A method for feature enhancement extraction of on-chip optical qubits

Abstract

The application discloses a kind of on-chip light quantum bit feature enhancement extraction methods, suitable for portable field detection in biomedical detection field.The method is realized by combining various image preprocessing and feature extraction techniques, and the rapid and high sensitivity recognition of light quantum bit feature is realized.The method mainly includes the following steps:1) image preprocessing: automatically analyze the brightness characteristics in the original image, identify and classify the brightness extreme image;2) image enhancement: wavelet transform denoising and non-local mean denoising technology are used to improve the image quality, and the image contrast is optimized by histogram equalization;3) feature extraction and enhancement: difference analysis and superposition processing are performed on the processed image, and the target feature is enhanced while the background noise interference is reduced.The application guarantees the efficiency and portability, significantly improves the accuracy and repeatability of detection, and is suitable for rapid biomedical detection in resource-limited environment.

Description

Technical Field

[0001] This invention relates to an image processing method for portable point-of-care testing (POCT) in the field of biomedical detection, and in particular to an on-chip optical quantum bit feature enhancement and extraction method. Background Technology

[0002] Fluorescent nanodiamonds (FNDs) are widely used in biomedical detection due to their unique optical properties. FNDs possess high brightness, long lifetime, and good biocompatibility, making them particularly suitable for biolabeling and imaging. However, traditional image processing methods face numerous challenges in detecting and enhancing the features of photonic qubits generated by FNDs, such as background noise interference, insufficient image contrast, and low feature extraction accuracy.

[0003] In existing technologies, image denoising and enhancement techniques have been widely applied in the field of medical imaging. For example, sparse coding and edge detection techniques are often used to enhance image features. Sparse coding, by learning an overcomplete dictionary, can effectively represent and reconstruct image features, thereby improving the accuracy of feature extraction (Zhou, W., Yi, Y., Bao, J., et al. 2019. Adaptive weighted locality-constrained sparse coding for glaucoma diagnosis. Med. Biol. Eng. Comput. 57, 2055-2067.). Edge detection techniques, by identifying edge pixels in an image, can remove most of the background noise while preserving key details (Vilimek, D., Kubikova, K., J., Barvik, D., Penhaker, M., Cerny, M., Augustynek, M., and Oczka, D. 2019. A quantitative and comparative analysis of edge detectors for biomedical image identification within dynamical noise effect. Med. Biol. Eng. Comput. 57, 2055-2067. However, these methods still have shortcomings when processing optical qubit features. Sparse coding has limited effectiveness in processing images with complex backgrounds because background noise can interfere with the dictionary learning process. While edge detection can effectively identify edges in images, its enhancement effect is not significant when processing low-contrast images. Furthermore, when applied to images with optical qubit features, these methods often fail to effectively highlight target features, resulting in insufficient feature extraction accuracy.

[0004] In terms of feature extraction, existing methods mainly focus on multi-scale feature extraction of images and deep learning techniques. For example, Scale Invariant Feature Transform (SIFT) is a commonly used multi-scale technique for image feature extraction. By detecting keypoints and descriptors, it can effectively extract local features in images (Zheng, L., and Qian, G. 2012. A SIFT-Based Approach for Image Registration. In Green Communications and Networks, Yang, Y. and Ma, M. (Eds.), Lecture Notes in Electrical Engineering, vol 113. Springer, Dordrecht, 39-44.). U-Net is a network structure commonly used for medical image segmentation. Through an encoder-decoder architecture and skip connections, it can achieve high-precision feature extraction while preserving image details (onneberger, O., Fischer, P., and Brox, T. 2015. U-Net: Convolutional Networks for Biomedical Image Segmentation. Med. Image Comput. Comput. Assist. Interv. 234-241.).

[0005] However, these methods also face some challenges in feature extraction of optical qubits generated by FND. For example, generating high-quality labeled data for deep learning models remains a challenge. Annotation of medical images typically requires specialized knowledge and significant time, leading to a lack of labeled data and inconsistent annotation quality (Vilimek, D., Kubikova, K., J., Barvik, D., Penhaker, M., Cerny, M., Augustynek, M., and Oczka, D. 2019. Towards a better understanding of annotation tools for medical imaging: a survey. Multimedia Tools and Applications 57, 2055-2067. Furthermore, deep learning models may still fail to achieve optimal performance when handling high-noise and low-contrast images. The high randomness of FND samples also presents a significant challenge for effective annotation. Traditional annotation methods require manual annotation by experts, which is time-consuming, labor-intensive, and susceptible to subjective influences. To address this issue, automated annotation techniques and active learning methods have begun to receive attention. For example, rule-based automatic annotation techniques can reduce the workload of manual annotation to some extent, while active learning methods can improve annotation efficiency and quality through selective annotation (Spanier, AB, and Joskowicz, L. 2014. Rule-based ventral cavity multi-organ automatic segmentation in CTscans. In Medical Computer Vision: Algorithms for Big Data (MCV 2014), Menze, B. et al. (Eds.), Lecture Notes in Computer Science, vol 8848. Springer, Cham, 39-44.).

[0006] In summary, existing technologies have many shortcomings in the extraction and labeling of optical quantum bit features, and a new method is urgently needed to solve these problems and improve the accuracy and reliability of detection. Summary of the Invention

[0007] The purpose of this invention is to provide a method for enhancing and extracting features of optical qubits on-chip, addressing the current difficulties in feature extraction and labeling.

[0008] The objective of this invention is achieved through the following technical solution: an on-chip optical quantum bit feature enhancement and extraction method, comprising the following steps:

[0009] (1) Image pre-screening: Brightness feature analysis is performed on several original fluorescence images collected, and images with extreme brightness values ​​are identified and classified by image screening algorithm, including images with the highest and lowest brightness.

[0010] (2) Preprocess the brightness extreme value image;

[0011] (3) Perform difference analysis on the highest and lowest brightness images in each pair of preprocessed brightness extreme value images, and then extract the features of photonic qubits by combining multi-image stacking technology.

[0012] Further, step (1) includes the following sub-steps:

[0013] (1.1) Set the modulation frequency f and sampling frequency T, define the size of the sliding window W and the time interval of the acquisition sequence, where the length of the sliding window is T / f, and segment the image sequence along the time axis to ensure that sufficient image data can be covered in each sampling period;

[0014] (1.2) Based on the sampling settings in step (1.1), analyze the image set contained in each sliding window W; calculate the brightness value of each image, i.e., the average value R of the red channel, and select representative brightness extreme value images; use brightness histogram analysis to automatically identify the images with the highest and lowest brightness in each window, and combine these images into a data group G. i Each pair contains one image with the highest brightness and one image with the lowest brightness.

[0015] Furthermore, step (1) also includes:

[0016] Preliminary color correction is performed on the brightness extreme value image pairs. Color balance adjustment technology is applied to compensate for the color deviation of the red channel caused by long-term shooting, so that the same type of extreme value images in the same sequence can achieve color consistency.

[0017] Furthermore, step (2) includes the following sub-steps:

[0018] (2.1) Perform wavelet transform denoising on the brightness extreme value image;

[0019] (2.2) Further perform nonlocal mean denoising on the denoised image obtained in step (2.1);

[0020] (2.3) Perform histogram equalization on the denoised image obtained in step (2.2).

[0021] Furthermore, step (3) includes the following sub-steps:

[0022] (3.1) Perform difference analysis on the preprocessed brightness extremum images, for each pair of brightness extremum image groups G i By subtracting the lowest brightness image from the highest brightness image in the same group, non-target background and noise in the image are removed, and the features of the photonic qubits are extracted to generate a set of photonic qubit feature maps.

[0023] (3.2) Stack the set of optical qubit feature maps obtained in step (3.1) to restore and enhance the feature details weakened during the preprocessing process;

[0024] (3.3) Perform pixel flipping on the stacked image in step (3.2) so that all pixel values ​​of the image are within a reasonable range, and further perform binarization to limit the pixel values ​​to 0 and 255, thereby significantly highlighting the features of the photonic qubit and generating a photonic qubit feature marker map.

[0025] Furthermore, the extracted quantum bit features are used as tag data; the original fluorescence image and the corresponding tag are used to train a neural network model, and the trained neural network is used for quantum bit feature extraction.

[0026] The present invention also provides an on-chip optical quantum bit feature enhancement and extraction device, comprising:

[0027] The image pre-screening module is used to perform brightness feature analysis on several acquired raw fluorescence images, and to identify and classify extreme brightness images, including the images with the highest and lowest brightness, through an image screening algorithm.

[0028] The preprocessing module is used to preprocess the brightness extreme value image;

[0029] The feature extraction module is used to perform difference analysis on the highest and lowest brightness images in each pair of brightness extreme value images after preprocessing, and then combine multi-image stacking technology to extract the features of photonic qubits.

[0030] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the above-described on-chip optical quantum bit feature enhancement extraction method.

[0031] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described on-chip optical quantum bit feature enhancement and extraction method.

[0032] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described on-chip optical quantum bit feature enhancement and extraction method.

[0033] The beneficial effects of this invention are that, through image pre-screening, comprehensive enhancement, and feature extraction and enhancement, it effectively reduces background noise interference, improves image contrast and feature extraction accuracy, and achieves efficient enhancement and extraction of photonic qubit features. Compared with existing technologies, this invention performs exceptionally well in processing high-noise and low-contrast image sequences, significantly improving the detection accuracy and reliability of photonic qubit features. High-quality labeled data is generated through pre-screening and enhancement of the images, effectively supporting the training of deep learning models. Combined with deep learning technologies, such as the U-Net network structure, this invention can serve as data labels, further improving the extraction efficiency of photonic qubit features and solving the problem of processing high-noise and low-contrast image sequences. The combination of automatic labeling technology and active learning methods solves the problem of high randomness in FND samples and the difficulty of manual labeling, improving labeling efficiency and quality. The multi-step comprehensive processing method of this invention reduces the workload of manual labeling while ensuring the integrity and recognizability of photonic qubit features, possessing strong versatility and simplicity. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a schematic diagram of the overall process of the on-chip optical quantum bit feature enhancement and extraction method of the present invention;

[0036] Figure 2 This is a simplified schematic diagram of the on-chip optical quantum bit feature enhancement and extraction method of the present invention;

[0037] Figure 3 These are sample data for the on-chip optical quantum bit feature enhancement and extraction method of the present invention; wherein, (a) is the original sample image and (b) is the feature-enhanced image;

[0038] Figure 4 The images are feature extraction images of the on-chip optical quantum bit feature enhancement extraction method of the present invention; wherein, (a) is a grayscale image of optical quantum bit features, (b) is an extracted image of optical quantum bit features, (c) is an extracted image of optical quantum bit features after pixel flipping, and (d) is a labeled image of optical quantum bit features.

[0039] Figure 5 For a concentration range of 1×10 -16 M~1×10 -9 The results of applying the on-chip optical quantum bit feature enhancement and extraction method of the present invention to sample M are shown;

[0040] Figure 6 This serves as a verification of the results of the on-chip optical quantum bit feature enhancement and extraction method of the present invention.

[0041] Figure 7 This is a schematic diagram of an electronic device structure provided in an embodiment of the present invention. Detailed Implementation

[0042] The present invention will now be described in detail with reference to the accompanying drawings. Unless otherwise specified, the features of the following embodiments and implementations can be combined with each other.

[0043] The core technology of this invention is to generate a binary image of the target photonic qubit image based on pre-screening, enhancement and feature extraction, and thereby achieve efficient photonic qubit feature enhancement and extraction to improve the accuracy and efficiency of biomedical detection.

[0044] This invention proposes an on-chip optical qubit feature enhancement and extraction method, comprising the following steps:

[0045] (1) Image pre-screening: Brightness feature analysis is performed on the acquired raw images, and images with extreme brightness values ​​are identified and classified through image screening algorithms, including images with the highest and lowest brightness.

[0046] In one embodiment, step (1) specifically includes the following sub-steps:

[0047] (1.1) Sample acquisition: To ensure sufficient data, the fluorescence signal of the FND photonic quantum bit sample is modulated at frequency f to generate a scintillation fluorescence signal, and the sampling frequency T of the image sensor is set.

[0048] (1.2) Extreme value image screening: For the scintillation image sequence of photoqubits obtained in step (1.1), the first L frames are truncated as the main interval for feature extraction to reduce the interference of thermal noise caused by the long-term operation of the image sensor. Subsequently, based on the frequency parameter settings in step (1.1) and considering the delay effect of image sensor sampling, the length of the sliding window W is set to T / f so that each window contains an image with the highest and lowest brightness to capture significant changes in photoqubit features.

[0049] The brightness values ​​of all images within the window are calculated using the following formula, which is equivalent to calculating the average pixel value of the red channel for each image:

[0050]

[0051] Among them, R l R is the average pixel value of the red channel of the l-th frame image within the window, N is the total number of pixels in the image, and R is the average pixel value of the red channel. xIt is the red channel value of the xth pixel.

[0052] Subsequently, a brightness histogram is constructed for each sliding window. Each interval b (bin) of the histogram represents a range of brightness values, calculated by taking the average brightness value I of all images within the window. l The sums are accumulated within the corresponding intervals. The formula can be expressed as:

[0053]

[0054] Where H(b) is the number of images whose brightness falls within the interval b, m is the total number of images in the window, and δ is the Kronecker function, when b = bin(I l When ), the value of δ is 1, representing the brightness value I of the image. l It falls exactly within interval b of the histogram; otherwise, the value of δ is 0. Find the highest brightness interval b in the histogram where the number of images is not zero. max and the lowest brightness range b min Within the range, the images with the highest and lowest brightness values ​​are selected as the standards for image color correction, and the average red channel values ​​are denoted as R0 and R1 respectively. max and R min Next, the images with the highest and lowest brightness within each window are identified, and these images are grouped into a dataset G. i ;

[0055] (1.3) Color Correction: Perform preliminary color correction on the extreme brightness image pairs obtained from step (1.2), paying particular attention to the adjustment of the red channel. Apply color balance adjustment techniques to compensate for the red channel color deviation caused by long-term shooting, and adjust the red channel value of each pixel in the extreme value image using the following formula:

[0056]

[0057] Among them, R max and R min These are the maximum and minimum red channel values ​​in the image pixels, typically 255 and 0, respectively. bright (i) and R' dark (i) are the corrected red channel values ​​of the i-th pixel in the brightest and darkest images, respectively. bright (i) and R dark (i) are the original red channel values ​​of the i-th pixel in the brightest and darkest images, respectively. and These are the average red channel values ​​for the brightest and darkest images, respectively.

[0058] By following the above method, you can obtain a set of extreme image data that are consistent in color within the same sequence.

[0059] (2) Preprocess the brightness extreme value image;

[0060] Specifically, comprehensive image enhancement involves performing wavelet transform denoising (by decomposing the image into different frequency levels, reducing high-frequency noise while enhancing low-frequency features) and non-local mean denoising (reducing texture and edge noise in the image while maintaining the details and structural integrity of the photonic qubits) on the selected extreme value image data set to initially improve image quality. Histogram equalization technology is then used to optimize image contrast and improve the recognizability and processing accuracy of photonic qubit features.

[0061] In one embodiment, step (2) specifically includes the following sub-steps:

[0062] (2.1) Wavelet transform denoising: Based on the extreme value image data group G obtained in step (1.3) i Wavelet transform denoising technology is applied to the images within the group. The image matrix A is converted to floating-point numbers and normalized to obtain the normalized image matrix A. norm ,Right now:

[0063]

[0064] Where min(A) is the minimum value of matrix A, and max(A) is the maximum value of matrix A.

[0065] Subsequently, the normalized image A was processed using two-dimensional discrete wavelet transform (DWT). norm Multi-level decomposition, which includes first decomposing the image matrix A norm Perform a one-dimensional discrete wavelet transform on each row to obtain the row-transformed matrix A. norm_r Then for A norm_r Perform a one-dimensional discrete wavelet transform on each column to obtain the final two-dimensional wavelet transform matrix A. norm_w The decomposition coefficient is set to 1, meaning only one decomposition is performed. During this process, the approximation coefficient *a* and detail coefficient *d* at each location *k* in the image are calculated using the following formula:

[0066]

[0067] Where φ and These are the scaling function and the wavelet function, respectively, where N is the normalized image matrix A. norm The number of elements, i.e., the total number of pixels in the image. This method uses the Haar wavelet basis function:

[0068]

[0069] Therefore, the approximation coefficient a and the detail coefficient d can be rewritten as:

[0070]

[0071] Before denoising, the noise level is first estimated. The standard deviation σ of the noise can be obtained using the median estimation method from the detail coefficients d[k]. The specific formula is as follows:

[0072]

[0073] Where κ is the theoretical scaling factor between the absolute deviation and the standard deviation in a Gaussian distribution, typically taken as 0.6745. Using the estimated noise level, a threshold θ is calculated to effectively remove noise while preserving as much important information in the image as possible.

[0074]

[0075] Where N is the total number of pixels in the image. This threshold is used to perform soft thresholding on the wavelet coefficients, removing detail coefficient variations caused by noise.

[0076]

[0077] This method ensures that only coefficients greater than a threshold are retained, while smaller coefficients are set to zero or reduced accordingly. Finally, the modified coefficients are used for inverse wavelet transform to reconstruct the denoised image A. recon :

[0078]

[0079] This method removes coarse structure and large-scale noise from images.

[0080] (2.2) Non-local mean denoising: Based on the extreme value image obtained after applying wavelet transform denoising in step (2.1), non-local mean denoising is applied. The value of each pixel is calculated by weighted averaging the pixel values ​​of similar blocks around the pixel. The weights are based on the similarity between blocks, with blocks of higher similarity having greater weights. The image is divided into multiple overlapping pixel blocks. For each pixel i, a small block v(n) centered on that pixel is defined. i The size is set to (2r+1)×(2r+1), where r is the block radius. For each pixel block v(n) i ) and v(n j Calculate the Euclidean distance between them. The smaller the distance, the more similar the blocks are. The weight w(i,j) of pixel i reflects the similarity between each pixel j and pixel i. The higher the similarity, the greater the contribution of the pixel to pixel i. It is calculated by the following formula:

[0081]

[0082] Where ||·||² represents the Euclidean distance between blocks, and h is a parameter that adjusts the filtering intensity. To ensure that the sum of the weights is 1, the calculated weights are normalized. Specifically, for the weight w(i,j) of each pixel i, normalization is performed according to the following formula:

[0083]

[0084] Where N(i) is the neighborhood of pixel i. The denoised pixel values ​​are obtained by weighting the image pixel values ​​using the normalized weights:

[0085]

[0086] Where I(j) is the value of pixel j in the original image, I NLM (i) is the pixel value after nonlocal mean denoising.

[0087] This method removes noise from an image while preserving its details and edge information.

[0088] (2.3) Histogram Equalization: Based on the extreme value image obtained after nonlocal mean denoising in step (2.2), histogram equalization is applied. First, the color image is converted from the BGR color space to the YCrCb color space. This conversion process allows the separation of the image's luminance component (Y channel) and chrominance component (Cr and Cb channels). For each pixel in the image, the conversion formula is as follows:

[0089] Y = 0.299R + 0.587G + 0.114B

[0090] Cb=-0.1687R-0.3313G+0.5B+128

[0091] Cr = 0.5R - 0.4187G - 0.0813B + 128

[0092] Where R, G, and B represent the red, green, and blue chromatic components in the original BGR color space, respectively. This method performs histogram equalization only on the luminance channel Y, while leaving the Cr and Cb channels unchanged. The mathematical expression for this operation is:

[0093]

[0094] Where H(v) is a cumulative distribution function, representing the cumulative sum of the number of pixels less than or equal to the brightness value v, H′(v) is the value of H(v) after equalization, N is the total number of pixels in the image, and L... max This is the maximum possible brightness value. This processing remaps the original brightness value to a new brightness level, resulting in a more uniform brightness distribution across the entire usable range.

[0095] After histogram equalization, the Y channel is merged back with the original color channels to form an adjusted YCrCb image, which is then converted back to the BGR color space to complete the entire histogram equalization process. The inverse conversion formula is as follows:

[0096] B = Y + 1.772(Cb - 128)

[0097] G=Y-0.3344136(Cb-128)-0.714136(Cr-128)

[0098] R = Y + 1.402(Cr - 128)

[0099] Following the method described above, such as Figure 3 As shown in (b), not only is the local contrast of the image improved, but the overall color balance is also maintained, thus obtaining a set of extreme image data with more obvious contrast after denoising.

[0100] (3) Perform difference analysis on the highest and lowest brightness images in each pair of brightness extreme value images after preprocessing, and then combine the multi-image stacking technology to extract the features of photonic qubits to generate feature maps.

[0101] In one embodiment, feature extraction and enhancement involves applying difference analysis and multi-image stacking techniques to effectively extract and enhance target features, significantly reducing background noise interference and ensuring that the extracted photonic qubit features have high detection accuracy and repeatability. Specifically, this includes the following sub-steps:

[0102] (3.1) Difference Analysis: Difference analysis is performed on the denoised and contrast-prone extreme image data obtained in step (2) to extract the photonic qubit features in the image sequence. Difference analysis is achieved by calculating the gray-level difference between corresponding pixels in two images. The calculation formula is as follows:

[0103] D(i)=I bright (i)-I dark (i)

[0104] Where D(i) is the value of the i-th pixel in the difference image, I bright (i) and I dark (i) represents the grayscale value of the i-th pixel in a pair of images with the highest and lowest brightness, respectively. This method calculates the difference image D for a pair of extreme brightness images within a group and extracts the photonic qubit features affected by brightness variations. The denoised image also contains some residual noise, so medium filtering is applied to the difference image. Finally, grayscale values ​​are adjusted to ensure that all pixel values ​​in the image are correctly limited to the range of 0 to 255.

[0105] Processed using this method, such as Figure 4 As shown in (a), a set of grayscale images of optical qubit features are obtained.

[0106] (3.2) Multi-image stacking: Multi-image stacking is applied to the grayscale images of the set of photonic qubit features obtained in step (3.1) for feature enhancement. This stacking process involves accumulating corresponding pixels from multiple images to highlight signals that appear consistently across all images. By accumulating the grayscale values ​​of corresponding pixels in the same sequence, the residual background noise is further reduced to highlight the features by utilizing the property of pixel values ​​overflowing to the limit (255). The formula is as follows:

[0107]

[0108] Where S(i) is the accumulated value of the i-th pixel in the stacked image, I j (i) is the value of the j-th image at the ith pixel, and NUM is the total number of images participating in the stacking. The target feature values ​​are relatively enhanced, while the background is suppressed due to value overflow. Subsequently, the accumulated pixel values ​​are cropped to ensure that the value range is within the standard grayscale range.

[0109] Processed using this method, such as Figure 4 As shown in (b) in the figure, the extraction diagram of the optical quantum bit features is obtained.

[0110] (3.3) Pixel Flipping and Binarization: The extracted image of the photonic qubit features obtained in step (3.2) is subjected to pixel flipping. Considering that in typical image processing scenarios, the target features should exhibit high pixel intensity while the background is low, the stacked image is subjected to pixel flipping, such as... Figure 4 As shown in (c), the extracted image of the photonic qubit features after pixel flipping is obtained. To present the photonic qubit features of all signal intensities, the image is binarized, that is, non-background pixels (pixel value of 0) are marked as target feature pixels (pixel value of 255).

[0111] Following the method described above, such as Figure 4 As shown in (d) in the figure, the characteristic labeling map of the optical qubit can be obtained.

[0112] Example 1:

[0113] On a machine equipped with an Intel Core i9-13900HX CPU, an Nvidia GTX3090 GPU, and 32GB of RAM, according to Figure 1 , Figure 2 The illustrated process implements one embodiment of the present invention. In this embodiment, the modulation frequency is set to 2Hz, the sampling frequency is 10Hz, and the concentration is 1×10⁻⁶. -9Sample the FND sample of M, such as Figure 3 As shown in (a), a 100-frame flashing image sequence was obtained. A sliding window of length 5 was applied to filter this image sequence, resulting in 96 sets of extreme value image data.

[0114] As described in step (2), wavelet transform denoising, nonlocal mean denoising, and histogram equalization are performed on all images in the extreme value image data set. These processes enhance the contrast between target features and background in the images, such as... Figure 3 As shown in (b) of the diagram.

[0115] like Figure 4 As shown, the enhanced image is subjected to difference analysis, multi-image stacking, pixel flipping and binarization processing to finally obtain the photonic qubit feature map corresponding to the image sequence.

[0116] The above process applies to a concentration range of 1×10 -16 M~1×10 -9 Features were extracted from samples M, and the results are as follows: Figure 5 As shown. To verify the rationality of the extracted features, in this embodiment, four concentrations are first randomly selected, and five sets of original image sequences are randomly chosen within each concentration to perform Fast Fourier Transform (FFT) to analyze the frequency components and specifically extract the 2Hz amplitude. Subsequently, the same FFT process is applied only to feature regions containing only those marked by this invention, and the results are as follows. Figure 6 As shown, it is evident that the amplitude of the 2Hz frequency component is significantly increased due to the substantial reduction in interference from background noise. This result verifies the effectiveness of the present invention in extracting and enhancing the features of optical qubits.

[0117] In one embodiment, the photonic qubit feature map extracted by the above method is used as tag data; the original fluorescence image and the corresponding tag are used to train a neural network model, and the trained neural network is used for photonic qubit feature extraction.

[0118] Neural network models include U-Net-ConvLSTM networks, etc.

[0119] The present invention also provides an on-chip optical quantum bit feature enhancement and extraction device, comprising:

[0120] The image pre-screening module is used to perform brightness feature analysis on several acquired raw fluorescence images, and to identify and classify extreme brightness images, including the images with the highest and lowest brightness, through an image screening algorithm.

[0121] The preprocessing module is used to preprocess the brightness extreme value image;

[0122] The feature extraction module is used to perform difference analysis on the highest and lowest brightness images in each pair of brightness extreme value images after preprocessing, and then combine multi-image stacking technology to extract the features of photonic qubits.

[0123] It should be noted that the device embodiment shown in this embodiment matches the content of the above method embodiment, and the content of the above method embodiment can be referred to, and will not be repeated here.

[0124] Figure 7 This is a schematic diagram of an electronic device structure provided in an embodiment of the present invention. Please refer to... Figure 5 The electronic device provided in this embodiment includes a memory and a processor. The memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions. When the program instructions are loaded and executed by the processor, they implement an on-chip optical quantum bit feature enhancement and extraction method of the present invention.

[0125] It should be noted that, in addition to Figure 7 In addition to the memory and processor shown, electronic devices may include other hardware depending on their actual functions, which will not be elaborated further.

[0126] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the above-described on-chip optical quantum bit feature enhancement and extraction method.

[0127] The computer-readable storage medium can be an internal storage unit of any data processing device described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be any data processing device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0128] The present invention also provides a computer program product, including a computer program, characterized in that, when the computer program is executed by a processor, it implements the above-described on-chip optical quantum bit feature enhancement and extraction method.

[0129] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0130] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0131] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0132] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0133] The above embodiments are only used to illustrate the design concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made based on the principles and design ideas disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A method for enhancing and extracting features of on-chip optical qubits, characterized in that, Includes the following steps: (1) Image pre-screening: The brightness features of several original fluorescence images are analyzed, and the images with extreme brightness values ​​are identified and classified by the image screening algorithm, including the images with the highest and lowest brightness. (2) Preprocess the brightness extreme value image; (3) Perform difference analysis on the highest and lowest brightness images in each pair of brightness extreme value images after preprocessing, and then extract the features of photonic qubits by combining multi-image stacking technology; Step (1) includes the following sub-steps: (1.1) Set the modulation frequency f and sampling frequency T Define a sliding window W The size and time interval of the acquisition sequence, where the length of the sliding window is... T / f, The image sequence is segmented along the time axis to ensure that sufficient image data is covered in each sampling period; (1.2) Based on the sampling settings in step (1.1), in each sliding window W Within the dataset, the system analyzes the included image set and calculates the brightness value for each image. ， That is, the average value of the red channel. R Representative images with extreme brightness values ​​were selected; brightness histogram analysis was used to automatically identify the images with the highest and lowest brightness within each window, and these images were grouped into a dataset. G i Each pair contains one image with the highest brightness and one image with the lowest brightness; Step (1) further includes: Preliminary color correction is performed on the brightness extreme value image pairs, and color balance adjustment technology is applied to compensate for the color deviation of the red channel caused by long-term shooting, so that the same type of extreme value images in the same sequence can achieve color consistency. Step (2) includes the following sub-steps: (2.1) Perform wavelet transform denoising on the brightness extreme value image; (2.2) Further perform nonlocal mean denoising on the denoised image obtained in step (2.1); (2.3) Perform histogram equalization on the denoised image obtained in step (2.2); Step (3) includes the following sub-steps: (3.1) Perform difference analysis on the preprocessed extreme brightness images, for each pair of extreme brightness image groups G i By subtracting the lowest brightness image from the highest brightness image in the same group, non-target background and noise in the image are removed, and the features of the photonic qubits are extracted to generate a set of photonic qubit feature maps. (3.2) Stack the set of photonic qubit feature maps obtained in step (3.1) to restore and enhance the feature details weakened during the preprocessing process; (3.3) Perform pixel flipping on the stacked image in step (3.2) so that all pixel values ​​of the image are within a reasonable range, and further perform binarization to limit the pixel values ​​to 0 and 255, thereby significantly highlighting the features of the photonic qubit and generating a photonic qubit feature marker map.

2. The on-chip optical quantum bit feature enhancement and extraction method according to claim 1, characterized in that, The features of the photonic qubits extracted in claim 1 are used as tag data; the original fluorescence image and the corresponding tag are used to train a neural network model, and the trained neural network is used for photonic qubit feature extraction.

3. An on-chip optical quantum bit feature enhancement and extraction device, characterized in that, A method for enhancing and extracting on-chip optical qubit features according to any one of claims 1-2 includes: The image pre-screening module is used to perform brightness feature analysis on several acquired raw fluorescence images, and to identify and classify extreme brightness images, including the images with the highest and lowest brightness, through an image screening algorithm. The preprocessing module is used to preprocess the brightness extreme value image; The feature extraction module is used to perform difference analysis on the highest and lowest brightness images in each pair of brightness extreme value images after preprocessing, and then combine multi-image stacking technology to extract the features of photonic qubits.

4. An electronic device, comprising a memory and a processor, characterized in that, The memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the on-chip optical quantum bit feature enhancement and extraction method according to any one of claims 1-2.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements an on-chip optical quantum bit feature enhancement and extraction method as described in any one of claims 1-2.

6. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the on-chip optical quantum bit feature enhancement and extraction method according to any one of claims 1-2.

Images