Quantum-inspired method and system for super-resolution reconstruction of plant microtubule images

Abstract

The present application belongs to the technical field of image processing, and particularly relates to a plant microtubule image super-resolution reconstruction method and system based on quantum inspiration. The method of the present application takes IRSRMamba model as a baseline model, constructs a feature fusion module and a quantum superposition embedding module in the shallow feature extraction layer thereof, the feature fusion module comprising a shallow quantum feature refinement module and a wavelet transform; the Mamba deep feature extraction backbone is composed of a plurality of residual state space groups, each residual state space group comprising a plurality of quantum state space blocks; the quantum state space block comprises a two-dimensional selective scanning, a quantum inspiration state space module, a deep quantum feature refinement module and the like; and the image reconstruction layer comprises a quantum measurement reconstruction module. The present application effectively improves the super-resolution reconstruction effect of the slender plant microtubule image with topological continuity, and improves the accuracy and robustness of the filamentous feature extraction in the field of biological imaging and related industrial detection.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, specifically relating to a quantum-inspired super-resolution reconstruction method and system for plant microtubule images. Background Technology

[0002] Microtubules, as a core component of the eukaryotic cytoskeleton system, are fundamental subcellular units that maintain the integrity of cellular spatial structure and perform complex physiological functions. High-precision imaging and feature extraction of microtubule structures are crucial for a deeper understanding of the microscopic mechanisms of life activities. However, limited by the physical diffraction limit of optical microscopy, raw fluorescence microscopy images often suffer from low contrast and insufficient spatial resolution, resulting in the appearance of delicate microtubule structures as blurry, diffuse spots or discontinuous line segments. To overcome the diffraction limit and obtain high-resolution images, image super-resolution (SR) technology has become a research hotspot in this field.

[0003] Traditional image super-resolution methods primarily focus on improving the visual representation of areas with clear textures in natural scenes. However, research on microtubule structures, which possess high topological complexity and densely overlapping branches, is relatively weak. In recent years, deep learning models, such as Convolutional Neural Networks (CNNs) and Transformers, have made significant progress in feature extraction. However, CNNs are limited by their local receptive fields, making it difficult to maintain the overall consistency of long-range continuous microtubule structures; Transformers, on the other hand, suffer from computational complexity that increases quadratically with image resolution, making them difficult to directly apply to large-scale microscopic image processing. Recently, large-scale model architectures, such as Mamba (State-Space Model, SSM), have demonstrated powerful long-range dependency modeling capabilities within linear complexity, providing a new technical path for microscopic image super-resolution tasks.

[0004] However, even advanced large-scale models like Mamba primarily employ feature extraction mechanisms geared towards the regionalized identification and reconstruction of general targets. When directly applied to microtubule image super-resolution reconstruction tasks, they are prone to significant morphological perception biases, making it difficult to accurately extract the topological skeleton of microtubules. They often misclassify overlapping microtubule structures as localized block artifacts or background noise. Furthermore, current mainstream super-resolution models mostly employ pixel-level residual mapping strategies, and pixel-level predictions are easily affected by background fluorescence spots, thus introducing numerous spurious branches and structural errors. Summary of the Invention

[0005] The purpose of this invention is to provide a quantum-inspired super-resolution reconstruction method and system for plant microtubule images. By introducing a quantum-inspired mechanism, the feature embedding method of existing large visual models represented by Mamba is improved, and the backbone network architecture of the IRSRMamba (infrared image super-resolution based on wavelet transform feature modulation model of Mamba) model is optimized. At the same time, Hessian regularization constraints are introduced to improve the geometric continuity constraints of microtubules from the pixel level to the topological level, effectively overcoming the structural breakage problem that is prone to occur in the super-resolution process of microtubule images in traditional methods, and significantly improving the accuracy and reliability of super-resolution reconstruction of biological microtubule images.

[0006] This invention is achieved through the following technical solution: A quantum-inspired super-resolution reconstruction method for plant microtubule images, comprising the following steps: A super-resolution reconstruction network for plant microtubule images was constructed with the IRSRMamba model as the backbone network. The backbone network consists of a shallow feature extraction layer, a Mamba deep feature extraction backbone, and an image reconstruction layer. A low-resolution image of plant microtubules is input into the shallow feature extraction layer, which includes a feature fusion module and a quantum superposition embedding module. The feature fusion module is used to extract multi-scale features of the input image and enhance cross-channel entanglement through a shallow quantum feature refinement module to output strongly coupled high-frequency features. The quantum superposition embedding module is used to map the high-frequency features to a complex Hilbert space to generate quantum superposition state features. The quantum superposition state features are input into the Mamba deep feature extraction backbone, which includes an image patch embedding layer, a residual state space group, and an image patch inverse embedding layer. Each residual state space group contains several quantum state space blocks. The quantum state space blocks first perform normalization, two-dimensional selective scanning, and quantum-inspired state space module processing on the input one-dimensional feature sequence in sequence. The quantum-inspired state space module is used to adaptively learn control parameters and correct and lock the phase of the input features. The corrected and locked features are then connected with the residual of the input one-dimensional feature sequence and normalized again. Then, linear entanglement cascade is performed through the deep quantum feature refinement module to achieve quantum state entanglement fusion in the channel dimension. Finally, weight calibration is performed through the channel attention module. The features output from the Mamba deep feature extraction backbone are input into the image reconstruction layer. The image reconstruction layer includes a quantum measurement reconstruction module, which is used to simulate the wave function collapse process and output a super-resolution image of plant microtubules by calculating the collapse probability and the expected value of the eigenvalue.

[0007] Further optimized, the quantum superposition embedding module inputs the input features into two parallel convolutional layers respectively. The absolute value of the output of one convolutional layer is taken to obtain the amplitude, which represents the low-frequency skeleton brightness of plant microtubules. The output of the other convolutional layer is mapped to the arctangent function to obtain the phase, which represents the high-frequency overlap direction. Then, the amplitude and phase are superimposed and recombined in complex Hilbert space using Euler's formula to obtain the quantum superposition state features, which serve as the core data sent to the Mamba deep feature extraction backbone.

[0008] In a further preferred embodiment, the quantum-inspired state-space module adaptively learns a control parameter from the current features through an activation function, and then uses the control parameter as a control signal to drive a controlled phase gate, thereby locking and correcting the phase of the master quantum state without changing the feature amplitude, thus preventing the breakage and information loss of the long microtube lines during the Mamba scanning process.

[0009] Further preferably, the shallow quantum feature refinement module and the deep quantum feature refinement module have the same structure. Both extract the independent phase of each channel through a single-bit rotating gate, and then introduce a controlled entanglement gate for linear entanglement cascade. In the shallow layer, it is used to enable cross-channel interaction of high-frequency edge features in different directions to remove background noise. In the deep layer, it is used to enable semantic-level entanglement fusion of features from different channels in high-dimensional space.

[0010] In a further preferred embodiment, the quantum measurement reconstruction module calculates two sets of matrices through parallel convolution. One set is constrained to the range of 0-1 by the Sigmoid function to obtain the collapse probability, while the other set remains unchanged as the eigenvalue. Then, the mathematical expectation of the collapse probability and the eigenvalue is calculated to drive the quantum state collapse and output a super-resolution image.

[0011] Further optimization involves employing the Hessian regularization loss function during the training phase of the plant microtubule image super-resolution reconstruction network. This elevates the geometric continuity constraint of plant microtubules from the pixel level to the topological level, guiding each module to learn feature distributions that conform to the physical properties of microtubules.

[0012] Further optimized, the Mamba deep feature extraction backbone has 6 layers, each consisting of an image patch embedding layer, a residual state space group, and an image patch inverse embedding layer. The residual state space group of each layer contains several quantum state space blocks. After the quantum superposition state features are input to each layer of the Mamba deep feature extraction backbone, they are first flattened into a one-dimensional feature sequence by the image patch embedding layer. After processing by multiple quantum state space blocks, they are restored into a two-dimensional feature map by the image patch inverse embedding layer. Then, they are sequentially processed by convolution and residual connections to complete the complete process of one layer of the Mamba deep feature extraction backbone. Finally, after processing by all 6 layers of the Mamba deep feature extraction backbone, the final output feature map of the Mamba deep feature extraction backbone is obtained.

[0013] This invention also provides a quantum-inspired super-resolution reconstruction system for plant microtubule images, comprising: Image input module, used to receive images of plant microtubules; The super-resolution module has a built-in super-resolution reconstruction network for plant microtubule images, which is used to execute the quantum-inspired super-resolution reconstruction method for plant microtubule images to perform super-resolution reconstruction of plant microtubule images. The output module is used to output the super-resolution image of plant microtubules obtained from super-resolution reconstruction.

[0014] The present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the quantum-inspired super-resolution reconstruction method for plant microtubule images.

[0015] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the quantum-inspired super-resolution reconstruction method for plant microtubule images as described above.

[0016] The present invention has the following advantages: 1. This invention overcomes the limitation of traditional super-resolution reconstruction models that only process pixels in the real number domain. Specifically, for the morphological features of plant microtubules, a shallow quantum feature refinement module is introduced within the feature fusion module of the shallow feature extraction layer based on the IRSRMamba model to remove redundant high-frequency features. Simultaneously, a quantum superposition embedding module is introduced, using Euler's formula to map the microtubule features in the input real number domain to a complex quantum superposition state space, decomposing the pixel state into amplitude features representing the low-frequency skeleton brightness of the microtubule and phase features representing high-frequency edge direction information. By combining the feature fusion module and the quantum superposition embedding module, a dual refined representation of plant microtubule features in both the frequency domain and the complex Hilbert space is achieved, significantly enhancing the ability to identify microscopic overlapping and intersecting regions, thereby effectively improving the network's decoupling ability for overlapping microtubules.

[0017] 2. To address the issue of morphological perception bias in traditional large-scale models when processing slender structures, this invention adds a quantum-inspired state-space module to the Mamba deep feature extraction backbone network. Immediately following selective scanning (SS2D), the extracted spatial state sequence undergoes dynamic phase modulation, effectively suppressing feature diffusion of long-range microtube topological information during deep network transmission. Simultaneously, a deep quantum feature refinement module is introduced into the feature refinement and residual feedback paths. Utilizing qubit rotation gates and controlled entanglement gates, high-frequency textures of microtubes are coherently enhanced in the frequency domain, strengthening the logical entanglement between different feature channels. Combined with the Hessian regularization loss function introduced during model training, each module learns feature distribution patterns consistent with the physical properties of microtubes. This comprehensively solves the common problems of microtube line breakage and adhesion in super-resolution reconstruction, improving the biophysical plausibility of the reconstruction results.

[0018] 3. This invention introduces a quantum measurement reconstruction module into the image reconstruction layer, simulating the measurement collapse process in quantum mechanics. It outputs the final super-resolution image by calculating the probability expectation of eigenvectors on the projection basis. This probabilistic collapse measurement mechanism effectively suppresses background noise and anomalous artifacts caused by deep network amplification while improving image resolution (the collapse probability of anomalous background noise approaches zero, and it is naturally filtered out during expectation calculation), thus achieving high topological fidelity while maintaining visual clarity.

[0019] 4. This invention effectively improves the super-resolution reconstruction effect of images of slender plant microtubules with topological continuity, and significantly enhances the accuracy and robustness of filament feature extraction in biological imaging and related industrial detection fields. Attached Figure Description

[0020] Figure 1 This is a network architecture diagram of a quantum-inspired super-resolution reconstruction method for plant microtubule images provided by the present invention; Figure 2 This is a schematic diagram of a quantum superposition embedding module; Figure 3 This is a schematic diagram of the quantum feature refinement module; Figure 4 This is a schematic diagram of a quantum-inspired state-space module; Figure 5 This is a schematic diagram of the quantum measurement reconstruction module; Figure 6 These are the original high-resolution plant microtubule images and the corresponding original low-resolution plant microtubule images; Figure 7 This is a super-resolution image of plant microtubules obtained by the present invention; Figure 8This image shows a partial comparison between a super-resolution image of plant microtubules obtained by this invention and a partial comparison between a raw low-resolution image of plant microtubules. Detailed Implementation

[0021] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0022] Reference Figures 1-5 This embodiment provides a quantum-inspired super-resolution reconstruction method for plant microtubule images. A super-resolution reconstruction network for plant microtubule images is constructed with the IRSRMamba model as the backbone network. The backbone network consists of three layers: a shallow feature extraction layer, a Mamba deep feature extraction backbone, and an image reconstruction layer. The first layer, the shallow feature extraction layer, comprises a feature fusion module (WTFM) and a quantum superposition embedding module (QSE). The feature fusion module includes a shallow quantum feature refinement module and wavelet transform. The core unit of the second layer, the Mamba deep feature extraction backbone, is a 6-layer residual state space group (RSSG). Each residual state space group (RSSG) stacks N quantum state space blocks (VSSBlocks). Each quantum state space block (VSS Block) includes a normalization layer, two-dimensional selective scanning (SS2D), a quantum-inspired state space module (QSSM), a deep quantum feature refinement module, and a channel attention module (CAB). The third layer, the image reconstruction layer, applies a quantum measurement reconstruction module (QMR).

[0023] The input low-resolution plant microtubule image enters the Feature Fusion Module (WTFM). First, wavelet transform (WT) extracts the size-scale features of the low-resolution plant microtubule image and obtains high-frequency subband features. Then, the high-frequency subband features enter the shallow quantum feature refinement module, which uses quantum single-qubit rotation and multi-qubit controlled entanglement to transform the multiple high-frequency subband features obtained from the wavelet transform into strongly coupled high-frequency features. Then, it reaches the quantum superposition embedding module, which extracts the amplitude and phase of the high-frequency features and maps them to complex Hilbert space using Euler's formula to obtain high-dimensional quantum superposition state features. The Mamba deep feature extraction backbone has six layers. Each layer consists of an image patch embedding layer, a residual state space group, and an image patch inverse embedding layer. The core module of the residual state space group is the quantum state space block (VSS Block). When high-dimensional quantum superposition features enter each layer of the Mamba deep feature extraction backbone, they are first flattened into a one-dimensional feature sequence (1D sequence) by the image patch embedding layer, and then enter the quantum state space block (VSS Block). Within this quantum state space block, the one-dimensional feature sequence passes through a normalization layer and a two-dimensional selective scan (SS2D) to obtain a one-dimensional implicit feature that completes a global long sequence scan but is prone to local phase dissipation. This feature then enters the quantum-inspired state space module (QSSM). The QSSM adaptively extracts a control parameter from the feature environment and uses it as a controlled phase gate to precisely correct and lock the phase angle of the one-dimensional feature sequence. The corrected and locked feature is then concatenated with the residual of the input one-dimensional feature sequence to restore the feature space attributes to two-dimensional (2D). After passing through a normalization layer, it reaches the deep quantum feature refinement module. In the deep quantum feature refinement module, linear entanglement cascades are performed, allowing the spatial features scanned by Mamba to undergo complete quantum state entanglement fusion in the channel dimension. This is then flattened into a 1D feature tensor. The strongly coupled spatial features reach the channel attention module (CAB) for final weight calibration, thus completing the evolution of a quantum state space block (VSS Block). This process continues through multiple quantum state space blocks (VSS Blocks). After the block evolution, the image patch is restored to a two-dimensional feature map by the PatchUnEmbedding layer. Finally, the dimension adjustment and detail injection are completed by the tail convolution and residual connection. After processing by 6 layers of Mamba deep feature extraction backbone, the final output feature map of Mamba deep feature extraction backbone is residually connected with the output of quantum superposition embedding module. Then, the spatial dimension of the pixel is enlarged by pixel rearrangement upsampling. The enlarged high-resolution and high-dimensional implicit quantum state features enter the quantum measurement reconstruction module (QMR). The quantum measurement reconstruction module calculates the collapse probability of the quantum state feature at this point belonging to the real microtube and calculates the corresponding classical feature value. Finally, the super-resolution image of plant microtubes is obtained in the form of probability expectation value.

[0024] To address the challenge of effectively separating the low-frequency skeleton from the overlapping high-frequency edges in plant microtubule images, while the original backbone network employs standard convolution operations in the shallow feature extraction layer, offering computational simplicity, this conventional convolution struggles to handle plant microtubule images with mixed multi-frequency information, easily leading to structural blurring in overlapping microtubule regions. To resolve this issue, this embodiment incorporates a quantum superposition embedding (QSE) module in the shallow feature extraction stage, such as... Figure 2As shown, the quantum superposition embedding module takes the high-frequency features output by the feature fusion module and passes them through two parallel convolutions, then performs absolute value extraction and mapping to... The amplitude and phase of the features are obtained, representing the low-frequency energy of the plant microtubule skeleton and the high-frequency direction of the edge, respectively. Then, the deterministic classical features are mapped to high-dimensional quantum superposition features using Euler's formula, thus preserving the topological information of the microtubules without loss in complex Hilbert space. The quantum superposition embedding module processing can be represented by formulas (1) and (2): (1) (2) in, This represents the high-frequency features output by the feature fusion module; , Parallel convolutional layers and convolutional layers Learnable weights; To represent the hyperbolic tangent function, restricting its value to... between; Indicates amplitude; Indicates phase; Indicates the characteristics of quantum superposition states; Represents the initial quantum state; and This represents a revolving door, rotating about the y-axis and z-axis respectively. Figure 2 In the middle, H represents the Hadamard gate, which is used to construct the initial state of quantum superposition.

[0025] To address the issues of loosely discrete high-frequency edges in shallow frequency domain images and a lack of semantic coherence in deep spatial dimensions, although the original backbone network employs simple feature stitching in shallow feature fusion and channel attention for weight labeling in deep layers, this linear stitching and scalar scaling benchmark cannot establish strong information interaction in high-dimensional space. This results in insufficient sensitivity of the network to plant microtube intersections and incomplete semantic fusion. To resolve this problem, this embodiment inserts quantum feature refinement modules after the wavelet transform in the feature fusion module and in the spatial reconstruction stage of the quantum state space block (VSSBlock). These are the shallow quantum feature refinement module and the deep quantum feature refinement module, which have the same structure, as shown below. Figure 3As shown, the quantum feature refinement module treats the input feature channels as independent qubits. First, it extracts the independent phase of each channel through a single-qubit rotation gate, and then introduces a controlled entanglement gate for linear entanglement cascade. In the shallow layer, this mechanism enables high-frequency edge features in different directions to interact across channels, thereby eliminating independent background noise. In the deep layer, it treats deep features as control bits and target bits and performs thorough semantic-level entanglement fusion in high-dimensional space. The input feature processing of the quantum feature refinement module can be represented by formula (3): (3) in, Represents the phase of the nth channel. For the input of the nth channel of the quantum feature refinement module, in the shallow layer The nth high-frequency sub-band feature obtained from wavelet transform decomposition, in deep layer The output of the QSSM module is the nth deep quantum state feature that has passed normalization; For the nth channel, it is a single-bit rotation gate; For controlled entanglement; For activation functions; For the output of the nth channel of the quantum feature refinement module, in the shallow layer The high-frequency characteristics of strong coupling in the output after strong physical entanglement across channels, at a deep level. This indicates that the spatial characteristic state after processing by the deep controlled entanglement gate has undergone complete semantic entanglement, meaning that the originally independent deep channels have become completely semantically entangled.

[0026] To address the issue of breakage and feature loss in long lines of plant microtubules during long-sequence scanning, although the original backbone network incorporated Mamba's 2D Selective Scanning (SS2D) mechanism for efficient global receptive field modeling, flattening the 2D spatial image into a one-dimensional feature sequence inevitably disrupts the continuity of local space, resulting in physical "local phase diffusion." To address this problem, this embodiment incorporates a quantum-inspired state-space module within each quantum state-space block (VSS Block) of the Mamba deep feature extraction backbone (i.e., after SS2D scanning) to resolve the structural breakage caused by the scanning. For example... Figure 4 As shown, the quantum-inspired state-space module utilizes a classical control bypass to adaptively learn a control parameter from the current features through an activation function. This parameter then serves as a control signal to drive the controlled phase gate on the quantum circuit, precisely locking and correcting the phase of the master quantum state without altering the feature amplitude (energy intensity). This process can be represented by the following formulas (4) and (5): (4) (5) in, This represents the implicit quantum state characteristics after being selectively scanned in two dimensions by Mamaba; and These represent the weights and biases of the linear layer, respectively. Indicates the activation function; These are the control parameters adaptively learned from the features; Indicates a controlled phase gate; This indicates the phase-locked state characteristics of the output.

[0027] To address the issue of severe residual background noise and artifacts in the final reconstructed image, although the original backbone network uses conventional convolutional layers at the end to directly map features to pixel output, this deterministic mapping mechanism lacks inherent filtering capability against high-frequency noise amplified by deep networks. To address this problem, this embodiment adds a quantum measurement reconstruction module at the network's exit point to solve the insufficient noise resistance during the reconstruction process. For example... Figure 5 As shown, the quantum measurement reconstruction module simulates the wave function measurement collapse mechanism in quantum mechanics. It calculates the collapse probability of each implicit feature (strictly constrained to between 0 and 1 by the Sigmoid function) and the eigenvalue through parallel convolution, and then drives the quantum state to collapse into a classical image by calculating the mathematical expectation of the probability and the eigenvalue. Since the collapse probability of anomalous background noise approaches zero, this mechanism naturally eliminates redundant noise when calculating the expectation. This process can be represented by the following formulas (6) and (7): (6) (7) in This represents the high-dimensional implicit quantum state features after pixel rearrangement upsampling and amplification. This represents the high-resolution feature tensor output by the pixel rearrangement upsampling module; This represents the convolutional layer Conv used to calculate probabilities. prob The weights; This represents the normalization function, used to constrain the probability between 0 and 1; This represents the probability of wave function collapse; Conv represents the convolutional layer that computes eigenvalues. val The weights; This represents the expected value after the quantum state collapses; Represents the eigenvalues. This represents the final output super-resolution image of plant microtubules.

[0028] To clearly demonstrate the super-resolution reconstruction effect of this method, both the original high-resolution plant microtubule images and the original low-resolution plant microtubule images were subjected to the following process: Figure 6 As shown, the plant microtubule image super-resolution reconstruction network of this invention was trained using a plant microtubule image dataset to obtain weights, and then these weights were used to test the test set. The test results are as follows. Figure 7 As shown, by Figure 6 and Figure 7 Comparison and Figure 8 The local comparison shows that the original low-resolution plant microtubule image has many microtubules stuck together due to insufficient resolution, and there are many halos in the dense areas of the image core, which makes the vertical overlapping relationship between microtubules lost. In contrast, the plant microtubule super-resolution image reconstructed by this invention has extremely high contrast even at the densest intersections, and the direction of the intersecting microtubule lines is clearly distinguishable. The background is extremely pure black and there is no ringing effect common in super-resolution algorithms. These phenomena show the great potential of the method of this invention in the field of image super-resolution reconstruction.

[0029] To further verify the effectiveness of the proposed plant microtubule image super-resolution reconstruction network, a comparative experiment was designed. The proposed network was trained and tested on the same dataset against five other super-resolution reconstruction models and the baseline model IRSRMamba. These models included the Efficient Frequency Domain Feature Aggregation Transformer Super-Resolution Reconstruction Network (EFATSR), the CNN-Transformer Embedded Unfolding Super-Resolution Reconstruction Network (CTUNet), the Frequency Domain Partitioning Stepwise Reconstruction Super-Resolution Network (FDSR), the Spectral Nonlocal Super-Resolution Network (SNLSR) which recovers high-frequency structural information by mining feature channels and spatial similarity, and the Dynamic Architecture Network (DSRNet) which dynamically adjusts the receptive field based on the features of different complex scenes. Five metrics commonly used in image super-resolution reconstruction were used for comparison: Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR), Structural Similarity (SSIM), Natural Image Quality Evaluation (NIQE), and BRISQUE (No Reference Image Quality Evaluation). The definitions and calculation methods of these metrics are well-known in the field and will not be elaborated here.

[0030] The experimental results for each model under each index are shown in Table 1. To clearly demonstrate the effectiveness of the method of this invention, the model indices proposed by this method have been bolded in the table.

[0031] Table 1 Comparison of Super-Resolution Reconstruction Results of Models

[0032] The data in Table 1 demonstrates that the model proposed in this invention exhibits advantages in objective pixel-level reconstruction accuracy and structural restoration. In terms of the PSNR metric, which reflects the absolute signal fidelity of the image, the model surpasses all other advanced comparative models, achieving the best performance. Furthermore, compared to the original baseline model IRSRMamba, this invention, after introducing a quantum heuristic module, achieves significant improvements in both SSIM and MSE metrics. This fully demonstrates that the network designed in this invention can more accurately reconstruct the complex topological relationships of images. Although it cannot surpass all comparative methods in metrics such as NIQE and BRISQUE, because the model of this invention prioritizes the ultimate pixel fidelity, this inevitably leads to a slight deviation in the texture preference generated by the algorithm in the unreferenced naturalness statistical features of the image. However, this slight difference in scores is not only imperceptible to the naked eye but also has virtually no impact on the practical application value of scientific images. In summary, the model of this invention achieves the best balance between maintaining the highest reconstruction fidelity and performance.

[0033] This invention also provides a quantum-inspired super-resolution reconstruction system for plant microtubule images, comprising: Image input module, used to receive images of plant microtubules; The super-resolution reconstruction module has a built-in plant microtubule image super-resolution reconstruction network, which is used to execute the quantum-inspired plant microtubule image super-resolution reconstruction method to perform super-resolution reconstruction of plant microtubule images. The output module is used to output the super-resolution image of plant microtubules obtained from super-resolution reconstruction.

[0034] The present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the quantum-inspired super-resolution reconstruction method for plant microtubule images.

[0035] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the quantum-inspired super-resolution reconstruction method for plant microtubule images as described above.

[0036] The above-described invention merely illustrates implementation methods of the present invention and should not be construed as limiting the scope of the invention patent, nor as imposing any form of limitation on the structure of the embodiments of the present invention. It should be noted that those skilled in the art can make various changes and improvements without departing from the concept of the embodiments of the present invention, and these all fall within the protection scope of the embodiments of the present invention.

Claims

1. A plant microtubule image super-resolution reconstruction method based on quantum inspiration, characterized in that, Includes the following steps: A super-resolution reconstruction network for plant microtubule images was constructed with the IRSRMamba model as the backbone network. The backbone network consists of a shallow feature extraction layer, a Mamba deep feature extraction backbone, and an image reconstruction layer. A low-resolution image of plant microtubules is input into the shallow feature extraction layer, which includes a feature fusion module and a quantum superposition embedding module. The feature fusion module is used to extract multi-scale features of the input image and enhance cross-channel entanglement through a shallow quantum feature refinement module to output strongly coupled high-frequency features. The quantum superposition embedding module is used to map the high-frequency features to a complex Hilbert space to generate quantum superposition state features. The quantum superposition state features are input into the Mamba deep feature extraction backbone, which includes an image patch embedding layer, a residual state space group, and an image patch inverse embedding layer. Each residual state space group contains several quantum state space blocks. The quantum state space blocks first perform normalization, two-dimensional selective scanning, and quantum-inspired state space module processing on the input one-dimensional feature sequence in sequence. The quantum-inspired state space module is used to adaptively learn control parameters and correct and lock the phase of the input features. The corrected and locked features are then connected with the residual of the input one-dimensional feature sequence and normalized again. Then, linear entanglement cascade is performed through the deep quantum feature refinement module to achieve quantum state entanglement fusion in the channel dimension. Finally, weight calibration is performed through the channel attention module. The features output from the Mamba deep feature extraction backbone are input into the image reconstruction layer. The image reconstruction layer includes a quantum measurement reconstruction module, which is used to simulate the wave function collapse process and output a super-resolution image of plant microtubules by calculating the collapse probability and the expected value of the eigenvalue.

2. The plant microtubule image super-resolution reconstruction method according to claim 1, characterized in that, The quantum superposition embedding module inputs the input features into two parallel convolutional layers respectively. The absolute value of the output of one convolutional layer is taken to obtain the amplitude. The output of the other convolutional layer is mapped to the arctangent function to obtain the phase. Then, the amplitude and phase are superimposed and recombined in complex Hilbert space using Euler's formula to obtain the quantum superposition state features.

3. The plant microtubule image super-resolution reconstruction method of claim 1, wherein, The quantum-inspired state-space module adaptively learns a control parameter from the current features through an activation function, and then uses the control parameter as a control signal to drive the controlled phase gate, thereby locking and correcting the phase of the main quantum state without changing the feature amplitude.

4. The plant microtubule image super-resolution reconstruction method of claim 1, wherein, The shallow quantum feature refinement module and the deep quantum feature refinement module have the same structure. Both extract the independent phase of each channel through a single-bit rotating gate, and then introduce a controlled entanglement gate for linear entanglement cascade. In the shallow layer, it is used to make high-frequency edge features in different directions interact across channels to remove background noise. At a deeper level, it is used to enable semantic-level entanglement and fusion of features from different channels in a high-dimensional space.

5. The method for super-resolution reconstruction of plant microtubule images according to claim 1, characterized in that, The quantum measurement reconstruction module calculates two sets of matrices through parallel convolution. One set is constrained to the range of 0-1 by the Sigmoid function to obtain the collapse probability, while the other set remains unchanged as the eigenvalues. Then, the mathematical expectation of the collapse probability and the eigenvalues ​​is calculated to drive the quantum state collapse and output a super-resolution image.

6. The method for super-resolution reconstruction of plant microtubule images according to claim 1, characterized in that, In the training phase of the super-resolution reconstruction network for plant microtubule images, the Hessian regularization loss function is used to elevate the geometric continuity constraint of plant microtubules from the pixel level to the topological level.

7. The method for super-resolution reconstruction of plant microtubule images according to claim 1, characterized in that, The Mamba deep feature extraction backbone has 6 layers. Each layer consists of an image patch embedding layer, a residual state space group, and an image patch inverse embedding layer. The residual state space group of each layer contains several quantum state space blocks. After the quantum superposition features are input to each layer of the Mamba deep feature extraction backbone, they are first flattened into a one-dimensional feature sequence by the image patch embedding layer. After being processed by multiple quantum state space blocks, they are restored to a two-dimensional feature map by the image patch inverse embedding layer. Then, they are sequentially processed by convolution and residual connections to complete the complete process of one layer of the Mamba deep feature extraction backbone. Finally, after processing by all 6 layers of the Mamba deep feature extraction backbone, the final output feature map of the Mamba deep feature extraction backbone is obtained.

8. A quantum-inspired super-resolution reconstruction system for plant microtubule images, characterized in that, include: Image input module, used to receive images of plant microtubules; A super-resolution reconstruction module with a built-in plant microtubule image super-resolution reconstruction network, wherein the plant microtubule image super-resolution reconstruction network is used to execute the plant microtubule image super-resolution reconstruction method according to any one of claims 1 to 7, so as to perform super-resolution reconstruction of plant microtubule images; The output module is used to output the super-resolution image of plant microtubules obtained from super-resolution reconstruction.

9. An electronic device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the quantum-inspired super-resolution reconstruction method for plant microtubule images as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the quantum-inspired super-resolution reconstruction method for plant microtubule images as described in any one of claims 1 to 7.

Images