Image classification method and system based on new quantum convolutional network

Abstract

The application provides an image classification method and system based on a novel quantum convolution network, without needing to perform complex dimension reduction operation on the image to be classified, only needing to rearrange data twice in a convolution kernel sliding mode and a binary inversion mode, and then encoding the data to the amplitude of a quantum state, compared with an existing quantum convolution neural network used for image classification, the consumption of computing resources is greatly reduced; and the rearrangement operation and the amplitude encoding operation on the image to be classified can retain edge information and position information, convolution operation is performed by using a parameterized quantum circuit, and the extraction effect of local features can be realized.

Description

Technical Field

[0001] This invention belongs to the field of image classification technology, and in particular relates to an image classification method and system based on a novel quantum convolutional network. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] In the era of big data, image data is growing at an astonishing rate, containing a wealth of visual information. Classifying and analyzing this massive amount of image data can help people extract valuable information. Image classification can provide more comprehensive and accurate data support for decision-making, thereby promoting development and progress in various fields.

[0004] Deep learning methods have demonstrated excellent performance in image classification, with convolutional neural network (CNN) models being particularly outstanding. As a crucial branch of artificial intelligence, deep learning has made significant progress over the past decade. It has achieved substantial advancements in algorithms, computing power, and data, enabling AI technology to overcome commercial limitations and expanding its industry applications. Classical CNNs, a commonly used network model in deep learning, exhibit significant performance in image classification. Their ability to extract local features is crucial for CNNs, allowing them to better capture local information, reduce computational load, and increase network depth. However, classical CNNs have limitations. Their large number of parameters increases model complexity, makes training time-consuming, and requires substantial computational resources. Furthermore, these limitations further restrict the performance of CNNs in image classification as image data sizes increase.

[0005] With the potential of quantum computers to outperform any foreseeable classical computer in solving certain computational tasks, quantum deep learning has emerged as a potential solution to the challenge of handling ever-increasing volumes of data. While some quantum convolutional neural networks (CNNs) for image classification have been proposed, existing CNNs lack the ability to extract local features, and they still require dimensionality reduction of classical data using methods such as PCA, consuming significant classical computational resources. Furthermore, their performance deteriorates as the number of classification categories increases.

[0006] Therefore, while utilizing quantum computing to solve the problem of classical convolutional neural networks consuming a large amount of computational resources, effectively preserving the important ability of classical convolutional neural networks to extract local features, and enabling them to exhibit better generalization ability in image classification problems, is a technical problem that needs to be solved. Summary of the Invention

[0007] To overcome the shortcomings of the prior art, this invention provides an image classification method and system based on a novel quantum convolutional network. The method performs two data rearrangements, one using a sliding convolution kernel and the other using a binary inversion method, and then encodes the data onto the amplitude of the quantum state. This significantly reduces the consumption of computational resources. Then, by using parameterized quantum circuits for convolution operations, the method can achieve the effect of extracting local features.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] In a first aspect, the present invention provides an image classification method based on a novel quantum convolutional network, comprising:

[0010] Acquire the image to be classified and convert the acquired image to be classified into one-dimensional data features;

[0011] The transformed one-dimensional data features are rearranged sequentially using convolution kernel sliding and binary inversion methods.

[0012] The rearranged one-dimensional data features are encoded into the amplitude of the qubit;

[0013] The input quantum state of a qubit is convolved using parameterized quantum circuits;

[0014] Quantum measurement operations are performed on the convolution result to obtain the corresponding feature values, and the image classification result is obtained based on the obtained feature values.

[0015] Secondly, the present invention provides an image classification system based on a novel quantum convolutional network, comprising:

[0016] The conversion module is configured to: acquire the image to be classified and convert the acquired image to be classified into one-dimensional data features;

[0017] The rearrangement module is configured to rearrange the transformed one-dimensional data features sequentially using a convolution kernel sliding method and a binary inversion method.

[0018] The encoding module is configured to encode the rearranged one-dimensional data features onto the amplitude of the qubit;

[0019] The convolution module is configured to perform convolution processing on the input quantum state of the qubit using parameterized quantum circuits;

[0020] The quantum measurement module is configured to perform quantum measurement operations on the result of convolution processing to obtain the corresponding feature values, and obtain the image classification result based on the obtained feature values.

[0021] Thirdly, the present invention provides an electronic device including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0022] Fourthly, the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in the first aspect.

[0023] The above one or more technical solutions have the following beneficial effects:

[0024] In this invention, there is no need to perform complex dimensionality reduction operations on the images to be classified. It only requires two data rearrangements, one using the convolution kernel sliding method and the other using the binary inversion method, and then encoding the data into the amplitude of the quantum state. Compared with existing quantum convolutional neural networks for image classification, this invention significantly reduces the consumption of computational resources. Moreover, the rearrangement and amplitude encoding operations on the images to be classified can preserve information such as edges and positions. By using parameterized quantum circuits for convolution operations, it is possible to achieve the effect of extracting local features.

[0025] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0026] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0027] Figure 1 This is a flowchart of a quantum convolutional neural network with local feature extraction capability in Embodiment 1 of the present invention;

[0028] Figure 2 For existing quantum convolutional neural networks used for image binary classification;

[0029] Figure 3 This refers to the quantum convolutional neural network with local feature extraction capability in Embodiment 1 of the present invention;

[0030] Figure 4 This is a partial matrix of the Ansatz parameterization circuit of the quantum convolutional neural network with local feature extraction capability in Embodiment 1 of the present invention;

[0031] Figure 5 This is the loss function for image binary classification processed by the quantum convolutional neural network with local feature extraction capability in Embodiment 1 of the present invention;

[0032] Figure 6 This is the loss function for image four-class classification using a quantum convolutional neural network with local feature extraction capability in Embodiment 1 of the present invention;

[0033] Figure 7 This is the loss function for image seven-classification processing using a quantum convolutional neural network with local feature extraction capability in Embodiment 1 of the present invention;

[0034] Figure 8 This is a schematic diagram of the quantum bit route designed in Embodiment 1 of the present invention. Detailed Implementation

[0035] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0036] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0037] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0038] Example 1

[0039] Local feature extraction is a concept in image processing and computer vision. It refers to the process of extracting information and features from local regions of an image, such as small pixel blocks or superpixels. These features typically include attributes such as edges, texture, color, and shape, which can describe the details and structure in the image.

[0040] Local feature extraction (LAN) can extract key information from raw image data, reducing data volume and facilitating subsequent processing. This method is adaptable to different image content and scenes, supporting various visual tasks and forming the basis for many innovative applications. The extracted local features can form feature representations at different levels, from simple edges and textures to complex shapes and structures. It exhibits good robustness to image noise and occlusion, allowing for the identification of major objects even if parts of the image are damaged or occluded. It is fundamental to pattern recognition and image classification, providing machine learning algorithms with the information needed to identify and classify image content. In Convolutional Neural Networks (CNNs), local feature extraction is achieved through convolutional layers. These layers use learnable convolutional kernels to automatically extract local features from images, a key factor in the success of CNNs in visual tasks, significantly improving computational efficiency.

[0041] Convolutional Neural Networks (CNNs) are deep learning models that excel in image recognition and processing tasks, primarily due to their powerful local feature extraction capabilities. CNNs automatically extract local features through convolutional layers. These layers use small convolutional kernels that slide across the input image, capturing key information about local regions, such as edges, textures, and shapes. The importance of local feature extraction lies in its ability to extract meaningful patterns from complex image data, which are fundamental to image recognition and classification. Because the parameters of the convolutional kernels are shared throughout the network, CNNs can learn rich feature representations with fewer parameters. This not only improves computational efficiency but also enhances the model's ability to generalize to new data.

[0042] The hierarchical structure of CNNs allows the model to learn features progressively from simple to complex. Lower layers may focus on basic edge features, while higher layers may recognize more complex shapes and object features. This hierarchical feature extraction is key to the success of CNNs in visual tasks and is the reason for their widespread application in many fields such as image classification, object detection, and image segmentation.

[0043] Quantum Neural Networks (QNNs) are a cutting-edge technology combining quantum computing and neural networks. They utilize qubits for information encoding and apply quantum gates to perform weight matrix operations similar to those in traditional neural networks. The core advantage of QNNs lies in their ability to leverage quantum mechanical properties, such as superposition and entanglement, to handle complex computational tasks. Quantum Convolutional Neural Networks (QCNNs) have shown unique potential in image classification. Compared to classical CNNs, QCNNs can perform efficient feature extraction in quantum state space, giving them a potential computational advantage when processing high-dimensional data.

[0044] Despite the immense theoretical potential of quantum convolutional neural networks (CNNs), they are currently in the early stages of research and development, with the limitations of quantum hardware being a major challenge. However, with the continuous advancement of quantum computing technology, CNNs are expected to play a significant role in future image processing and classification tasks.

[0045] This embodiment discloses an image classification method based on a novel quantum convolutional network, including:

[0046] Acquire the image to be classified and convert the acquired image to be classified into one-dimensional data features;

[0047] The transformed one-dimensional data features are rearranged sequentially using convolution kernel sliding and binary inversion methods.

[0048] The rearranged one-dimensional data features are encoded into the amplitude of the qubit;

[0049] The input quantum state of a qubit is convolved using parameterized quantum circuits;

[0050] Quantum measurement operations are performed on the convolution result to obtain the corresponding feature values, and the image classification result is obtained based on the obtained feature values.

[0051] This embodiment draws inspiration from classic convolutional neural networks, emphasizing the role of local feature extraction capabilities in classic convolutional neural networks. It references the important method of local feature extraction and implements it through data preprocessing and the design of parameterized circuits in Ansatz. Figure 1 The main process of this model can be simply divided into two parts: data preprocessing and training the model to obtain prediction results.

[0052] The following sections will provide a detailed introduction to these two parts.

[0053] Step 1: Data Preprocessing

[0054] As shown in Table 1, this embodiment uses data from a single image in the MNIST dataset as an example for detailed explanation:

[0055] Step 11: Expand the image data from 28*28 to 32*32 using zero-padding, which helps the encoded data conform more closely to the logistic wave function; then flatten the image data into 1024-dimensional one-dimensional data, resulting in the initial two rows in Table 1, with indices 0 to 1023 representing image data a0 to a0 respectively. 1023 .

[0056] Step 12: First image data reordering.

[0057] Following the classic convolutional neural network approach, using a 4x4 convolutional kernel with a stride of 4, the image data is grouped into 64 groups of 16 data points for each convolution. For example, the first group of 16 data points is a0, a1, a2, a3, a... 32 a 33 a 34 a 35 a 64 a 65 a 66 a 67 a 96 a 97 a 98 a 99 These 16 data points are arranged sequentially in the first 16 positions. The convolution kernel slides to the right with a stride of 4, again containing 16 data points and arranging them in positions 17-32, and so on, resulting in the two rows in Table 1 after one sorting. As can be seen from Table 1, after one sorting, some image data at certain indices have changed; for example, the data at position 96 has changed from a... 96 Change to a 24 The data at position 97 is from a 97 Change to a 25 The data at position 512, being the last data in group 32, remained unchanged.

[0058] Step 13: Second data reordering.

[0059] This section primarily addresses the programming framework used in this embodiment. Since the high and low bits are reversed in this framework, the binary string corresponding to the index needs to be reversed. For example, the binary string at position 1 is 0000000001, which, after reversal, becomes 1000000000, corresponding to 512. Therefore, the data at position 1 and position 512 need to be swapped. This operation results in the two rows of the secondary sort in Table 1.

[0060] This concludes the data preprocessing section. The above steps have processed the MNIST dataset into the data required by the model of this invention, facilitating subsequent operations.

[0061] Table 1:

[0062]

[0063]

[0064] Step 2: Model training and testing.

[0065] The following describes the structure of the quantum convolutional neural network model with local feature extraction capability of this invention:

[0066] 1. Data encoding.

[0067] Data encoding is the efficient mapping of classical data onto quantum states to leverage the advantages of quantum computing. Common quantum data encoding methods include direct encoding, angle encoding, and amplitude encoding. This embodiment uses amplitude encoding to encode 1024-dimensional image data onto the amplitude of 10 qubits, which is a major advantage of quantum computing.

[0068] 2. Design of parameterized quantum circuit Ansatz.

[0069] like Figure 2 As shown, this quantum convolutional neural network is an existing network model for image binary classification. Aside from the data encoding and measurement parts, what remains is the parameterized quantum circuitry. It can be seen that the parameterized quantum circuitry of this model is quite complex.

[0070] like Figure 8 As shown, the parameterized quantum circuit in this embodiment uses a four-qubit circuit with an All-to-All structure. R is added to each of the four qubits respectively. X and R Z After two parameterized quantum gates, each qubit is used as a control bit to apply controlled R to the remaining three qubits. X Operations that enable full connectivity.

[0071] The network model in this embodiment, such as Figure 3 As shown, since the data has been reordered, only one U operation with 20 parameters is needed to perform a convolution process on the data.

[0072] like Figure 4The image shows the matrix representation of the parameterized quantum circuit of the model. The matrix is ​​1024*1024 in size, with 16*16 parameter-related values ​​only on the diagonal, and zero values ​​in all other positions. Each row of this 16*16 matrix is ​​equivalent to a 4*4 convolution kernel, which is multiplied by 16 image data points to perform convolution. The 16 rows are 16 different convolution kernels.

[0073] After processing with this matrix, the image data is equivalent to undergoing a classic convolution process with a kernel size of 4*4, a stride of 4, and 16 channels.

[0074] 3. Quantum measurement.

[0075] To obtain the feature values ​​extracted after convolution of the image data, the model uses a probabilistic measurement method to obtain the squares of the 1024 amplitudes of the 10 qubits. The measured values ​​are the feature values ​​after convolution with 16 convolution kernels, with 64 feature values ​​per channel. Single-channel or multi-channel feature values ​​can be selected as the basis for classification and post-processing.

[0076] Once the model is built, it can be trained and tested using the processed image data. In this example, the initial learning rate is set to 0.1, the Adam optimizer is used, and the cross-entropy loss function is used. The model is trained for 10 epochs and then tested. The specific results will be shown later.

[0077] Quantum convolutional neural networks for image classification are an important research direction combining quantum computing and deep learning. This embodiment has the following advantages compared to other quantum convolutional neural networks for image classification:

[0078] (1) No complex dimensionality reduction processing is required for the data; only two simple reordering steps are needed. The preprocessed data is encoded onto the amplitude of the quantum state using amplitude encoding. Compared with other existing quantum convolutional neural networks for image classification, this embodiment significantly reduces the consumption of computational resources.

[0079] (2) It has the ability to extract local features. The processed data is zero-filled at the edges. The image data is sequentially encoded into the amplitude of each quantum state by amplitude encoding, preserving edge information, position information, etc. The parameterized quantum circuit Ansatz is designed to realize local feature extraction.

[0080] (3) Lower model complexity. Existing quantum convolutional neural networks require multiple convolutional blocks, such as... Figure 2 In this embodiment, the parameterized quantum circuit only requires one convolutional block, such as... Figure 3 As shown.

[0081] (4) The model has stronger generalization ability. Existing quantum convolutional neural network models only show good results for binary or quadruple classification of the MNIST dataset, but perform poorly for more categories. In contrast, this model shows good classification results for binary, quadruple, and seven-class classification of the MNIST dataset.

[0082] In this example, the MNIST dataset was used to test this quantum convolutional neural network model, measuring its accuracy on binary, quadrilateral, and septial classification problems. Compared to existing quantum convolutional neural networks (binary classification accuracy 98.4%, quadrilateral classification accuracy 90.03%, septial classification accuracy approximately 50%), the model in this embodiment demonstrates superior performance with lower circuit complexity as the number of image classification categories increases. Specific results are shown below:

[0083] In image binary classification, this embodiment performs exceptionally well, achieving a final accuracy of 99.80%. The loss function decreases rapidly and converges quickly. Figure 5 As shown in the figure. The model's prediction results are shown in Table 2.

[0084] Table 2: Binary Classification Prediction Results

[0085]

[0086] This embodiment still performs well in image four-class classification, achieving an accuracy of 92.74%. The loss function decreases smoothly and gradually converges. Figure 6 As shown in Table 3, the final prediction results for the four-class classification of images by the model are shown in Table 3.

[0087] Table 3: Four-category Prediction Results

[0088]

[0089] In classification problems involving multiple categories where other quantum convolutional neural networks perform poorly, this invention demonstrates significantly better performance, such as... Figure 7 As shown, the classification accuracy of the seven-class classification can reach 86.69%, and the loss function decreases smoothly, although it fluctuates relatively large after convergence. Specific prediction results are shown in Table 4.

[0090] Table 4: Seven-category Prediction Results

[0091]

[0092] Example 2

[0093] The purpose of this embodiment is to provide an image classification system based on a novel quantum convolutional network, including:

[0094] The conversion module is configured to: acquire the image to be classified and convert the acquired image to be classified into one-dimensional data features;

[0095] The rearrangement module is configured to rearrange the transformed one-dimensional data features sequentially using a convolution kernel sliding method and a binary inversion method.

[0096] The encoding module is configured to encode the rearranged one-dimensional data features onto the amplitude of the qubit;

[0097] The convolution module is configured to perform convolution processing on the input quantum state of the qubit using parameterized quantum circuits;

[0098] The quantum measurement module is configured to perform quantum measurement operations on the result of convolution processing to obtain the corresponding feature values, and obtain the image classification result based on the obtained feature values.

[0099] In further embodiments, the following is also provided:

[0100] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When executed by the processor, the computer instructions perform the method described in Embodiment 1. For brevity, further details are omitted here.

[0101] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0102] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0103] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in Embodiment 1.

[0104] The method in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0105] A computer program product includes a computer program that, when executed by a processor, implements the method described in Embodiment 1.

[0106] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the processes / methods described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided among program modules as needed. The machine-executable instructions for the program modules can execute within a local or distributed device. In a distributed device, the program modules can reside in both local and remote storage media.

[0107] The computer program code used to implement the methods of the present invention may be written in one or more programming languages. This computer program code may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the computer or other programmable data processing device, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a computer, partially on a computer, as a stand-alone software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0108] In the context of this invention, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals may include electrical, optical, radio, sound, or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0109] Those skilled in the art will recognize that the units and algorithm steps described in conjunction with the embodiments herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0110] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for image classification based on a novel quantum convolutional network, characterized in that, The method comprises the following steps: acquiring an image to be classified, and converting the acquired image to be classified into one-dimensional data features; rearranging the converted one-dimensional data features in turn according to a convolution kernel sliding mode and a binary inversion mode; specifically, determining the size of the convolution kernel and the sliding step; taking the data for each convolution as a group of data, and sequentially performing index arrangement to obtain the first rearranged one-dimensional data features; inverting the binary string corresponding to the index of the first rearranged data features in turn, and interchanging the data features under the index after inversion with the data features under the index before inversion; encoding the rearranged one-dimensional data features onto the amplitude of a quantum bit; performing convolution processing on the input quantum state of the quantum bit by using a parameterized quantum circuit; performing quantum measurement operation on the result of the convolution processing to obtain corresponding feature values, and obtaining the image classification result according to the obtained feature values.

2. The image classification method based on the novel quantum convolutional network according to claim 1, wherein, Before converting the acquired image to be classified into one-dimensional data features, the method further comprises performing an expansion operation on the image to be classified, so that the data encoding conforms to the logic wave function.

3. The image classification method based on the novel quantum convolutional network according to claim 1, wherein, The rearranged one-dimensional data features are encoded onto the amplitude of a quantum bit in an amplitude encoding mode.

4. The image classification method based on the novel quantum convolutional network according to claim 1, wherein, The parameterized quantum circuit uses a four-qubit circuit of All-to-All structure, and adds parameterized quantum gates R to four qubits respectively X and R Z After that, each qubit is taken as a control bit to apply a controlled quantum gate R X Operation to the remaining three qubits, realizing full connection.

5. A novel quantum convolutional network-based image classification system, characterized by, The method comprises the following steps: a conversion module configured to acquire an image to be classified, and convert the acquired image to be classified into one-dimensional data features; a rearrangement module configured to rearrange the converted one-dimensional data features in turn according to a convolution kernel sliding mode and a binary inversion mode; specifically, determining the size of the convolution kernel and the sliding step; taking the data for each convolution as a group of data, and sequentially performing index arrangement to obtain the first rearranged one-dimensional data features; inverting the binary string corresponding to the index of the first rearranged data features in turn, and interchanging the data features under the index after inversion with the data features under the index before inversion; an encoding module configured to encode the rearranged one-dimensional data features onto the amplitude of a quantum bit; a convolution module configured to perform convolution processing on the input quantum state of the quantum bit by using a parameterized quantum circuit; a quantum measurement module configured to perform quantum measurement operation on the result of the convolution processing to obtain corresponding feature values, and obtain the image classification result according to the obtained feature values.

6. An electronic device, comprising: The computer program product comprises a memory and a processor, and computer instructions stored in the memory and running on the processor, and when the computer instructions are run by the processor, the method of any one of claims 1-4 is completed.

7. A computer readable storage medium characterized in that, The computer program product is used for storing computer instructions, and when the computer instructions are executed by the processor, the method of any one of claims 1-4 is completed.

Images