A method and related apparatus for target region segmentation of large-scale images

By combining a quantum classification model with a classical segmentation network, false positive segmentation is suppressed, improving the accuracy of target region segmentation in large-scale images and solving the problem of false positive segmentation in existing technologies.

CN121033404BActive Publication Date: 2026-11-17ORIGIN QUANTUM COMPUTING TECH (HEFEI) CO LTD +1
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411187233.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2024-05-21
Filing Date
2024-08-28
Publication Date
2026-11-17
Estimated Expiration
2044-08-28

AI Technical Summary

Technical Problem

Existing techniques are prone to false positives in large-scale image segmentation, especially in mammograms and satellite images, leading to a decrease in segmentation accuracy.

Method used

A quantum classification model is used to classify each sub-image of a large-scale image. The results of the classical segmentation network are combined to perform suppression calculations. False positive segmentation results are suppressed by multiplying the classification value with the class probability with the classical segmentation network result.

Benefits of technology

It effectively suppresses false positive segmentation, improves the accuracy of target region segmentation in large-scale images, and ensures that the segmentation results include only the target region.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121033404B_ABST
    Figure CN121033404B_ABST
Patent Text Reader

Abstract

This application discloses a method and related apparatus for target region segmentation of large-scale images, relating to the field of quantum computing technology. The method includes: classifying each sub-image obtained from the transformation of a large-scale image using a quantum classification model to obtain a classification result, wherein the classification result indicates whether the sub-image includes a target region; performing suppression calculations based on the classification result and the result of a classical segmentation network to obtain a target region segmentation result for the large-scale image, wherein the classical segmentation network result represents the segmentation result for each sub-image, and the target region segmentation result includes only the target region, or includes either the target region or a non-target region. This method can effectively suppress false positive segmentation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application belongs to the field of quantum computing technology, and in particular relates to a method and related apparatus for target region segmentation of large-scale images. Background Technology

[0002] A quantum computer is a physical device that performs high-speed mathematical and logical operations, stores and processes quantum information in accordance with the laws of quantum mechanics. When a device processes and calculates quantum information and runs quantum algorithms, it is a quantum computer. Because of its ability to process mathematical problems more efficiently than ordinary computers—for example, reducing the time to crack RSA keys from hundreds of years to hours—quantum computers have become a key technology under research.

[0003] Image segmentation is the process of dividing an image into multiple regions with independent properties. This process can separate foreground objects from the background or segment different objects in an image into different regions. Image segmentation has wide applications in computer vision and image processing, such as object detection, image editing, and medical image analysis.

[0004] To ensure image clarity, image segmentation for large-scale images such as mammograms and satellite images often employs a block-based segmentation method, dividing the large-scale image into multiple sub-images for processing. However, some sub-images may lack the target region to be segmented. This can easily lead to false positives, where non-target regions in sub-images without a target region are incorrectly segmented, thus affecting the accuracy of target region segmentation in large-scale images. Therefore, a mechanism is needed to suppress false positives. Summary of the Invention

[0005] The purpose of this application is to provide a method and related apparatus for target region segmentation of large-scale images, which aims to suppress false positive segmentation.

[0006] To achieve the above objectives, a first aspect of the embodiments of this application provides a target region segmentation method for large-scale images, the method comprising:

[0007] Each sub-image obtained from the large-scale image transformation is classified using a quantum classification model to obtain a classification result, which is used to indicate whether the sub-image includes the target region.

[0008] Suppression calculations are performed based on the classification results and the results of the classical segmentation network to obtain the target region segmentation results of the large-scale image. The classical segmentation network results are used to represent the segmentation results of each sub-image. The target region segmentation results only include the target region. The segmentation results include the target region or non-target regions.

[0009] In another embodiment of this application, the classification result includes classification values ​​0 and 1 and their corresponding class probabilities; the step of performing suppression calculation based on the classification result and the classical segmentation network result to obtain the target region segmentation result of the large-scale image includes:

[0010] The target region segmentation result of the large-scale image is obtained by multiplying the classification value with the class probability with the result of the classical segmentation network, and fusing the suppressed target region segmentation result of each sub-image obtained by multiplication.

[0011] In another embodiment of this application, the encoder and decoder of the classic segmentation network are arranged in a U-shape.

[0012] In another embodiment of this application, the classical segmentation network output by the U-shaped classical segmentation network includes multiple segmentation results with progressively decreasing scales. Multiplying the classification value with the higher class probability by the classical segmentation network result includes:

[0013] The classification value with the highest category probability is multiplied one by one with the segmentation results of the classic segmentation network, which includes segmentation results of multiple scales decreasing sequentially, to obtain the suppressed target region segmentation result for each sub-image.

[0014] In another embodiment of this application, the quantum classification model includes a dropout layer, a pooling layer, a variable quantum circuit, and an activation layer connected in sequence.

[0015] In another embodiment of this application, the variational module of the variational quantum circuit includes a CNOT gate that acts sequentially on a single quantum bit (Rz, Ry, Rz) and sequentially on adjacent quantum bits (Rz, Ry, Rz) and sequentially on a single quantum bit (Rz, Ry, Rz).

[0016] In another embodiment of this application, the classical segmentation network and the quantum classification model are trained using the controlled variable method and the gradient descent method.

[0017] A second aspect of this application provides a target region segmentation apparatus for large-scale images, the apparatus comprising:

[0018] The classification module is used to classify each sub-image obtained by large-scale image transformation using a quantum classification model, and obtain a classification result, which is used to indicate whether the sub-image includes the target region.

[0019] The calculation module is used to perform suppression calculation based on the classification result and the classical segmentation network result to obtain the target region segmentation result of the large-scale image. The classical segmentation network result is used to represent the segmentation result of each sub-image. The target region segmentation result only includes the target region. The segmentation result includes the target region or non-target region.

[0020] A third aspect of this application provides a storage medium storing a computer program, wherein the computer program is configured to execute the steps of the method described in any of the first aspects above when running.

[0021] A fourth aspect of this application provides an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps of the method described in any of the first aspects above.

[0022] Based on the above technical solution, this application acquires multiple sub-images of the image to be segmented, and performs image classification on each sub-image obtained from the large-scale image transformation using a quantum classification model to obtain classification results. The classification results are used to indicate whether the sub-image includes the target region. The classical segmentation network results are used to represent the segmentation results of each sub-image. The target region segmentation results only include the target region, and the segmentation results include either the target region or non-target regions. Furthermore, suppression calculations are performed based on the classification results and the classical segmentation network results to obtain the target region segmentation results of the large-scale image. Suppression calculations are performed on the classical segmentation network results of sub-images that do not include the target region, thereby eliminating the interference of the segmentation results of non-target regions in sub-images that do not include the target region, thus improving the image segmentation accuracy of the image to be segmented and effectively suppressing false positive segmentation. Attached Figure Description

[0023] Figure 1 This is a network block diagram of a target region segmentation system for a large-scale image, according to an exemplary embodiment.

[0024] Figure 2 This is a flowchart illustrating a target region segmentation method for a large-scale image according to an exemplary embodiment.

[0025] Figure 3 This is a schematic diagram of the structure of a Unet network according to an exemplary embodiment.

[0026] Figure 4 This is a schematic diagram of a variational module of a variational quantum circuit according to an exemplary embodiment.

[0027] Figure 5 This is a schematic diagram illustrating target region segmentation of a sub-image based on a Unet network and a quantum classification model, according to an exemplary embodiment.

[0028] Figure 6 This is a block diagram illustrating a target region segmentation device for a large-scale image according to an exemplary embodiment.

[0029] Figure 7 This is a block diagram illustrating a computer device according to an exemplary embodiment. Detailed Implementation

[0030] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0031] Figure 1 This is a network block diagram of a target region segmentation system for large-scale images provided in an embodiment of this application. The target region segmentation system for large-scale images may include a network 110, a server 120, a wireless device 130, a client 140, storage 150, a classical computing unit 160, a quantum computing unit 170, and may also include additional memory, a classical processor, a quantum processor, and other devices not shown.

[0032] Network 110 is a medium used to provide communication links between various devices and computers connected together within a target region segmentation system for large-scale images, including but not limited to the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. The connection method can be wired, wireless communication links, or fiber optic cables.

[0033] Server 120, wireless device 130, and client 140 are conventional data processing systems that may contain data and application programs or software tools that perform conventional computational processes. Client 140 may be a personal computer or a network computer, so the data may also be provided by server 120. Wireless device 130 may be a smartphone, tablet, laptop, smart wearable device, etc. Storage unit 150 may include database 151, which can be configured to store data such as qubit parameters, quantum logic gate parameters, quantum circuits, and quantum programs.

[0034] The classical computing unit 160 (quantum computing unit 170) may include a classical processor 161 (quantum processor 171) for processing classical data (quantum data) and a memory 162 (memory 172) for storing classical data (quantum data). The classical data (quantum data) may be a boot file, an operating system image, and an application program 163 (application program 173). The application program 163 (application program 173) may be used to implement a quantum algorithm compiled according to the target region segmentation method for large-scale images provided in the embodiments of this application.

[0035] Any data or information stored or generated in the classical computing unit 160 (quantum computing unit 170) can also be configured to be stored or generated in another classical (quantum) processing system in a similar manner, and any application executed therein can also be configured to be executed in another classical (quantum) processing system in a similar manner.

[0036] It should be noted that a true quantum computer has a hybrid structure, which includes at least... Figure 1 The system consists of two main parts: the classical computing unit 160, which is responsible for performing classical calculations and control; and the quantum computing unit 170, which is responsible for running quantum programs to achieve quantum computing.

[0037] The aforementioned classical computing unit 160 and quantum computing unit 170 can be integrated into a single device or distributed across two different devices. For example, a first device including the classical computing unit 160 runs a classical computer operating system, providing quantum application development tools and services, as well as the storage and network services required for quantum applications. Users develop quantum programs using the quantum application development tools and services on the second device, and send these quantum programs to a second device including the quantum computing unit 170 via the network services. The second device runs a quantum computer operating system, which parses and compiles the quantum program's code into instructions that the quantum processor 170 can recognize and execute. The quantum processor 170 then implements the quantum algorithm corresponding to the quantum program based on these instructions.

[0038] The computing units of the classic processor 161 within the classic computing unit 160 are based on CMOS transistors on a silicon chip. These computing units are not limited by time or coherence; that is, they are available at any time without time constraints. Furthermore, the number of such computing units in a silicon chip is sufficient; currently, a single classic processor 161 contains tens of thousands of computing units. Given this sufficient number of computing units and the fixed selectable computing logic of the CMOS transistors (e.g., AND logic), computational performance is achieved by combining a large number of CMOS transistors with a limited set of logic functions during operation.

[0039] In the quantum computing unit 170, the basic computing unit of the quantum processor 171 is the qubit. The input of a qubit is limited by coherence and coherence time; that is, a qubit is limited by its available usage time and is not always readily available. Making full use of qubits within their available usage time is a key challenge in quantum computing. Furthermore, the number of qubits in a quantum computer is one of the representative indicators of its performance. Each qubit performs computational functions through on-demand configured logical functions. Given the limited number of qubits and the diverse logical functions available in quantum computing, such as Hadamard gates (H gates), Pauli-X gates (X gates), Pauli-Y gates (Y gates), Pauli-Z gates (Z gates), X gates, RY gates, RZ gates, CNOT gates, CR gates, iSWAP gates, Tofoli gates, etc., quantum computing requires combining a limited number of qubits with diverse logical function combinations to achieve computational effects.

[0040] Based on these differences, the design of classical logic functions applied to CMOS transistors and the design of quantum logic functions applied to qubits are significantly and fundamentally different. The design of classical logic functions applied to CMOS transistors does not need to consider the individuality of CMOS transistors. For example, the representation of a CMOS transistor in a silicon chip is its individual identifier, location, and usable time of each CMOS transistor. Therefore, classical algorithms composed of classical logic functions only express the operational relationship of the algorithm, not the algorithm's dependence on individual CMOS transistors.

[0041] Quantum logic functions applied to qubits need to consider the individuality of each qubit, such as its position within the quantum chip, its relationship with surrounding qubits, and the duration of its usable time. Therefore, quantum algorithms composed of quantum logic functions not only express the computational relationships within the algorithm but also its dependence on the individual qubits.

[0042] For example:

[0043] Quantum Algorithm 1: H1, H2, CNOT(1,3), H3, CNOT(2,3);

[0044] Quantum Algorithm 2: H1, H2, CNOT(1,2), H3, CNOT(2,3);

[0045] Quantum Algorithm 1 and Quantum Algorithm 2 use a combination of quantum logic gates and Arabic numerals to represent the qubits that function under the quantum logic gates. The Arabic numerals 1, 2, and 3 can represent three qubits Q1, Q2, and Q3 connected in sequence or qubits Q1, Q2, and Q3 connected to each other.

[0046] An exemplary explanation of how quantum algorithms are affected by the coherence time of qubits is as follows:

[0047] Define the execution time of a single-qubit logic gate as t, and the execution time of two single-qubit logic gates operating on adjacent qubits as 2t; then:

[0048] When Q1, Q2, and Q3 are interconnected, the computation of Quantum Algorithm 1 requires 6t, which is divided into 4 time periods. The duration of each time period is t, 2t, t, and 2t, respectively. The operations performed in each time period are: H1, H2; CNOT(1,3); H3; CNOT(2,3);

[0049] The computation of Quantum Algorithm 1 requires 5t, which is divided into 3 time periods. The duration of each time period is t, 2t, and 2t respectively. The operations performed in each time period are: H1, H2, H3; CNOT(1,2); CNOT(2,3);

[0050] When Q1, Q2, and Q3 are connected sequentially, Quantum Algorithm 1 needs to be equivalent to: H1, H2; swap(1,2), CNOT(2,3), swap(1,2); H3; CNOT(2,3). The computation of the equivalent Quantum Algorithm 1 requires 10t, divided into 4 time periods, with each time period requiring durations of t, 6t, t, and 2t respectively. The operations performed in each time period are: H1, H2; swap(1,2), CNOT(2,3), swap(1,2); H3; CNOT(2,3).

[0051] Therefore, the application of quantum logic functions in the design of qubits (including the design of whether qubits are used and the design of the efficiency of each qubit) is key to improving the computational performance of quantum computers and requires special design. This is the unique characteristic of quantum algorithms based on quantum logic functions, and is fundamentally and significantly different from classical algorithms based on classical logic functions. The aforementioned design considerations for qubits are technical problems that ordinary computing devices do not need to consider or address. This application proposes a target region segmentation method and related apparatus for large-scale images to suppress false positive segmentation.

[0052] In most medical image segmentation processes, false positives are unavoidable in images lacking target regions. This is primarily due to background noise residing in shallower layers, leading to false positives. Mammograms and satellite images are typically large, and to ensure image clarity, block-based segmentation is often employed. However, some sub-images may lack the target region required for segmentation. In such cases, false positives are more likely, necessitating a mechanism to suppress them.

[0053] See Figure 2 , Figure 2 This is a flowchart illustrating a target region segmentation method for a large-scale image according to an exemplary embodiment. The method includes the following steps:

[0054] S201. Classify each sub-image obtained from the large-scale image transformation using a quantum classification model to obtain the classification result.

[0055] The classification result is used to indicate whether a sub-image includes the target region.

[0056] In this embodiment of the application, due to the limitations of the image processing capabilities of classical segmentation networks, image segmentation of large-scale images usually requires first converting the large-scale image into multiple smaller-scale sub-images, and then performing image segmentation on each sub-image. Each sub-image may contain the target region in the large-scale image, or it may only contain the non-target region in the large-scale image without containing the target region in the large-scale image.

[0057] For example, for a large-scale image P, the large-scale image P can be pre-converted into four sub-images, sub-image A1, sub-image A2, sub-image A3 and sub-image A4. Sub-images A1, A2, A3 and A4 can be stitched together to obtain the large-scale image P. The large-scale image P includes the target region. Any one of the sub-images A1, A2, A3 and A4 may include the target region or a part of the target region.

[0058] The quantum classification model provided in this application embodiment can classify each sub-image, and the classification result indicates whether the sub-image includes the target region.

[0059] S202. Based on the classification results and the results of the classical segmentation network, suppression calculation is performed to obtain the target region segmentation results of the large-scale image.

[0060] The classical segmentation network result is used to represent the segmentation result of each sub-image, and the target region segmentation result includes only the target region. The segmentation result includes either the target region or a non-target region.

[0061] In this embodiment, the classical segmentation network can segment each sub-image to obtain the classical segmentation network result. The classification result output by the quantum classification model can be used to perform suppression calculation with the classical segmentation network result to suppress the segmentation result of non-target regions in the classical segmentation network result of each sub-image.

[0062] In one implementation, the classic segmentation network in the above embodiments can be a Unet network and a transformer network.

[0063] This application acquires multiple sub-images of an image to be segmented, and performs image classification on each sub-image obtained from the large-scale image transformation using a quantum classification model to obtain classification results. The classification results indicate whether a sub-image includes a target region. The classical segmentation network results represent the segmentation results of each sub-image, with the target region segmentation results including only the target region or non-target regions. Furthermore, suppression calculations are performed based on the classification results and the classical segmentation network results to obtain the target region segmentation results of the large-scale image. Suppression calculations are also performed on the classical segmentation network results of sub-images that do not include the target region, thereby eliminating interference from the segmentation results of non-target regions in sub-images that do not include the target region. This improves the image segmentation accuracy of the image to be segmented and effectively suppresses false positive segmentation.

[0064] In another embodiment of this application, the classification result includes classification values ​​0 and 1 and their corresponding category probabilities; S202 in the above embodiment, performing suppression calculation based on the classification result and the classical segmentation network result to obtain the target region segmentation result of the large-scale image, can be specifically implemented as follows:

[0065] The target region segmentation result of the large-scale image is obtained by multiplying the classification value with the class probability with the result of the classical segmentation network, and fusing the suppressed target region segmentation result of each sub-image obtained by multiplication.

[0066] In this classification, a classification value of 0 indicates that the sub-image does not contain the target region, and a classification value of 1 indicates that the sub-image includes the target region. The classification result includes classification value 0 and classification value 1, as well as the classification probability corresponding to classification value 0 and the classification probability corresponding to classification value 1.

[0067] For example, if the classification result of sub-image A1 includes a classification probability of 0.6 for classification value 0 and 0.4 for classification value 1, and the classification probability of 0.6 for classification value 0 is greater than the classification probability of 0.4 for classification value 1, then sub-image A1 does not include the target region. Therefore, classification value 0 can be multiplied by the classical segmentation network result of sub-image A1. However, if the classical segmentation network result of sub-image A1 includes non-target regions, meaning the classical segmentation network produced false positives in the image segmentation of sub-image A1, multiplying classification value 0 by the classical segmentation network result of sub-image A1 yields a suppressed target region segmentation result for each sub-image, thus suppressing the false positive segmentation result of sub-image A1.

[0068] Similarly, following the method in example A1 above, the classification results of sub-images A2, A3, and A4 are multiplied with the results of the classic segmentation network. The suppressed target region segmentation results of each sub-image obtained by multiplication are then fused to obtain the target region segmentation result of the large-scale image P.

[0069] In another embodiment of this application, the encoder and decoder of the classic segmentation network in the above embodiments form a U-shaped structure. The Unet network with a U-shaped structure will be used as an example for illustration.

[0070] In this embodiment of the application, the classic segmentation network result output by the Unet network includes multiple segmentation results with progressively decreasing scales. Based on the Unet network, the target region of each sub-image is segmented by multiple progressively decreasing mask segments to obtain the classic segmentation network result of each sub-image.

[0071] The Unet network is a deep learning architecture for image segmentation tasks, designed to solve problems in medical image segmentation.

[0072] like Figure 3 As shown, Figure 3 This is a schematic diagram of the structure of a Unet network provided in an embodiment of this application. The Unet network architecture consists of a symmetrical encoder and decoder. The encoder part is similar to a traditional convolutional neural network, which gradually downsamples the input image through multiple convolution and pooling operations and extracts the high-level features of the image. The decoder part gradually restores the features extracted by the encoder to the original size of the input image through upsampling and convolution operations and generates the segmentation result.

[0073] The UNet network is characterized by the introduction of skip connections in the decoder section, which connects feature maps of different scales in the encoder with corresponding feature maps of the decoder. This allows for the fusion of low-level and high-level features, enabling the network to utilize both global and local information for segmentation, thus improving segmentation accuracy.

[0074] The UNet network has been widely used in medical image segmentation, especially excelling in handling small sample sizes and class imbalance. It has become one of the classic network architectures in image segmentation, widely applied to tasks such as medical image segmentation, road segmentation, and face segmentation.

[0075] In another embodiment of this application, the quantum classification model includes a dropout layer, a pooling layer, a variable quantum circuit, and an activation layer connected in sequence.

[0076] Dropout layers are a technique used to prevent overfitting in neural networks. During training, the Dropout layer randomly sets the output of a subset of neurons to 0. These neurons do not participate in forward or backward propagation calculations during that training iteration. This forces the network to learn multiple independent feature representations, increasing the robustness of the quantum classification model, reducing its dependence on certain specific inputs, and thus improving its generalization ability.

[0077] The Dropout layer works by randomly deactivating a portion of neurons to reduce the number of parameters in the neural network, thereby reducing model complexity. During testing, the neurons in the Dropout layer are not deactivated, but their outputs are multiplied by a retention rate, typically 0.5 or 0.2, to compensate for the output of neurons deactivated during training, ensuring that the expected output of each neuron during testing is consistent with that during training.

[0078] Dropout layers can be applied to various types of neural networks, such as fully connected layers, convolutional layers, and recurrent neural networks. In practice, dropout layers are typically placed after fully connected layers to significantly improve model performance and generalization ability.

[0079] Pooling layers are a common type of layer in convolutional neural networks (CNNs) used to reduce the spatial size of feature maps, thereby reducing the number of parameters and computational cost, while extracting the position invariance of features.

[0080] Pooling layers divide the feature map into non-overlapping regions (such as 2x2 or 3x3 windows) and aggregate the features within each region, typically by taking the maximum value (max pooling) or calculating the average value (average pooling). Pooling reduces the size of the feature map, preserving key features and reducing redundant information.

[0081] The pooling layer serves several purposes:

[0082] Reducing the spatial size of feature maps lowers model complexity, reduces computational load and the number of parameters, and accelerates training and inference.

[0083] Position invariance of extracted features means that when the input image is transformed by translation, rotation, or other methods, pooling operations can maintain the invariance of the extracted features.

[0084] Increase the robustness of the model by reducing the spatial resolution of the features, thereby reducing the impact of noise and subtle changes in the input image on the model.

[0085] Pooling layers are typically used alternately with convolutional layers. Convolutional layers extract features, while pooling layers reduce the size of the feature map and the number of parameters. Common pooling layer parameters include window size, stride, and padding. Common pooling operations include max pooling and average pooling. Max pooling is often used to extract key features, while average pooling is often used to preserve overall features.

[0086] The activation layer is implemented using the softmax function, a commonly used activation function that transforms a multi-dimensional real-number vector into a probability distribution. It takes the power of each element in the vector and then normalizes it so that the sum of all elements is 1. The softmax function is frequently used in multi-class classification problems, as it transforms the model's output into a probability distribution, facilitating classification. In deep learning, the softmax function is often used in the output layer to convert the neural network's output into probabilities. The softmax function's characteristic is that it amplifies large values ​​and shrinks small values ​​in the vector, thus enhancing the model's ability to predict high-probability classes. Furthermore, since the sum of the softmax function's outputs is 1, it can also be seen as a normalization operation, allowing the output to be interpreted as a probability.

[0087] Optionally, the activation layer can also be implemented based on the sigmoid function. The sigmoid function is frequently used in machine learning and neural networks to map the result of a linear transformation to a probability value, and is commonly used as an activation function in binary classification problems. Its output value can be interpreted as a probability estimate of the input value in a certain category, and therefore can be used to calculate the loss function and perform probability prediction when training a classification model.

[0088] In another embodiment of this application, such as Figure 4 As shown, Figure 4 This is a schematic diagram of a variational module for a variational quantum circuit, exemplarily shown in an embodiment of this application. The variational module of the aforementioned variational quantum circuit includes a CNOT gate that sequentially acts on a single qubit (Rz, Ry, Rz), sequentially acts on adjacent qubits, and sequentially acts on a single qubit (Rz, Ry, Rz). Figure 4 A measurement module for measuring two qubits is also shown. Figure 4 The quantum coding module is not shown.

[0089] Variational quantum circuits typically include a quantum encoding module, a variational module, and a measurement module. The quantum encoding module encodes classical feature data onto qubits, the variational module evolves the encoded quantum states to achieve feature classification, the rotational logic gates constituting the variational module include trainable parameters, and the measurement layer measures the evolved quantum states to obtain the classification results.

[0090] In another embodiment of this application, the classical segmentation network output by the aforementioned U-shaped classical segmentation network includes multiple segmentation results with progressively decreasing scales. Multiplying the classification value with the higher class probability by the classical segmentation network result includes:

[0091] The classification value with the highest category probability is multiplied one by one with the segmentation results of the classic segmentation network, which includes segmentation results of multiple scales decreasing sequentially, to obtain the suppressed target region segmentation result for each sub-image.

[0092] like Figure 5 As shown, Figure 5 This is a schematic diagram illustrating target region segmentation of a sub-image based on a Unet network and a quantum classification model, as provided in an embodiment of this application. Figure 5 The quantum classification model consists of a Dropout layer, a pooling layer, a quantum layer, and an activation layer (Softmax) connected in sequence. The output of quantum classification includes classification values ​​0 and 1, as well as classification probabilities of 0.6 for classification value 0 and 0.4 for classification value 1. The classification results are processed by the argmax function to obtain the classification value 0 with the higher classification probability.

[0093] The Unet network performs mask segmentation on the target region of the sub-image at multiple scales that decrease sequentially, resulting in the classical segmentation network result of the sub-image. The classical segmentation network result includes the segmentation results at multiple scales that decrease sequentially. The classification value 0, which has a high classification probability, is multiplied one by one with the segmentation results at multiple scales that decrease sequentially, to obtain the suppressed target region segmentation result of each sub-image.

[0094] As can be seen from the above embodiments, since classification value 0 is used to indicate that the sub-image does not include the target region and classification value 1 is used to indicate that the sub-image includes the target region, the segmentation results of multiple scales are multiplied one by one based on classification value 0. When the segmentation result includes non-target regions but does not include the target region, the target region segmentation result of each sub-image after suppression is 0, which can suppress false positive segmentation.

[0095] Multiplying the segmentation results based on the classification value 1 and the segmentation results of multiple scales decreasing one by one, when the segmentation result includes the target region, the target region segmentation result of each sub-image after suppression is still the segmentation result that includes the target region, without affecting the target region segmentation.

[0096] In another embodiment of this application, the classical segmentation network and the quantum classification model described above are trained using the control variable method and the gradient descent method.

[0097] The specific training steps are as follows:

[0098] Step 1: Initialize the parameters of the classical segmentation network and the quantum classification model.

[0099] Step 2: Keeping the parameters of the quantum classification model unchanged, train the classical segmentation network and update the parameters of the classical segmentation network based on the gradient descent method.

[0100] Step 3: Keeping the parameters of the classical segmentation network unchanged, train the quantum classification model and update the parameters of the quantum classification model based on the gradient descent method.

[0101] The embodiments of this application do not specifically limit the execution order of steps two and three. Step two can be executed first and then step three, or step three can be executed first and then step two.

[0102] As can be seen from the above embodiments, this application embodiment trains a classical segmentation network and a quantum classification model. In the application stage, the quantum classification model can classify each sub-image of a large-scale image. The classification result of each sub-image can indicate whether the sub-image includes a target region. The classical segmentation network can segment each sub-image of the large-scale image. The classical segmentation network result is used to represent the segmentation result of each sub-image. The segmentation result can include a target region or a non-target region. Based on the classification result of each sub-image and the classical segmentation network result of each sub-image, suppression calculation can be performed to suppress the segmentation result including non-target regions. Then, the suppressed target region segmentation result of each sub-image can be fused to obtain the target region segmentation result of the large-scale image, thus achieving the suppression of false positive segmentation.

[0103] Based on the same inventive concept, embodiments of this application provide a target region segmentation device for large-scale images, such as... Figure 6 As shown, the device includes:

[0104] The classification module 601 is used to classify each sub-image obtained by large-scale image transformation using a quantum classification model, and obtain a classification result, wherein the classification result is used to indicate whether the sub-image includes the target region.

[0105] The calculation module 602 is used to perform suppression calculation based on the classification result and the classical segmentation network result to obtain the target region segmentation result of the large-scale image. The classical segmentation network result is used to represent the segmentation result of each sub-image. The target region segmentation result only includes the target region. The segmentation result includes the target region or non-target region.

[0106] The specific functions and effects of the target region segmentation device for large-scale images can be explained by referring to other embodiments in this specification, and will not be repeated here. Each module in the target region segmentation device for large-scale images can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0107] Please see Figure 7 This specification also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the target region segmentation method for large-scale images in any of the above embodiments. Please refer to... Figure 7 The computer device can be a classical computer or a quantum computer.

[0108] This specification also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a computer, causes the computer to perform the target region segmentation method for large-scale images in any of the above embodiments.

[0109] This specification also provides a computer program product containing instructions that, when executed by a computer, cause the computer to perform the target region segmentation method for large-scale images in any of the above embodiments.

[0110] It is understood that the specific examples in this specification are only intended to help those skilled in the art better understand the implementation methods described herein, and are not intended to limit the scope of this application.

[0111] It is understood that in the various embodiments of this specification, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not limit the implementation process of the embodiments of this specification in any way.

[0112] It is understood that the various implementation methods described in this specification can be implemented individually or in combination, and the implementation methods in this specification are not limited in this respect.

[0113] Unless otherwise stated, all technical and scientific terms used in the embodiments of this specification have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this specification. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items. The singular forms "a," "the," and "the" as used in the embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0114] It is understood that the processor in the embodiments of this specification can be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method embodiments can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this specification. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this specification can be directly implemented by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. 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 methods.

[0115] It is understood that the memory in the embodiments of this specification may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may be random access memory (RAM). It should be noted that the memory in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0116] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed 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 specification.

[0117] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the aforementioned method implementations, and will not be repeated here.

[0118] In the several embodiments provided in this specification, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0119] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0120] In addition, the functional units in the various embodiments of this specification can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0121] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of this specification, in essence, or the parts that contribute to the prior art, or parts of the technical solutions, can be embodied in the form of software products. These computer software products are stored in a storage medium and include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this specification. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0122] The above description is merely a specific embodiment of this specification, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this specification should be included within the scope of protection of this specification. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for segmenting a target region of a large-scale image, characterized in that, The method includes: Each sub-image obtained from the large-scale image transformation is classified using a quantum classification model to obtain a classification result. The classification result indicates whether the sub-image contains the target region. The classification result includes classification values ​​0 and 1 and their corresponding class probabilities. The quantum classification model includes a dropout layer, a pooling layer, a variable quantum circuit, and an activation layer connected in sequence. The variational module of the variable quantum circuit includes Rz, Ry, and Rz gates acting sequentially on a single qubit, CNOT gates acting sequentially on adjacent qubits, and Rz, Ry, and Rz gates acting sequentially on a single qubit. The classification value with a high category probability is multiplied with the result of the classical segmentation network, and the suppressed target region segmentation result of each sub-image obtained by multiplication is fused to achieve suppression calculation based on the classification result and the classical segmentation network result, so as to obtain the target region segmentation result of the large-scale image. The classical segmentation network result is used to represent the segmentation result of each sub-image, and the target region segmentation result only includes the target region. The segmentation result includes the target region or non-target region.

2. The method of claim 1, wherein, The encoder and decoder of the classic segmentation network have a U-shaped structure.

3. The method of claim 2, wherein, The classical segmentation network output by the U-shaped structure includes segmentation results at multiple scales decreasing sequentially. Multiplying the classification value with the highest class probability by the classical segmentation network result includes: The classification value with the highest category probability is multiplied one by one with the segmentation results of the classic segmentation network, which includes segmentation results of multiple scales decreasing sequentially, to obtain the target region segmentation result of each sub-image after suppression.

4. The method according to any of claims 1 to 3, characterized in that, The classical segmentation network and the quantum classification model are trained using the controlled variable method and the gradient descent method.

5. An apparatus for segmenting a target region of a large-scale image, characterized by comprising: The device includes: A classification module is used to classify each sub-image obtained from the large-scale image transformation using a quantum classification model, and obtain a classification result. The classification result indicates whether the sub-image contains the target region. The classification result includes classification values ​​0 and 1 and their corresponding class probabilities. The quantum classification model includes a dropout layer, a pooling layer, a variable quantum circuit, and an activation layer connected in sequence. The variational module of the variable quantum circuit includes Rz, Ry, and Rz gates acting sequentially on a single qubit, CNOT gates acting sequentially on adjacent qubits, and Rz, Ry, and Rz gates acting sequentially on a single qubit. The calculation module is used to multiply the classification value with the high category probability with the result of the classical segmentation network, and to fuse the suppressed target region segmentation result of each sub-image obtained by multiplication. This enables the suppression calculation based on the classification result and the classical segmentation network result to obtain the target region segmentation result of the large-scale image. The classical segmentation network result is used to represent the segmentation result of each sub-image. The target region segmentation result only includes the target region. The segmentation result includes the target region or non-target region.

6. A storage medium, characterized by The storage medium stores a computer program, wherein the computer program is configured to execute the method described in any one of claims 1 to 4 when it is run.

7. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method as described in any one of claims 1 to 4.

Citation Information

Patent Citations

  • Quantum multi-target clustering-based remote sensing image segmentation method

    CN102567963A

  • Segmentation strengthening method for damage detection image of aerospace composite material

    CN112819775A