focus annotation in the mapping relationship and a preset evaluation index to obtain an evaluation result;

[0016] The model parameters of the quantum neural network to be trained are optimized according to the evaluation results until the quantum neural network to be trained reaches convergence or a preset training round, thereby obtaining a trained quantum neural network.

[0017] In some embodiments, inputting the training image features in the mapping relationship into the quantum neural network to be trained to obtain a focus prediction result includes:

[0018] Assign corresponding weights to the training quantum bits, the training quantum dynamic trajectories, and the training quantum depth of field distribution, respectively;

[0019] Performing a weighted summation on the training quantum bits, the training quantum dynamic trajectory, and the training quantum depth of field distribution based on a preset quantum attention mechanism and assigned weights to obtain a comprehensive feature representation;

[0020] The comprehensive feature representation is decoded into a corresponding focus prediction result.

[0021] In some embodiments, before extracting the quantum image features corresponding to the continuous image sequence, the focus optimization method further includes:

[0022] Identifying the scene category of the target object based on a preset quantum classification algorithm and the continuous image sequence; wherein the scene category includes low illumination scene, high dynamic range scene, and multi-target interference scene;

[0023] The preset quantum feature extraction strategy is adjusted according to the scene category, and the adjusted quantum feature extraction strategy is used to extract quantum image features corresponding to the continuous image sequence.

[0024] In some embodiments, the quantum bits are obtained by converting the image pixels and edge information of the continuous image sequence through quantum state encoding, the quantum dynamic trajectory is obtained by using a preset quantum optical flow analysis algorithm to predict the dynamic change trend of the target object in the continuous image sequence, and the quantum depth of field distribution is the quantized distribution of the depth of the scene in which the target object is located.

[0025] In some embodiments, the photographic device includes a single camera or multiple cameras; and obtaining a sequence of continuous images of the target object includes:

[0026] photographing the target object using a photographic device including a single camera to obtain the continuous image sequence; or;

[0027] A photographic device including multiple cameras is used to obtain a sequence of continuous images of the target object at different viewing angles.

[0028] According to the second aspect, an embodiment provides a focusing optimization device based on quantum deep learning, comprising:

[0029] An image data acquisition module is used to acquire a sequence of continuous images of a target object; wherein the sequence of continuous images is acquired by photographing the target object using a photographic device;

[0030] An image feature extraction module, configured to extract quantum image features corresponding to the continuous image sequence; wherein the quantum image features include quantum bits, quantum dynamic trajectories, and quantum depth of field distribution;

[0031] An object focus prediction module, configured to input the quantum image features into a pre-trained quantum neural network to obtain a focus prediction result of the target object;

[0032] The device focal length adjustment module is used to adjust the focal length of the photographic device according to the focus prediction result.

[0033] According to the third aspect, an embodiment provides a focusing optimization device based on quantum deep learning, including:

[0034] Memory, used to store programs;

[0035] A processor is used to implement the focusing optimization method by executing the program stored in the memory.

[0036] According to a fourth aspect, an embodiment provides a computer program product, comprising a computer program and / or instructions, which implement the focusing optimization method when executed by a processor.

[0037] The quantum deep learning-based focus optimization method, apparatus, device, and computer program product described in the above embodiments extract quantum image features from a sequence of continuous images of the target object and, in combination with a quantum neural network, predict the focus of the target object in real time. Leveraging the ultrafast reasoning capabilities of quantum computing, they break through the limits of classical computing, significantly improving focusing speed and accuracy, making them suitable for ultra-high-speed photography and complex scene capture. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of a focusing optimization method based on quantum deep learning according to an embodiment of the present application;

[0039] Figure 2 A flowchart of a method for training a pre-trained quantum neural network according to one embodiment;

[0040] Figure 3 A flowchart of an embodiment of training a quantum neural network to be trained based on a mapping relationship to obtain a trained quantum neural network;

[0041] Figure 4 A flowchart of an embodiment of inputting training image features in a mapping relationship into a quantum neural network to be trained to obtain a focus prediction result;

[0042] Figure 5 A flowchart of a method for extracting quantum image features corresponding to a continuous image sequence according to an embodiment;

[0043] Figure 6 Schematic diagram of the structure of a focusing optimization device based on quantum deep learning in one embodiment. DETAILED DESCRIPTION

[0044] The present invention will be further described in detail below by means of specific embodiments in conjunction with the accompanying drawings. Similar elements in different embodiments are numbered with associated similar elements. In the following embodiments, many detailed descriptions are provided to enable the present application to be better understood. However, those skilled in the art will readily appreciate that some of the features may be omitted in different circumstances, or may be replaced by other elements, materials, or methods. In some cases, some operations related to the present application are not shown or described in the specification. This is to avoid the core portion of the present application being overwhelmed by excessive descriptions, and for those skilled in the art, it is not necessary to describe these related operations in detail. They will fully understand the related operations based on the description in the specification and the general technical knowledge in the art.

[0045] In addition, the features, operations, or characteristics described in the specification may be combined in any appropriate manner to form various embodiments. Furthermore, the steps or actions in the method description may be reordered or adjusted in a manner readily apparent to those skilled in the art. Therefore, the various sequences in the specification and drawings are provided solely for the purpose of clearly describing a particular embodiment and are not intended to be mandatory, unless otherwise specified.

[0046] The serial numbers assigned to components herein, such as "first," "second," etc., are used solely to distinguish the objects being described and do not convey any sequential or technical meaning. References to "connection" and "coupling" herein, unless otherwise specified, include both direct and indirect connections (couplings).

[0047] Existing autofocus technologies are limited by classical computing speeds, making it difficult to achieve millisecond-level focus prediction in ultra-high-speed continuous shooting or complex scenes. For example, in live sports events, when athletes are in high-speed motion, the camera must switch focus within milliseconds. Traditional algorithms suffer from computational delays, resulting in blurred images.

[0048] To improve the speed and accuracy of focusing to meet the needs of ultra-high-speed photography and shooting in complex scenes, an embodiment of the present application provides a focusing optimization method based on quantum deep learning. In this focusing optimization method, a sequence of continuous images of a target object is obtained; the target object is photographed using a photographic device to obtain the continuous image sequence; quantum image features corresponding to the continuous image sequence are extracted; the quantum image features include quantum bits, quantum dynamic trajectories, and quantum depth of field distribution; the quantum image features are input into a pre-trained quantum neural network to obtain a focus prediction result of the target object; and the focal length of the photographic device is adjusted based on the focus prediction result.

[0049] The following describes the focusing optimization method based on quantum deep learning provided by the embodiments of the present application in conjunction with the accompanying drawings.

[0050] Figure 1 A flow chart of a focusing optimization method based on quantum deep learning provided by an embodiment of the present application is shown, which is described in detail below:

[0051] Step S10: Acquire a sequence of continuous images of the target object.

[0052] Specifically, a photographic device is used to photograph a target object to obtain a continuous image sequence, wherein the continuous image sequence has a preset frame rate.

[0053] For example, the photographic equipment can be a quantum camera, an ultra-high-speed imaging device, or a scientific photography device. In the transient shooting scenario of ultra-high-speed scientific photography, a continuous image sequence of high-speed particle motion is captured in a laboratory; in the dynamic target tracking scenario of quantum imaging equipment, a multi-camera quantum system is used to capture a continuous image sequence of a high-speed moving target (such as an aircraft); in microscopic imaging in complex environments, a continuous image sequence of the target object in a microscopic photography task is captured.

[0054] In the embodiments of this application, the preset frame rate determines the time interval between two adjacent frames in a continuous image sequence. A higher frame rate can record the target object's motion state with extremely small time intervals, thereby accurately depicting the target object's motion trajectory. By acquiring a continuous image sequence of the target object, detailed image information of the target object is recorded, which serves as the data basis for subsequent focus prediction.

[0055] Step S20: extracting quantum image features corresponding to the continuous image sequence.

[0056] Specifically, quantum image features include quantum bits, quantum dynamic trajectories, and quantum depth of field distribution. Quantum state encoding is used to convert image pixels and edge information from a continuous image sequence into quantum bits. Quantum optical flow analysis is used to estimate the dynamic changes of the target object in the continuous image sequence to obtain the quantum dynamic trajectories. A quantum neural network is used to evaluate the quantized distribution of the depth of the scene in which the target object is located to obtain the quantum depth of field distribution.

[0057] In the embodiments of the present application, quantum bits are the basis of quantum computing and information processing. Quantum bits can be used to more efficiently represent the complex information of a continuous image sequence, such as texture, pixels, edges, etc., providing richer information for subsequent image analysis and processing. Quantum dynamic trajectories can accurately describe the motion changes of the target object at the quantum level. By extracting quantum dynamic trajectories, the motion characteristics of the target object can be more comprehensively understood, providing a more accurate basis for motion analysis and focus prediction. Depth of field is an important indicator of image clarity, and the quantum depth of field distribution takes into account the influence of the quantum characteristics of photons on the depth of field. Therefore, the quantum depth of field distribution can more realistically reflect the clarity of different areas in the continuous image sequence, which helps to improve the quality and readability of the continuous image sequence.

[0058] Step S30: Inputting the quantum image features into a pre-trained quantum neural network to obtain a focus prediction result of the target object.

[0059] Specifically, quantum neural network is a new computing model that combines quantum computing with neural networks. It uses the parallelism, superposition and entanglement characteristics of quantum computing to quantize traditional neural networks to achieve efficient information processing.

[0060] For example, when shooting high-speed particle motion in a laboratory environment, a pre-trained quantum neural network can analyze the continuous image sequence in real time, predict the focal position of the particles at different time points, and provide accurate data support for scientific experiments.

[0061] In the embodiments of the present application, the pre-trained quantum neural network utilizes the superposition state characteristics of quantum bits to process multiple quantum image features simultaneously, significantly improving computing efficiency, especially when processing large-scale continuous image sequences.

[0062] Step S40: adjusting the focal length of the photographic device according to the focus prediction result.

[0063] Specifically, the focus prediction result can be a focus position or a focus parameter. The focus position refers to the coordinates or area range of the area in the picture that needs to be clearly imaged, while the focus parameter is a parameter used to describe the imaging characteristics of the focus area and is used to optimize the focusing effect.

[0064] In the embodiments of this application, dynamic adjustment of the camera's focal length based on focus prediction results significantly improves shooting quality and efficiency, enabling precise focusing. Real-time focus adjustment based on focus prediction results leverages the ultrafast reasoning capabilities of quantum computing, breaking through the limits of classical computing and significantly improving focusing speed and accuracy. This approach is suitable for ultra-high-speed photography and complex scene capture.

[0065] In this embodiment, quantum image features are extracted from a sequence of continuous images of the target object and combined with a quantum neural network to predict the focus of the target object in real time. Leveraging the ultra-fast reasoning capabilities of quantum computing, this method breaks through the limits of classical computing and significantly improves focusing speed and accuracy, making it suitable for ultra-high-speed photography and complex scene capture.

[0066] In some embodiments, please refer to Figure 2 The pre-trained quantum neural network is obtained through training from step S31 to step S33, which is described in detail below.

[0067] Step S31: Obtain a training data set.

[0068] Specifically, the training dataset includes training image sequences and their corresponding focus annotations under different environmental conditions. The environmental conditions include the movement speed of the training object, the scene complexity of the scene in which the training object is located, and the lighting conditions of the training object.

[0069] For example, the motion speeds of training objects include low, medium, and high speeds. A low-speed training object might be a robotic arm moving at a constant speed in a laboratory; a medium-speed training object might be a product sliding at a constant speed of 1 m / s on an industrial assembly line; and a high-speed training object might be a high-speed particle in a laboratory environment. The scene complexity of the training object reflects the number and diversity of background interference elements, which directly affects the encoding accuracy of the quantum depth of field distribution. Scene complexity can be categorized as low, medium, or high. Low complexity scenarios might include objects against a solid background or particle motion in a laboratory vacuum; medium complexity scenarios might include objects in an indoor office environment or biological sample observation under natural light; and high complexity scenarios might involve target identification in a combat environment. Lighting conditions determine the signal-to-noise ratio of the qubit superposition state in quantum image features, directly impacting the reliability of the input data. Lighting conditions can be categorized as strong, stable, low, or dynamic. A strong and stable light environment can refer to a surgical scene under a shadowless lamp in an operating room, a weak light environment can refer to nighttime city monitoring, and a dynamic light environment can refer to the alternating light and dark in a tunnel where a car is driving.

[0070] In the embodiment of the present application, by obtaining a diverse training data set to train the model, the generalization ability and robustness of the model can be improved.

[0071] Step S32: extracting a training feature set corresponding to the training data set, and establishing a mapping relationship between the training image features in the training feature set and their corresponding focus annotations through a preset quantum variational algorithm.

[0072] Specifically, the training image features include training quantum bits, training quantum dynamic trajectories, and training quantum depth of field distributions. A mapping relationship between the training image features in the training feature set and their corresponding focus annotations is established through a preset quantum variational algorithm to perform focus standardization in the training image features.

[0073] Step S33: training the quantum neural network to be trained based on the mapping relationship to obtain a trained quantum neural network.

[0074] In some embodiments, please refer to Figure 3 Step S33: training the quantum neural network to be trained based on the mapping relationship to obtain a trained quantum neural network, including steps S331 to S333, which are described in detail below.

[0075] Step S331: input the training image features in the mapping relationship into the quantum neural network to be trained to obtain the focus prediction result.

[0076] Step S332: Evaluate the focus prediction result based on the focus label in the mapping relationship and the preset evaluation index to obtain an evaluation result.

[0077] Specifically, the preset evaluation indicators include focus prediction error and millisecond focusing success rate. Among them, the focus prediction error includes but is not limited to two-dimensional focus prediction error, three-dimensional focus prediction error and normalized focus error. Two-dimensional focus prediction error refers to the deviation in two-dimensional space between the predicted focus position in the focus prediction result and the real focus position in the preset focus reference position, and the Euclidean distance between the two can be calculated. Three-dimensional focus prediction error refers to the deviation in three-dimensional space between the predicted focus position in the focus prediction result and the real focus position in the preset focus reference position. Normalized focus error refers to scaling the error between the predicted focus position in the focus prediction result and the real focus position in the preset focus reference position to between 0 and 1 or a specific range to eliminate the dimensional effect. The millisecond focusing success rate is used to evaluate the proportion of successfully predicted real focus positions in the focus prediction results.

[0078] In an embodiment of the present application, the evaluation result is used to measure the error between the focus reference position and the focus prediction position to facilitate subsequent optimization and training of the model.

[0079] Step S333: Optimizing the model parameters of the quantum neural network to be trained according to the evaluation results until the quantum neural network to be trained reaches convergence or a preset number of training rounds, thereby obtaining a trained quantum neural network.

[0080] In some embodiments, please refer to Figure 4, step S331: input the training image features in the mapping relationship into the quantum neural network to be trained to obtain the focus prediction result, including steps S331a to S331c, which are described in detail below.

[0081] Step S331a: assign corresponding weights to the training quantum bits, training quantum dynamic trajectories, and training quantum depth of field distribution respectively.

[0082] Specifically, the weights assigned to training quantum bits, training quantum dynamic trajectories, and training quantum depth of field distributions can be dynamically adjusted.

[0083] Step S331b: Based on the preset quantum attention mechanism and the assigned weights, weighted summation is performed on the training quantum bits, the training quantum dynamic trajectories, and the training quantum depth of field distribution to obtain a comprehensive feature representation.

[0084] Step S331c: Decode the comprehensive feature representation into the corresponding focus prediction result.

[0085] Specifically, the comprehensive feature representation is decoded into classical data, that is, the focus prediction result of the target object. The focus prediction result can be further optimized or adjusted according to actual application requirements to improve the accuracy and robustness of the focus prediction.

[0086] In some embodiments, please refer to Figure 5 , step S20: before extracting the quantum image features corresponding to the continuous image sequence, the focus optimization method further includes steps S21 to S22, which are described in detail below.

[0087] Step S21: identifying the scene category of the scene where the target object is located based on a preset quantum classification algorithm and the continuous image sequence.

[0088] Specifically, each frame of the continuous image sequence is quantum-encoded to generate a spatiotemporal feature quantum state, which is then input into a trained quantum classifier to output a quantum probability distribution of the scene category. The deterministic classification result corresponding to the quantum probability distribution is obtained through quantum measurement to obtain the scene category of the scene where the target object is located. Among them, the scene categories include low-light scenes, high dynamic range scenes, and multi-target interference scenes. A low-light scene refers to a scene with extremely low ambient light intensity, resulting in an insufficient number of photons captured by the imaging device. A high dynamic range scene refers to a scene with both extremely bright and extremely dark areas. A multi-target interference scene refers to a scene with multiple targets of similar or different categories.

[0089] For example, the low-light scene can be a nighttime surveillance scene, the high-dynamic-range scene can be backlit photography, where the background is too bright and the face is too dark when taking a portrait, and the multi-target interference scene can be a dense crowd under crowd monitoring.

[0090] Step S22: adjusting the preset quantum feature extraction strategy according to the scene category, and using the adjusted quantum feature extraction strategy to extract quantum image features corresponding to the continuous image sequence.

[0091] Specifically, the quantum feature extraction strategy includes the priority of extracting quantum image features, and the quantum feature extraction strategy is dynamically adjusted by analyzing the scene category.

[0092] For example, in ultra-high-speed scenarios, motion-related quantum features are preferentially extracted, while in complex scenarios, focus priority is optimized based on quantum state distribution.

[0093] In an embodiment of the present application, combined with deep learning technology, the quantum feature extraction strategy is dynamically adjusted by analyzing the scene categories (such as high-speed objects and microscopic structures) in the continuous image sequence. By analyzing the scene categories, auxiliary support can be provided for focus prediction and the focusing effect can be improved.

[0094] In some embodiments, quantum bits are obtained by converting image pixels and edge information of a continuous image sequence through quantum state encoding, and the quantum dynamic trajectory is obtained by predicting the dynamic change trend of the target object in the continuous image sequence using a preset quantum optical flow analysis algorithm. The quantum depth of field distribution is the quantized distribution of the depth of the scene in which the target object is located.

[0095] In some embodiments, the photographic device includes a single camera or multiple cameras. A photographic device including multiple cameras can be used to capture a sequence of continuous images of a target object from different perspectives, thereby obtaining more comprehensive visual data. By synchronously capturing continuous image sequences from different perspectives and converting them into quantum states, focus prediction is collaboratively optimized. When using multiple cameras to capture a target object, the cameras can be divided into a primary camera and a secondary camera, with the primary and secondary cameras working collaboratively.

[0096] For example, in ultra-high-speed scientific experiment shooting scenarios, in order to capture transient phenomena (such as explosions and particle motion), quantum focusing is performed through the main camera, and the secondary camera provides multi-perspective quantum feature correction.

[0097] In some embodiments, when the target object's speed changes, a preset quantum optical flow algorithm is used to detect the target object's speed and adjust the focus to follow the target object. When the target object's lighting conditions change, the brightness distribution is analyzed in conjunction with quantum image features, and the focal length of the camera is optimized. When the capture task type is determined, the focus strategy is adjusted based on preset parameters (such as speed priority or detail priority).

[0098] In some embodiments, to ensure focusing accuracy, capturing a sequence of continuous images of a target object requires the following imaging conditions: 1. The image resolution must be high enough to support quantum state encoding. 2. The dynamic range must meet a minimum threshold to avoid distortion of quantum features. 3. The lighting conditions must be compatible with the quantum computing hardware to ensure data input stability.

[0099] In some embodiments, the quantum deep learning-based focusing optimization method provided by the present application is applicable to future quantum cameras, ultra-high-speed imaging equipment, and scientific photography devices. By embedding a quantum neural network into a quantum computing chip, combined with optical sensors and traditional GPUs, real-time computing is achieved. This increases focusing speed to sub-millisecond levels, significantly improving the success rate of ultra-high-speed photography.

[0100] In some embodiments, the focusing optimization method based on quantum deep learning of the embodiments of the present application can be applied in the following scenarios. Scenario 1: Transient shooting for ultra-high-speed scientific photography. Shoot high-speed particle motion in the laboratory, analyze quantum characteristics through quantum neural networks, adjust the focus to the particle trajectory in real time, and generate clear transient images. Scenario 2: Dynamic target tracking for quantum imaging equipment. Use a multi-camera quantum system to capture high-speed moving targets (such as aircraft), and the quantum neural network predicts the focus position to ensure picture continuity and detail integrity. Scenario 3: Microscopic imaging in complex environments. In microscopic photography tasks, optimize the focus through quantum deep learning, dynamically adapt to changes in lighting and target depth, and obtain high-quality images.

[0101] Corresponding to the focusing optimization method based on quantum deep learning in the above embodiment, Figure 6 A structural schematic diagram of a focusing optimization device based on quantum deep learning provided in an embodiment of the present application is shown. For the sake of convenience, only the parts related to the embodiment of the present application are shown.

[0102] Please refer to Figure 6 The focusing optimization device based on quantum deep learning includes an image data acquisition module 10, an image feature extraction module 20, an object focus prediction module 30 and a device focal length adjustment module 40, which are described in detail below.

[0103] In some embodiments, the object focus prediction module 30 is further configured to train a pre-trained quantum neural network:

[0104] Obtain a training dataset; wherein the training dataset includes training image sequences and their corresponding focus annotations under different environmental conditions, and the environmental conditions include the motion speed of the training object, the scene complexity of the scene in which the training object is located, and the lighting conditions of the training object.

[0105] A training feature set corresponding to the training data set is extracted, and a mapping relationship between the training image features in the training feature set and their corresponding focus annotations is established through a preset quantum variational algorithm; wherein the training image features include training quantum bits, training quantum dynamic trajectories, and training quantum depth of field distribution.

[0106] The quantum neural network to be trained is trained based on the mapping relationship to obtain a trained quantum neural network.

[0107] In some embodiments, the object focus prediction module 30 is further configured to train the quantum neural network to be trained based on the mapping relationship to obtain a trained quantum neural network, including:

[0108] The training image features in the mapping relationship are input into the quantum neural network to be trained to obtain the focus prediction results.

[0109] The focus prediction result is evaluated based on the focus annotation in the mapping relationship and the preset evaluation index to obtain the evaluation result.

[0110] The model parameters of the quantum neural network to be trained are optimized according to the evaluation results until the quantum neural network to be trained reaches convergence or a preset number of training rounds, thereby obtaining a trained quantum neural network.

[0111] In some embodiments, the object focus prediction module 30 is further configured to input the training image features in the mapping relationship into the quantum neural network to be trained to obtain a focus prediction result, including:

[0112] Corresponding weights are assigned to training quantum bits, training quantum dynamic trajectories, and training quantum depth of field distribution respectively.

[0113] Based on the preset quantum attention mechanism and the assigned weights, the training quantum bits, training quantum dynamic trajectories and training quantum depth of field distributions are weightedly summed to obtain a comprehensive feature representation.

[0114] Decode the comprehensive feature representation into the corresponding focus prediction result.

[0115] In some embodiments, before extracting quantum image features corresponding to the continuous image sequence, the image data acquisition module 10 is further configured to:

[0116] The scene category of the target object is identified based on a preset quantum classification algorithm and a continuous image sequence; the scene categories include low illumination scenes, high dynamic range scenes, and multi-target interference scenes.

[0117] The preset quantum feature extraction strategy is adjusted according to the scene category, and the adjusted quantum feature extraction strategy is used to extract quantum image features corresponding to the continuous image sequence.

[0118] In some embodiments, quantum bits are obtained by converting image pixels and edge information of a continuous image sequence through quantum state encoding, and the quantum dynamic trajectory is obtained by predicting the dynamic change trend of the target object in the continuous image sequence using a preset quantum optical flow analysis algorithm. The quantum depth of field distribution is the quantized distribution of the depth of the scene in which the target object is located.

[0119] In some embodiments, the photographic device includes a single camera or multiple cameras; the image data acquisition module 10 is used to acquire a sequence of continuous images of the target object, including:

[0120] photographing a target object using a photographic device comprising a single camera to obtain a sequence of continuous images; or

[0121] A photographic device including multiple cameras is used to obtain a sequence of continuous images of a target object at different viewing angles.

[0122] In some embodiments, the present application further provides a focusing optimization device based on quantum deep learning, comprising:

[0123] Memory, used to store programs;

[0124] The processor is configured to implement the focusing optimization method by executing a program stored in the memory.

[0125] In some embodiments, the present application also provides a computer program product, including a computer program and / or instructions, which implements a focusing optimization method when the computer program and / or instructions are executed by a processor.

[0126] Those skilled in the art will appreciate that all or part of the functions of the various methods in the above embodiments can be implemented by hardware or by computer program. When all or part of the functions in the above embodiments are implemented by computer program, the program can be stored in a computer-readable storage medium, and the storage medium can include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to implement the above functions. For example, the program is stored in the memory of the device, and when the program in the memory is executed by the processor, all or part of the above functions can be implemented. In addition, when all or part of the functions in the above embodiments are implemented by computer program, the program can also be stored in a storage medium such as a server, another computer, disk, optical disk, flash disk or mobile hard disk, and saved in the memory of the local device by downloading or copying, or the system of the local device is updated. When the program in the memory is executed by the processor, all or part of the functions in the above embodiments can be implemented.

[0127] The above examples are used to illustrate the present invention, which are only used to help understand the present invention and are not intended to limit the present invention. Those skilled in the art can make several simple deductions, modifications or substitutions based on the concept of the present invention.

Claims

1. A focusing optimization method based on quantum deep learning, characterized in that: include: Acquire a sequence of continuous images of a target object; wherein the target object is photographed using a photographic device to acquire the sequence of continuous images; Extracting quantum image features corresponding to the continuous image sequence; wherein the quantum image features include quantum bits, quantum dynamic trajectories, and quantum depth of field distribution; Inputting the quantum image features into a pre-trained quantum neural network to obtain a focus prediction result of the target object; The focal length of the photographic device is adjusted according to the focus prediction result.

2. The focusing optimization method according to claim 1, wherein: The pre-trained quantum neural network is trained in the following way: Acquire a training data set; wherein the training data set includes training image sequences and corresponding focus annotations under different environmental conditions, wherein the environmental conditions include the motion speed of the training object, the scene complexity of the scene in which the training object is located, and the lighting conditions of the training object; Extracting a training feature set corresponding to the training data set, and establishing a mapping relationship between training image features in the training feature set and their corresponding focus annotations using a preset quantum variational algorithm; wherein the training image features include training quantum bits, training quantum dynamic trajectories, and training quantum depth of field distributions; The quantum neural network to be trained is trained based on the mapping relationship to obtain a trained quantum neural network.

3. The focusing optimization method according to claim 2, wherein: The step of training the quantum neural network to be trained based on the mapping relationship to obtain a trained quantum neural network includes: Inputting the training image features in the mapping relationship into the quantum neural network to be trained to obtain a focus prediction result; Evaluating the focus prediction result based on the focus annotation in the mapping relationship and a preset evaluation index to obtain an evaluation result; The model parameters of the quantum neural network to be trained are optimized according to the evaluation results until the quantum neural network to be trained reaches convergence or a preset training round, thereby obtaining a trained quantum neural network.

4. The focusing optimization method according to claim 3, wherein: The step of inputting the training image features in the mapping relationship into the quantum neural network to be trained to obtain a focus prediction result includes: Assign corresponding weights to the training quantum bits, the training quantum dynamic trajectories, and the training quantum depth of field distribution, respectively; Performing a weighted summation on the training quantum bits, the training quantum dynamic trajectory, and the training quantum depth of field distribution based on a preset quantum attention mechanism and assigned weights to obtain a comprehensive feature representation; The comprehensive feature representation is decoded into a corresponding focus prediction result.

5. The focusing optimization method according to claim 1, wherein: Before extracting the quantum image features corresponding to the continuous image sequence, the focus optimization method further includes: Identifying the scene category of the target object based on a preset quantum classification algorithm and the continuous image sequence; wherein the scene category includes low illumination scene, high dynamic range scene, and multi-target interference scene; The preset quantum feature extraction strategy is adjusted according to the scene category, and the adjusted quantum feature extraction strategy is used to extract quantum image features corresponding to the continuous image sequence.

6. The focusing optimization method according to claim 1, wherein: The quantum bits are obtained by converting the image pixels and edge information of the continuous image sequence through quantum state encoding. The quantum dynamic trajectory is obtained by predicting the dynamic change trend of the target object in the continuous image sequence using a preset quantum optical flow analysis algorithm. The quantum depth of field distribution is the quantized distribution of the depth of the scene in which the target object is located.

7. The focusing optimization method according to claim 1, wherein: The photographic device includes a single camera or multiple cameras; the step of obtaining a sequence of continuous images of a target object includes: photographing the target object using a photographic device including a single camera to obtain the continuous image sequence; or; A photographic device including multiple cameras is used to obtain a sequence of continuous images of the target object at different viewing angles.

8. A focusing optimization device based on quantum deep learning, characterized in that: include: An image data acquisition module is used to acquire a sequence of continuous images of a target object; wherein the sequence of continuous images is acquired by photographing the target object using a photographic device; An image feature extraction module, configured to extract quantum image features corresponding to the continuous image sequence; wherein the quantum image features include quantum bits, quantum dynamic trajectories, and quantum depth of field distribution; An object focus prediction module, configured to input the quantum image features into a pre-trained quantum neural network to obtain a focus prediction result of the target object; The device focal length adjustment module is used to adjust the focal length of the photographic device according to the focus prediction result.

9. A focusing optimization device based on quantum deep learning, characterized in that: include: Memory, used to store programs; A processor, configured to implement the focusing optimization method according to any one of claims 1 to 7 by executing the program stored in the memory.

10. A computer program product comprising a computer program and / or instructions, characterized in that When the computer program and / or instructions are executed by a processor, the focusing optimization method according to any one of claims 1 to 7 is implemented.

Images