Brand-new quantum intelligent calculation method specially designed for smart city

Abstract

The invention relates to a brand-new quantum intelligent calculation method specially designed for a smart city, and the method comprises the steps: constructing a quantum convolution layer, and extracting image features through quantum state coding and quantum gate operation; designing a quantum convolution kernel, and realizing parallel computing by using quantum superposition; the feature learning ability is enhanced through quantum entanglement; optimizing a quantum circuit to reduce computational complexity; according to the method provided by the invention, the excellent performance of the quantum gate and the quantum circuit in the aspect of image feature extraction is fully utilized, and by virtue of the characteristics of quantum superposition and quantum entanglement, the computing power of the quantum convolutional neural network is remarkably improved when complex data and large-scale parallel computing are processed, and the method is suitable for various application scenes of smart cities, and has a wide application prospect. The method provides powerful technical support for efficient data processing and real-time decision making of the smart city, and has a wide application prospect.

Description

Technical Field

[0001] The present invention relates to the field of computer science and technology, and in particular to a new quantum intelligent computing method designed specifically for smart cities. Background Art

[0002] With the acceleration of global urbanization, the concept of smart cities has gradually become an important direction for the development of modern cities. Smart cities aim to improve urban management efficiency, optimize resource allocation, and improve the quality of life of residents by integrating advanced technologies such as the Internet of Things, big data, and artificial intelligence. Traditional convolutional neural networks (CNNs) have achieved remarkable success in image recognition and pattern recognition tasks and are widely used in various fields of smart cities, such as intelligent transportation systems, public safety monitoring, and environmental monitoring. However, with the expansion of urban scale and the surge in data volume, traditional convolutional neural network methods have gradually exposed problems such as insufficient computing power, slow processing speed, and high energy consumption. Especially when processing high-resolution images or large-scale data sets, the computational complexity grows exponentially, resulting in long training and inference time, and high energy consumption does not meet the green and low-carbon requirements of smart cities. At the same time, traditional computing models rely on the serial computing architecture of classical computers, which makes it difficult to fully utilize the advantages of parallel computing, resulting in low efficiency when processing large-scale data, and the limitations are becoming more and more obvious.

[0003] As an emerging computing paradigm, quantum computing, compared with traditional convolutional neural networks, has demonstrated superior performance in image feature extraction and great potential in complex data and large-scale parallel computing due to its unique quantum superposition and quantum entanglement characteristics, which gives quantum computing a clear advantage in making real-time decisions; therefore, quantum convolutional neural networks have shown great potential in various application scenarios of smart cities, such as traffic management, public safety monitoring, and environmental monitoring. Summary of the invention

[0004] In view of this, the present invention provides a new quantum intelligent computing method designed specifically for smart cities, which effectively solves the problems of insufficient computing power, slow processing speed and high energy consumption of traditional data processing methods when facing complex data, and significantly improves the computing power when processing complex data and large-scale parallel computing.

[0005] To achieve the above objectives, a new quantum intelligent computing method designed for smart cities includes the following steps: S1. Establish image dataset And preprocess to get the input image data ; S2, construct the quantum convolution layer QConv; S201, quantum state encoding and feature extraction, mapping the input image data into the quantum state of quantum bits through amplitude encoding , the quantum state passes through the parameterized quantum gate and quantum circuits Complete feature extraction and nonlinear transformation, and output quantum state features ; S202. Constructing quantum convolution kernel , through quantum gate operation Perform feature mapping and characterize the quantum state Perform convolution operation, the expression is: ; in, represents a quantum gate, represents the optimization parameters; S203, using quantum superposition to parallelly calculate all channel characteristics; S204, using quantum entanglement to enhance feature learning capabilities to obtain image feature representation, by constructing an entangled state Introducing nonlocal correlations between multiple features, the entangled state The expression is: ; ; in, Represents the original state Apply transformation The new quantum state obtained is represents the transformation operator acting on the quantum state; S205, Integrated quantum circuit optimization computational complexity; S3, build a fully connected layer; S4. Construct an output layer, use the Softmax function to process the output of the fully connected layer and output the result.

[0006] Preferably, the FashionMNIST grayscale image dataset is divided into 10 categories and used as an input dataset, and the preprocessing includes the following steps: Resize the FashionMNIST grayscale image to 28×28; Convert the FashionMNIST grayscale image to a tensor format; Divide the dataset into training and validation sets; The training set and validation set are converted into batch data through the data loader.

[0007] Preferably, the quantum state characteristic The expression is: ; in, , Represents an index variable, , represents the ground state; The expression for parameterized quantum gate processing is: ; in, represents the parameterized angle, represents the Hermitian; The quantum superposition state is obtained through the quantum superposition operation, and the expression of the quantum superposition state is: ; in, Indicates that the quantum state is in the ground state The probability amplitude on The quantum circuit Including single-qubit gates and multi-qubit gates , the optimization goal of the integrated quantum circuit is to minimize the depth D and the number of quantum gates G, expressed as: ; in, Denotes the maximum number of quantum gates that can be represented.

[0008] The quantum convolution layer is optimized by quantum gradient descent method, and the expression is: ; in, represents the loss function, represents the learning rate, Represents the parameter value in the tth iteration.

[0009] Preferably, the quantum entanglement extracts features in the image through a gate operation, and the gate operation includes a rotation-Z gate and a rotation-Y gate, and the rotation-Y gate expression is: ; Where I represents the identity matrix, represents the rotation-Y gate, and Y represents the matrix of the Pauli-Y gate; The expression of the rotating-Z gate is: ; in, represents the rotation-Z gate, and Z represents the matrix of the Pauli-Z gate.

[0010] Preferably, a Pauli-Z gate is used to convert the features extracted by the spin-Z gate and the spin-Y gate from a quantum state to a classical output to obtain a feature representation of the image, and the matrix of the Pauli-Z gate is: ; in, represents the standard Pauli-Z matrix in quantum computing.

[0011] Preferably, the fully connected layer uses the LeakyReLU function, expressed as: ; in, Represents the input value of the fully connected layer.

[0012] Preferably, the output layer includes 10 neurons, corresponding to the 10 categories of FashionMNIST, and the output values ​​of the 10 neurons are converted into probabilities using the Softmax function. , and satisfies the following expression: ; ; The expression of the Softmax function is: ; in, Indicates The output of each neuron; the FashionMNIST category with the highest probability value is selected as the classification result output.

[0013] Compared with the prior art, the present invention has the following beneficial effects: The present invention introduces the characteristics of quantum superposition and quantum entanglement, so that the quantum convolutional neural network can process multiple computing tasks at the same time, significantly improving the parallel computing capability. Compared with the traditional convolutional neural network, the quantum convolution adopted by the present invention reduces the computational complexity and greatly improves the computational efficiency. At the same time, due to the parallelism of quantum computing, the present invention can complete complex image feature extraction and large-scale data processing tasks in a shorter time, and is suitable for application scenarios that require real-time response in smart cities, such as intelligent transportation systems and public safety monitoring. The present invention optimizes the operation sequence of quantum gates by designing efficient quantum circuits, reduces the depth of quantum circuits and the number of gates, and thus reduces the consumption of computing resources. Compared with traditional computing models, the quantum convolutional Shenjin network significantly reduces energy consumption while maintaining high performance, meeting the green and low-carbon requirements of smart cities. The present invention can capture high-dimensional features and complex relationships in images through the superposition and entanglement characteristics of quantum states, extract more accurate image features, and at the same time can not only capture local features of the image, but also capture global features through quantum entanglement characteristics, thereby improving the generalization ability and classification performance of the model. The present invention has the advantages of high precision and high efficiency under low resource conditions, significantly improves the accuracy in image classification tasks, and can better balance the complexity and generalization ability of the model, reduce the risk of overfitting, and improve the performance of the model on unseen data; The method proposed in the present invention has a positive effect on many aspects of smart cities, such as intelligent transportation systems, public safety monitoring and environmental monitoring; in terms of intelligent transportation systems, it can predict traffic flow in real time, optimize signal control strategies, reduce traffic congestion, and improve the overall efficiency of the transportation system; in terms of public safety monitoring, the present invention can improve the real-time response capability of the monitoring system and enhance the effect of public safety monitoring; in terms of environmental monitoring, the present invention can efficiently process large-scale environmental data, achieve more accurate environmental monitoring and prediction, and provide strong support for the environmental management of smart cities.

[0014] Compared with the existing technology, the present invention significantly improves computing efficiency, reduces energy consumption, enhances feature extraction capabilities, and improves model accuracy and generalization capabilities by introducing the unique advantages of quantum computing. These beneficial effects enable the present invention to demonstrate excellent performance in a variety of application scenarios in smart cities, providing strong technical support for future urban management and services. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] FIG1 is a framework diagram of a quantum convolutional neural network of the present invention; Figure 2 Schematic diagram of quantum circuit in the quantum convolution layer of the present invention; Figure 3 Schematic diagram of the quantum convolution kernel of the present invention; Figure 4 A schematic diagram of features extracted from images in the data set of the present invention; Figure 5 This is a graph of experimental results of the CNN model of the present invention; Figure 6 This is a diagram of the experimental results of the LeNet model of the present invention; Figure 7 This is a diagram of the experimental results of the QCNN model of the present invention; Figure 8 This is a schematic diagram comparing the effects of the QCNN and CNN models of the present invention. DETAILED DESCRIPTION

[0016] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation mode, structure, characteristics and effects of the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments.

[0017] Example 1 With the continuous development of smart cities, the continuous increase in data volume has also led to a growing demand for advanced computing capabilities. Traditional machine learning algorithms are difficult to cope with various complex data and demands in a dynamic environment. Therefore, a new quantum intelligent computing method designed specifically for smart cities is proposed to improve computing power and efficiency through quantum convolutional neural networks.

[0018] The quantum convolutional neural network used in the present invention achieves excellent performance in image feature extraction through quantum gates and quantum circuits, utilizing the characteristics of quantum superposition and quantum entanglement. Compared with traditional convolutional neural networks, quantum convolutional neural networks can process large amounts of data in parallel and extract more accurate image features in a shorter time, which makes the quantum convolutional neural network have obvious advantages in processing complex data and making implementation decisions. In addition, the structural design of the quantum convolutional neural network is simpler, which reduces the consumption of computing resources and improves the practicality and scalability of the model. These advantages enable quantum convolutional neural networks to be used in various application scenarios in smart cities, such as intelligent traffic management, public safety monitoring and environmental monitoring.

[0019] Specifically, the present invention provides the following technical solution: a new quantum intelligent computing method designed specifically for smart cities, comprising the following steps: S1. Establish image dataset And preprocess to get the input image data ; The FashionMNIST grayscale image dataset is used as the input dataset. The dataset includes 60,000 training samples and 10,000 test samples, which are divided into 10 categories. The preprocessing includes the following steps: Resize the FashionMNIST grayscale image to 28×28; Convert FashionMNIST grayscale images to tensor format; The dataset is divided into a training set and a validation set, which occupy 80% and 20% of the original dataset respectively; Convert the training and validation sets into batch data through data loaders for efficient processing during model training and validation. Preprocessing ensures that the input data is standardized and structured, which helps in efficient training and evaluation of the model.

[0020] S2, construct the quantum convolution layer QConv; The quantum circuit implementation of the quantum convolution kernel is as follows Figure 2 As shown in the figure, similar to the classical convolution kernel, the quantum convolution kernel follows a certain receptive field size and step size to traverse the entire image, but its internal calculation process is different, specifically including the following steps: S201, quantum state encoding and feature extraction, input image data through the amplitude encoding of the encoding module to encode the digitized image information into the quantum state of the quantum bit , ( represents the quantum state of the input image data X after quantum encoding, and The Hadamard transformation acts on n quantum bits to make them enter a superposition state, thereby enhancing the parallelism and feature expression capabilities of quantum computing. This encoding process utilizes the superposition and superposition properties of quantum states to embed image information into quantum states in a high-dimensional and parallel manner. Quantum states pass through parameterized quantum gates and quantum circuits Complete feature extraction and nonlinear transformation to quantify input image data The deep feature representation of the output quantum state feature ; Quantum state characteristics The expression is: ; in, , Represents an index variable, , represents the ground state; The expression for parameterized quantum gate processing is: ; in, represents the parameterized angle, represents the Hermitian; S202. Constructing quantum convolution kernel , which is equivalent to the traditional convolution kernel, through quantum gate operation Perform feature mapping and characterize the quantum state Perform convolution operation, the expression is: ; in, represents a quantum gate, represents the optimization parameters; S203, using quantum superposition to parallelly calculate all channel characteristics, and obtaining a quantum superposition state through quantum superposition operation. The expression of the quantum superposition state is: ; in, Indicates that the quantum state is in the ground state The probability amplitude on S204, using quantum entanglement to enhance feature learning capabilities to obtain image feature representation, by constructing an entangled state Non-local correlation is introduced between multiple features to capture the high-dimensional feature relationship between complex data. In the convolution process, the entangled state is used for feature enhancement operations across channel interactions. The expression is: ; ; in, Represents the original state Apply transformation The new quantum state obtained is represents the transformation operator acting on the quantum state; Quantum entanglement extracts features from images through gate operations, which include spin-Z gate and spin-Y gate. The spin-Z gate is generated by the generating elements in the form of Pauli-Y matrix. The spin-Y gate expression is: ; Where I represents the identity matrix, represents the rotation-Y gate, and Y represents the matrix of the Pauli-Y gate; The spin-Y gate is also called a phase-shift gate. It is generated by the spin-Z gate as a generating element. The expression of the spin-Z gate is: ; in, represents the rotation-Z gate, and Z represents the matrix of the Pauli-Z gate; These gate operations enable interactions between quantum states by adjusting amplitude and phase, thereby capturing complex relationships and local features in the image. This interaction plays a key role in quantum convolutional neural networks, helping the model understand and extract abstract features of images, thereby achieving more accurate and efficient image recognition and classification tasks. Then use the Pauli-Z gate to convert the features extracted by the spin-Z gate and the spin-Y gate from the quantum state to the classical output to obtain the feature representation of the image. The matrix of the Pauli-Z gate is: ; in, Represents the standard Pauli-Z matrix in quantum computing, used to operate on the phase of a quantum bit; S205, Integrated quantum circuit optimization computational complexity; Quantum Circuits Including single-qubit gates and multi-qubit gates , the optimization goal of the integrated quantum circuit is to minimize the depth D and the number of quantum gates G, expressed as: ; in, Indicates the maximum number of quantum gates, used to optimize the complexity of quantum circuits; Using quantum superposition to parallelize all channel features in the same quantum circuit, the computational complexity is reduced from 100% to 100% of the traditional convolution. Down to , the computational complexity of quantum circuits is significantly better than traditional convolution calculations; The quantum convolution layer is optimized by quantum gradient descent method, and the expression is: ; in, represents the loss function, represents the learning rate, Represents the parameter value in the tth iteration; The execution process of quantum convolutional neural network is as follows Figure 3 As shown, the quantum state is converted into classical output, and the output of each quantum bit forms a feature map of a channel. In this way, the quantum convolution kernel can effectively process image data and extract rich feature information, providing strong support for subsequent classification or other tasks.

[0021] The algorithm for the quantum circuit is as follows:

[0022] S3, build a fully connected layer; The activation function is a nonlinear function that receives input signals and generates outputs as the activation values ​​of each neuron in the neural network. In convolutional neural networks, the role of the activation function is to introduce nonlinear characteristics so that the neural network can learn complex input-output mapping relationships, which helps to improve the network's expressiveness and adaptability, so that it can better adapt to complex data distributions and solve nonlinear problems that linear models cannot handle. Therefore, in the present invention, the fully connected layer uses the LeakyReLU function, which is expressed as: ; in, Represents the input value of the fully connected layer.

[0023] Compared with the traditional ReLu function, the LeakyReLU function has a small slope in the negative interval, ensuring that the activation function is not completely zero in the negative interval, which helps to maintain the gradient flow during training and increase the stability and nonlinear expression ability of the model. In addition, it also reduces the interaction between neurons during training, making the model more flexible. Through multiple experiments to adjust the parameters, ensure that the model effectively captures the key features and patterns of the data during training, thereby improving the performance and accuracy of the model. This fine-tuning parameter setting enables the model to better adapt to the complexity of the data and converge to the optimal solution faster during training, so the generalization ability and performance of the model are improved. After adjusting the parameters, the optimal parameter settings are obtained as shown in Table 1. Table 1:

[0024] As shown in Table 1, in the quantum convolution layer, the parallelism of quantum bits and the characteristics of quantum computing are used to efficiently extract and process image features. By using a quantum convolution kernel of size 4×4 and stride 2, the computational complexity and memory consumption are reduced while maintaining the effectiveness of the model, thereby accelerating training and improving operational efficiency. The quantum convolution layer is combined with the fully connected layer, so that the model can fully utilize the advantages of quantum computing and classical neural networks, thereby enhancing classification performance and generalization ability.

[0025] S4, constructing an output layer, using the Softmax function to process the output of the fully connected layer and output the result; After the fully connected layer, an output layer is set up, which includes 10 neurons, corresponding to the 10 categories of FashionMNIST. The output values ​​of these 10 neurons represent the unnormalized scores of each category. The Softmax function is used to convert the output values ​​of the 10 neurons into probability values. , and satisfies the following expression: ; ; The model can predict the probability that the input image belongs to each category, and finally select the FashionMNIST category with the highest probability value as the classification result output; The expression of the Softmax function is: ; in, Indicates The output of a neuron; Through the Softmax function, the model can perform predictions for multi-classification tasks, ensuring that the results have good interpretability and accuracy. The output layer can effectively convert the original output of the model into a probability value, making the classification decision clearer and more reliable.

[0026] In the traffic scene, traffic flow data Matrix mapping to quantum states , predicted by QCNN, the prediction result is: ; The prediction results are used to optimize traffic flow distribution and signal control strategies, providing technical support for traffic management in smart cities.

[0027] Example 2 Apply the FashionMNIST dataset to the model. First, split the loaded training dataset into a training set and a validation set. In this process, 80% of the data is used for training and the remaining 20% ​​of the data is used for validation. In order to evaluate the deviation between the model's predicted probability for each category and the actual category, the cross entropy loss function is used. , the expression is: ; in, represents a discrete random variable, Indicates the expected output, Represents the actual output, and the probability distribution function expression is: ; in, express A collection of values; A maximum of ten training iterations are set to efficiently evaluate the performance and effectiveness of the model within a limited time, prevent overfitting on the training set, and ensure that the model can generalize well to unseen data. The quantum convolutional Shenjin network model starts with 10 iterations, such as Figure 4 As shown in Figure 2, the image features are extracted through the quantum convolution layer, and the CNN model and LeNet model are applied to the FashionMNIST dataset. The experimental results of the three models are shown in Table 2: Table 2:

[0028] The traditional convolutional neural network was trained for 10 iterations. The experimental results are as follows: Figure 5 As shown in Table 2 and Figure 5 , as the number of iterations increases, the loss continues to decrease, but the gap between the training loss and the test loss also widens significantly, which indicates that the model is overfitting on the training data and may perform poorly on unseen test data; Figure 5 In the right figure, as the number of iterations increases, the accuracy continues to improve, but the gap between the training accuracy and the test accuracy is also getting bigger and bigger, which means that the model may pay too much attention to the details of the training data during the learning process, resulting in poor performance on the test data, and further optimization is needed to improve the generalization ability; through in-depth comparative research, it can be concluded that: under the same model structure, the quantum convolution layer has a stronger image feature extraction ability than the traditional convolution layer, which not only proves the superiority of the quantum convolution layer in computer vision tasks, but also highlights the potential of quantum computing in processing complex data; because the quantum convolution layer can utilize the parallelism of quantum states and quantum superposition states, it can more fully explore the diversity and complexity of image data, thereby providing more flexible and efficient solutions for tasks such as image recognition and classification.

[0029] The classic convolutional neural network LeNet was trained for 10 iterations. The experimental results are as follows: Figure 6 As shown in Table 2 and Figure 6 , the loss decreases consistently with the increase of the number of iterations; in addition, Figure 6 The right figure shows that the accuracy increases steadily with the increase in the number of iterations. After ten iterations, the accuracy of the model reaches 86%. Compared with the quantum convolutional neural network model proposed in this invention, LeNet has a slight advantage in accuracy. However, during the iteration process, the gap between the test loss and the training loss of the LeNet model is small, and the gap between the training accuracy and the test accuracy is also small, which indicates that the LeNet model has better stability and generalization ability during the training process.

[0030] The loss and accuracy of the quantum convolutional neural network model on the dataset after 10 iterations are shown in Figure 2. Figure 7 As shown in Table 2 and Figure 7 , as the number of iterations increases, the iterations continue to decrease, and at the same time Figure 7 The right figure shows that the accuracy rate continues to rise with the increase of iteration number. After ten iterations, the accuracy rate of the model reaches 87%, which indicates that the model gradually learns the data features during the training process, enabling it to effectively classify images. Compared with the QCNN model, the complexity of LeNet is reflected in its deeper network structure and more parameters. Therefore, the remarkable performance of the quantum convolution layer shows its great potential and application prospects in the field of neural networks in the future, indicating that the quantum convolution layer has a broad development space. Due to the same structure of QCNN and CNN, such as Figure 8As shown in the figure, the loss rate and accuracy of QCNN and CNN on the training set are summarized; by comparing these data, it can be seen that under the same network structure, the quantum convolution layer is more effective than the traditional convolution layer in image feature extraction; over time, the feature extraction ability of the traditional convolution layer is gradually limited, and it is difficult to capture complex data patterns. Although the training accuracy continues to improve, the slow increase in test accuracy indicates that the model has poor generalization performance on new data. The traditional convolution layer may face challenges such as overfitting and local feature optimization, resulting in poor performance on unseen data; On the contrary, quantum convolution layers exhibit stronger feature extraction capabilities and better generalization performance; quantum convolution layers are better at capturing abstract features and global patterns in images, rather than limited to local information; utilizing the characteristics of quantum computing, these convolution layers can process more complex data structures and maintain stable performance during the training and testing phases, reducing the risk of overfitting; in addition, quantum convolution layers may provide higher computational efficiency and lower resource consumption, and have potential advantages in large-scale data processing and deep learning tasks; at the same time, the proposed quantum convolution layer is compared with the classical convolutional neural network model LeNet, and it is found that although the quantum convolution layer has a simpler structure, its performance is comparable to LeNet.

[0031] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A new quantum intelligent computing method designed for smart cities, characterized by: The following steps are involved: S1. Establish image dataset And preprocess to get the input image data ; S2, construct the quantum convolution layer QConv; S201, quantum state encoding and feature extraction, mapping the input image data into the quantum state of quantum bits through amplitude encoding , the quantum state passes through the parameterized quantum gate and quantum circuits Complete feature extraction and nonlinear transformation, and output quantum state features ; S202. Constructing quantum convolution kernel , through quantum gate operation Perform feature mapping and characterize the quantum state Perform convolution operation, the expression is: ； in, represents a quantum gate, represents the optimization parameters; S203, using quantum superposition to parallelly calculate all channel characteristics; S204, using quantum entanglement to enhance feature learning capabilities to obtain image feature representation, by constructing an entangled state Introducing nonlocal correlations between multiple features, the entangled state The expression is: ； ； in, Represents the original state Apply transformation The new quantum state obtained is represents the transformation operator acting on the quantum state; S205, Integrated quantum circuit optimization computational complexity; S3, build a fully connected layer; S4. Construct an output layer, use the Softmax function to process the output of the fully connected layer and output the result.

2. According to claim 1, a new quantum intelligent computing method designed for smart cities is characterized by: The FashionMNIST grayscale image dataset is divided into 10 categories and used as the input dataset. The preprocessing includes the following steps: Resize the FashionMNIST grayscale image to 28×28; Convert the FashionMNIST grayscale image to a tensor format; Divide the dataset into training and validation sets; The training set and validation set are converted into batch data through the data loader.

3. According to claim 1, a new quantum intelligent computing method designed for smart cities is characterized by: The quantum state characteristics The expression is: ； in, , Represents an index variable, , represents the ground state, Indicates that the quantum state is in the ground state , The probability amplitude on The expression for parameterized quantum gate processing is: ； in, represents the parameterized angle, represents the Hermitian; The quantum superposition state is obtained through the quantum superposition operation, and the expression of the quantum superposition state is: ； in, Indicates that the quantum state is in the ground state The probability amplitude on The quantum circuit Including single-qubit gates and multi-qubit gates , the optimization goal of the integrated quantum circuit is to minimize the depth D and the number of quantum gates G, expressed as: ； in, Indicates the maximum number of quantum gates; The quantum convolution layer is optimized by quantum gradient descent method, and the expression is: ； in, represents the loss function, represents the learning rate, Represents the parameter value in the tth iteration.

4. According to claim 1, a new quantum intelligent computing method designed for smart cities is characterized by: The quantum entanglement extracts features in the image through gate operations, and the gate operations include a rotation-Z gate and a rotation-Y gate. The rotation-Y gate expression is: ； Where I represents the identity matrix, represents the rotation-Y gate, Y represents the matrix of the Pauli-Y gate, represents the unitary evolution operator in the exponential form of the rotation-Y gate matrix; The expression of the rotating-Z gate is: ； in, represents the rotation-Z gate, Z represents the matrix of the Pauli-Z gate, Represents the unitary evolution operator in exponential form of the rotation-Z gate matrix.

5. According to claim 4, a new quantum intelligent computing method designed for smart cities is characterized in that: The features extracted by the spin-Z gate and the spin-Y gate are converted from quantum state to classical output using the Pauli-Z gate to obtain the feature representation of the image. The matrix of the Pauli-Z gate is: ； in, represents the standard Pauli-Z matrix in quantum computing.

6. According to claim 1, a new quantum intelligent computing method designed for smart cities is characterized by: The fully connected layer uses the LeakyReLU function, expressed as: ； in, Represents the input value of the fully connected layer.

7. According to claim 1, a new quantum intelligent computing method designed for smart cities is characterized by: The output layer includes 10 neurons, corresponding to the 10 categories of FashionMNIST, and the output values ​​of the 10 neurons are converted into probabilities using the Softmax function. , and satisfy the following expression: ； ； The expression of the Softmax function is: ； in, Indicates The output of a neuron, Indicates The output of a neuron; Select the FashionMNIST category with the highest probability value as the classification result output.

Images