A Correlation Imaging Method Based on Transfer Learning

Abstract

The present application provides a correlation imaging method based on transfer learning, which relates to the field of correlation imaging technology, including: constructing a speckle network based on differential correlation imaging, obtaining optimized speckles after training; using the optimized speckle modulated light source to illuminate the target object to obtain the bucket detector value; performing differential correlation imaging to obtain the initial target pixel matrix; performing a convolution operation on the target pixel matrix and the optimized speckle to obtain an estimated bucket signal; and updating the value of each pixel of the target pixel matrix. The present invention ultimately only needs to update the pixel matrix to obtain a high-quality target image, without updating millions or even tens of millions of network parameters, which greatly improves the computing efficiency and reduces the requirements for computing resources of the terminal device.

Description

Technical Field

[0001] The present application relates to the technical field of correlation imaging, and in particular to a correlation imaging method based on transfer learning. Background Art

[0002] Correlation imaging is a new imaging technology that uses the intensity fluctuation correlation of the light field to image the target object. Compared with traditional imaging technology, it has the advantages of anti-scattering imaging, super-resolution imaging, and multi-band imaging. Achieving high-quality image reconstruction under low sampling conditions has always been the research focus of correlation imaging. In recent years, deep learning technology has been widely used in the field of optics. There are currently two main types of correlation imaging algorithms based on deep learning: one is a trained correlation imaging algorithm, and the other is an untrained correlation imaging algorithm, which is used to achieve high-quality image reconstruction under low sampling conditions.

[0003] However, in practical applications, trained correlation imaging algorithms require a large amount of data sets for training, and the imaging quality is poor for target scenes that are not in the training data sets. Untrained correlation imaging algorithms do not require training data sets, and can restore high-quality images by using the loss of estimated bucket signals and real bucket signals. However, existing correlation imaging algorithms that do not require training data sets have low imaging efficiency, poor comprehensive capabilities of imaging efficiency and imaging quality, and high requirements for computing resources of terminal devices. Summary of the invention

[0004] The present application is made in view of the above-mentioned problems, and its purpose is to provide a correlation imaging method based on transfer learning, which obtains optimized speckle based on differential correlation imaging, and migrates the optimized speckle and the differential correlation imaging results based on the optimized speckle to unsupervised correlation imaging. It only needs to update the pixel matrix to obtain a high-quality target image, without updating millions or even tens of millions of network parameters, which greatly improves the computing efficiency and reduces the requirements for computing resources of terminal devices.

[0005] Specifically, the first aspect of the present application provides an association imaging method based on transfer learning, comprising the following steps:

[0006] Step 1: Construct a trainable speckle network based on differential correlation imaging and train the network to obtain optimized speckles;

[0007] Speckle is an optical phenomenon usually formed by the random interference pattern of light waves when passing through an inhomogeneous medium. In correlation imaging, speckle is used as a random light field to reconstruct the target image by correlating with the reflected or transmitted light of the target object.

[0008] Step 2: Use the optimized speckle modulated light source to illuminate the target object to obtain the bucket detector value;

[0009] Bucket detector is a detector used in single-pixel imaging technology. Its basic principle is to achieve imaging by receiving total light intensity information. This detector does not have the ability to resolve spatial information, so during the imaging process, it only records the sum of the light intensity rather than the specific light intensity distribution.

[0010] Step 3: Perform differential correlation imaging on the collected bucket signal and the optimized speckle to obtain the initial target pixel matrix;

[0011] Step 4: Convolve the target pixel matrix with the optimized speckle to obtain the estimated bucket signal;

[0012] Step 5: By minimizing the error between the estimated bucket signal and the actual bucket signal, the value of each pixel in the target pixel matrix is ​​updated to quickly restore a high-quality target object image.

[0013] Furthermore, the step 1 specifically includes:

[0014] Step 1.1: Use a trainable convolutional layer to transform the speckle Indicates that the batch size of the input images is set and the differential correlation imaging algorithm is added to the convolutional network;

[0015] Step 1.2: Prepare a dataset consisting of identical images As the input of the network, and divided into training set and test set;

[0016] Step 1.3: Make a copy of the dataset to calculate the loss function. The formula is as follows:

[0017] ;

[0018] in: is the loss function;

[0019] It is the result of differential correlation imaging after convolution operation on the input image and the convolution layer;

[0020] For input picture;

[0021] is the weight of the loss term;

[0022] is the structural similarity index;

[0023] is the mean square error;

[0024] Step 1.4: Use the Adam optimizer to train the speckle using the dataset. When the performance does not improve significantly in multiple rounds of testing on the test set, terminate the training and obtain the trained speckle.

[0025] Step 1.5: Binarize the trained speckle pattern using the following formula:

[0026] ;

[0027] in: is the speckle after binarization;

[0028] is the sign function, used for positive and negative binarization;

[0029] is the speckle after training;

[0030] Step 1.6: Replace the convolutional layer with the binarized speckle, add an additive binarized convolution, and retrain to obtain the optimized speckle.

[0031] Furthermore, the step 2 specifically includes:

[0032] The optimized speckle pattern is loaded into the DMD, and the light source is modulated by the DMD and irradiated to the object. The light passing through the object is received by a bucket detector without spatial resolution to obtain the real bucket signal value.

[0033] Furthermore, the step three specifically includes:

[0034] The collected bucket signal value and the optimized speckle are differentially correlated to obtain the initial target pixel matrix. The formula is as follows:

[0035] ;

[0036] ;

[0037] ;

[0038] ;

[0039] in: is the target pixel matrix;

[0040] is the total number of sampling times;

[0041] is the optimized speckle at the i-th detection;

[0042] is the total light intensity detected by the bucket detector for the i-th time;

[0043] is the ensemble average of the optimized speckle at the i-th detection;

[0044] is the ensemble average of the total light intensity detected by the bucket detector for the i-th time;

[0045] For the i-th The sum of the light intensities;

[0046] For the i-th The ensemble average of the sum of light intensities.

[0047] Furthermore, the step 4 specifically includes: performing a convolution operation on the target pixel matrix and the optimized speckle to obtain an estimated bucket signal, and the formula is as follows:

[0048] ;

[0049] in: is the estimated bucket signal;

[0050] is the optimized speckle at the i-th detection;

[0051] is the convolution operation;

[0052] is the target pixel matrix.

[0053] Furthermore, the step five specifically includes: updating the value of each pixel of the target pixel matrix by minimizing the error between the estimated bucket signal and the real bucket signal, and restoring the high-quality target object image, the formula is as follows:

[0054] ;

[0055] ;

[0056] in: is the restored high-quality target object image;

[0057] is the target image that needs to be reconstructed;

[0058] The parameter value for which the function obtains the minimum value within the domain;

[0059] The estimated bucket signal when the error between the estimated bucket signal and the real bucket signal is the smallest;

[0060] is the estimated bucket signal;

[0061] is the real barrel signal;

[0062] is the optimized speckle at the i-th detection;

[0063] is the convolution operation.

[0064] In the second aspect, the present application also provides a computing device, which has the function of implementing the method described in the first aspect above. The beneficial effects can be found in the description of the first aspect, which will not be repeated here. The function can be implemented by hardware, or by executing the corresponding software implementation through hardware. The hardware or software includes one or more modules corresponding to the above functions. In one possible design, the structure of the device includes an acquisition module, a training module, and optionally, a construction module. These modules can implement the function of the training node in the method example of the first aspect above. Please refer to the detailed description in the method example for details, which will not be repeated here.

[0065] In the third aspect, the present application also provides a computing device, which is used to implement the functions of the method described in the first aspect above. The beneficial effects can be found in the description of the first aspect and will not be repeated here. The structure of the computing device includes a processor and a memory, and the memory is used to store instructions and / or data. The memory is coupled to the processor, and when the processor executes the program instructions stored in the memory, the function of the training node in the example of the first aspect above can be implemented. The structure of the computing device also includes a communication interface for communicating with other devices.

[0066] In a fourth aspect, the present application also provides a computer-readable storage medium, in which instructions are stored, and when the computer-readable storage medium is run on a computer, the computer executes the method in the above-mentioned first aspect and various possible designs of the first aspect.

[0067] In a fifth aspect, the present application also provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the method in the above-mentioned first aspect and various possible designs of the first aspect.

[0068] In a sixth aspect, the present application also provides a computing chip, which is connected to a memory, and the chip is used to read and execute a software program stored in the memory, and to execute the methods in the above-mentioned first aspect and various possible implementation methods of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the embodiments of the present drawings or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present drawings. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying any creative work.

[0070] Figure 1 A flowchart of the steps of this application;

[0071] Figure 2 A schematic diagram of a method for obtaining optimized speckle by training based on differential correlation imaging is provided for an embodiment of the present invention;

[0072] Figure 3 This is the principle diagram of the associated imaging of this application;

[0073] Figure 4 A schematic diagram of an associative imaging method based on transfer learning in the present application.

[0074] The purpose, features and advantages of this figure will be further described in conjunction with the embodiments with reference to the accompanying drawings. DETAILED DESCRIPTION

[0075] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments provided in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0076] Obviously, the drawings described below are only some examples or embodiments of the present application. For ordinary technicians in this field, the present application can also be applied to other similar scenarios based on these drawings without creative work. In addition, it can also be understood that although the efforts made in this development process may be complicated and lengthy, for ordinary technicians in this field related to the content disclosed in this application, some changes in design, manufacturing or production based on the technical content disclosed in this application are just conventional technical means, and should not be understood as insufficient content disclosed in this application.

[0077] Unless otherwise specified, all embodiments and optional embodiments of the present application can be combined with each other to form a new technical solution.

[0078] Unless otherwise specified, all technical features and optional technical features of this application can be combined with each other to form a new technical solution.

[0079] If there is no special explanation, all steps of the present application can be performed sequentially or randomly, preferably sequentially. For example, the method includes steps (a) and (b), which means that the method may include steps (a) and (b) performed sequentially, or may include steps (b) and (a) performed sequentially. For example, the method may further include step (c), which means that step (c) may be added to the method in any order. For example, the method may include steps (a), (b) and (c), or may include steps (a), (c) and (b), or may include steps (c), (a) and (b), etc.

[0080] If there is no special explanation, the "include" and "comprising" mentioned in this application are open-ended or closed-ended. For example, the "include" and "comprising" may mean that other components not listed may also be included or only the listed components may be included or only the listed components may be included.

[0081] If not specifically stated, in this application, the term "or" is inclusive. For example, the phrase "A or B" means "A, B, or both A and B". More specifically, any of the following conditions satisfies the condition "A or B": A is true (or exists) and B is false (or does not exist); A is false (or does not exist) and B is true (or exists); or both A and B are true (or exist).

[0082] In order to better understand the solutions of the embodiments of the present application, some relevant terms and concepts that may be involved in the embodiments of the present application are first introduced below.

[0083] (1) Artificial intelligence (AI), also known as intelligent machinery or machine intelligence, refers to machines that are made by humans and can display intelligence. Generally speaking, AI refers to the technology that displays human intelligence through ordinary computer programs.

[0084] (2) Machine learning (ML). Machine learning is the core of artificial intelligence. The theory of machine learning mainly designs and analyzes some algorithms that allow computers to learn automatically. Machine learning algorithms are a type of algorithm that automatically analyzes data to obtain patterns and uses the patterns to predict unknown data. Therefore, the core of machine learning is data, algorithms (models), and computing power (computer computing power). The application areas of machine learning are very wide, including data mining, data classification, computer vision, natural language processing (NLP), biometric recognition, search engines, medical diagnosis, detection of credit card fraud, securities market analysis, DNA sequencing, speech and handwriting recognition, strategic games and robot applications. Machine learning is to design an algorithm model to process data and output the results that users want. Users can continuously tune the algorithm model to form more accurate data processing capabilities.

[0085] (3) Deep learning (DL). Deep learning is a method of machine learning. Its concept originates from the study of artificial neural networks. Multilayer perceptrons with multiple hidden layers are a type of deep learning structure, so deep learning is often referred to as deep neural networks. Compared with general machine learning, deep learning can automatically extract features, that is, automatically combine simple features into more complex features, and use these combinations to perform multi-layer weight learning to solve problems. The motivation for studying deep learning is to establish a neural network that simulates the human brain for analysis and learning. It imitates the mechanism of the human brain to interpret data, such as images, sounds, and text. Deep learning first emerged in image recognition, but in just a few years, deep learning has been extended to various fields of machine learning and has performed well. It has been applied in various fields such as image recognition, speech recognition, audio processing, natural language recognition, robot bioinformatics processing, search engines, human-computer games, online advertising targeting, medical automatic diagnosis, and finance.

[0086] (4) Transfer learning (TL): Transfer learning is another method of machine learning, which refers to the reapplication of a pre-trained model in another task. In the process of using deep learning technology to solve problems, the most common obstacle is the massive amount of data required in the model training process. The reason why so much data is needed is that the machine will encounter a large number of parameters in the model during the learning process. When facing specific problems in a certain field, it is usually impossible to obtain the data of the scale required to build the model. However, the relationship obtained for a certain type of data in a model training task can also be applied to other problems in the same field. In other words, transfer learning is to transfer the knowledge of one field (source domain) to another field (target domain), so that the target field can achieve better learning results on the basis of limited data sets for solving downstream tasks.

[0087] (5) Neural networks (NN) are mathematical models that imitate the behavioral characteristics of animal neural networks and perform distributed parallel information processing. The purpose of information processing can be achieved by adjusting the interconnected relationships between a large number of nodes within the neural network. Neural networks have the ability to self-learn and adapt. Neural networks are usually used in artificial intelligence model training and data derivation processing.

[0088] Specifically, a neural network may usually include multiple layers connected end to end, such as convolutional layers, fully connected layers (FC), activation layers or pooling layers.

[0089] (6) Deep neural network (DNN), also known as multi-layer neural network, can be understood as a neural network with multiple hidden layers. According to the position of different layers, DNN can be divided into three categories: input layer, hidden layer, and output layer. Generally speaking, the first layer is the input layer, the last layer is the output layer, and the layers in between are all hidden layers. The layers are fully connected, that is, any neuron in the i-th layer must be connected to any neuron in the i+1-th layer. Although DNN looks complicated, the work of each layer is not complicated. Simply put, it is the following linear relationship expression: where is the input vector, is the output vector, is the offset vector, W is the weight matrix (also called coefficient), and α() is the activation function. Each layer simply performs such a simple operation on the input vector to obtain the output vector.

[0090] (7) Convolutional neural network (CNN) is a deep neural network with a convolutional structure. A convolutional neural network contains a feature extractor consisting of a convolution layer and a subsampling layer, which can be regarded as a filter. A convolution layer refers to a neuron layer in a convolutional neural network that performs convolution processing on the input signal. In a convolutional layer of a convolutional neural network, a neuron can only be connected to some neurons in adjacent layers. A convolutional layer usually contains several feature planes, each of which can be composed of a number of rectangularly arranged neural units. The neural units in the same feature plane share weights, and the shared weights here are convolution kernels. Shared weights can be understood as a way of extracting image information that is independent of position. The convolution kernel can be initialized in the form of a matrix of random size, and the convolution kernel can obtain reasonable weights through learning during the training process of the convolutional neural network. In addition, the direct benefit of shared weights is to reduce the connections between the layers of the convolutional neural network, while reducing the risk of overfitting.

[0091] (8) Loss function. In the process of training a deep neural network, we hope that the output of the deep neural network is as close as possible to the value we really want to predict. Therefore, we can compare the predicted value of the current network with the target value we really want, and then update the weight vector of each layer of the neural network according to the difference between the two (of course, there is usually an initialization process before the first update, that is, pre-configuring parameters for each layer in the deep neural network). For example, if the predicted value of the network is high, adjust the weight vector to make it predict a lower value, and continue to adjust until the deep neural network can predict the target value we really want or a value very close to the target value we really want. Therefore, it is necessary to pre-define "how to compare the difference between the predicted value and the target value", which is the loss function or objective function. They are important equations used to measure the difference between the predicted value and the target value. Among them, taking the loss function as an example, the higher the output value (loss) of the loss function, the greater the difference, so the training of the deep neural network becomes a process of minimizing this loss as much as possible.

[0092] (9) Differential Ghost Imaging (DGI) is a technique used to improve imaging quality, especially by using differential signals in traditional ghost imaging (GI) to improve the signal-to-noise ratio and reduce the impact of external noise on imaging results. In a ghost imaging system, the light beam emitted by a light source is usually divided into a signal beam and a reference beam, where the total light intensity of the signal beam is measured by a bucket detector, while the intensity distribution of the reference beam is measured by a detector with spatial resolution.

[0093] The basic idea of ​​differential correlation imaging is to use the AC component of the bucket detector instead of the traditional total light intensity value for calculation, so as to obtain a higher signal-to-noise ratio. This method uses the differential value after the bucket detector to replace the original light intensity value, thereby removing the average intensity fluctuation of the light field, significantly suppressing background noise and extracting effective object information.

[0094] In this embodiment, Figure 1 As shown, a correlation imaging method based on transfer learning includes the following steps:

[0095] Step 1: Construct a trainable speckle network based on differential correlation imaging and train the network to obtain optimized speckles;

[0096] Speckle is an optical phenomenon usually formed by the random interference pattern of light waves when passing through an inhomogeneous medium. In correlation imaging, speckle is used as a random light field to reconstruct the target image by correlating with the reflected or transmitted light of the target object.

[0097] Step 2: Use the optimized speckle modulated light source to illuminate the target object to obtain the bucket detector value;

[0098] Bucket detector is a detector used in single-pixel imaging technology. Its basic principle is to achieve imaging by receiving total light intensity information. This detector does not have the ability to resolve spatial information, so during the imaging process, it only records the sum of the light intensity rather than the specific light intensity distribution.

[0099] Step 3: Perform differential correlation imaging on the collected bucket signal and the optimized speckle to obtain the initial target pixel matrix;

[0100] Step 4: Convolve the target pixel matrix with the optimized speckle to obtain the estimated bucket signal;

[0101] Step 5: By minimizing the error between the estimated bucket signal and the actual bucket signal, the value of each pixel in the target pixel matrix is ​​updated to quickly restore a high-quality target object image.

[0102] Furthermore, step one specifically includes:

[0103] Step 1.1: Use a trainable convolutional layer to transform the speckle Indicates that the batch size of the input images is set and the differential correlation imaging algorithm is added to the convolutional network;

[0104] In this embodiment, the batch size of the input images is 32.

[0105] Step 1.2: Prepare a dataset consisting of identical images As the input of the network, and divided into training set and test set;

[0106] In this embodiment, the process of making a data set is as follows:

[0107] First, download the public data set STL-10, transform the size of the pictures in the data set, such as transforming them to 128*128 pixels or 64*64 pixels or other sizes. In this embodiment, they are transformed to 128*128 pixels; then divide the data set into a training set and a test set ratio of 8:2; finally, copy both the training set and the test set into two copies, one as the input of the network and the other for calculating the loss function.

[0108] Step 1.3: Make a copy of the dataset to calculate the loss function. The formula is as follows:

[0109] ;

[0110] Convolution operation is performed on each input image to obtain the corresponding bucket signal, and the bucket signal and the convolution layer are differentially correlated to obtain the differential correlation imaging result. The mean square error and structural similarity between the input image and the differential correlation imaging result are used as the loss function. In this embodiment, the setting The initial value of is 0.7, and is fine-tuned according to the training situation to determine the best value.

[0111] Step 1.4: Use the Adam optimizer to train the speckle using the dataset. When the performance does not improve significantly in multiple rounds of testing on the test set, terminate the training and obtain the trained speckle.

[0112] In this embodiment, the Adam optimizer is used for training, the learning rate is set to 0.1, and the learning rate is reduced by 4 times every 10 rounds of training. When the performance is not significantly improved in multiple rounds of testing of the test set, the training is terminated to obtain the optimized speckle.

[0113] Step 1.5: Binarize the trained speckle pattern using the following formula:

[0114] ;

[0115] Since DMD processes binary speckle faster, the optimized speckle is binarized to save training and processing time.

[0116] Step 1.6: Replace the convolutional layer with the binarized speckle, add an additive binarized convolution, and retrain to obtain the optimized speckle.

[0117] In this embodiment, the schematic diagram of the method for obtaining the optimized speckle is as follows: Figure 2 shown.

[0118] The speckle used for binarization is not equal to the speckle, which will lead to poor quality of image reconstruction. In order to solve this problem, the binarized speckle should be compensated to make it as close to the floating point speckle as possible. So replace the convolution layer with the binarized speckle and add an additive binarized convolution. , retrain and obtain the optimized speckle.

[0119] Furthermore, step 2 specifically includes:

[0120] The optimized speckle pattern is loaded into the DMD, and the light source is modulated by the DMD and irradiated to the object. The light passing through the object is received by a bucket detector without spatial resolution to obtain the real bucket signal value.

[0121] DMD (digital micromirror device) is an optical semiconductor module composed of many small aluminum mirrors (micromirrors) that can be independently controlled to form a one-dimensional or two-dimensional array. Each micromirror corresponds to a pixel, and its rotation angle is controlled by electrical addressing to achieve modulation of the optical signal. The working principle of DMD is to use the rotation of the micromirror to change the reflection direction of the light. In this way, the incident light can be binary amplitude modulated.

[0122] Furthermore, step three specifically includes:

[0123] The collected bucket signal value and the optimized speckle are differentially correlated to obtain the initial target pixel matrix. The formula is as follows:

[0124] ;

[0125] ;

[0126] ;

[0127] ;

[0128] Furthermore, step 4 specifically includes: performing a convolution operation on the target pixel matrix and the optimized speckle to obtain an estimated bucket signal, and the formula is as follows:

[0129] ;

[0130] Furthermore, step five specifically includes: updating the value of each pixel of the target pixel matrix by minimizing the error between the estimated bucket signal and the actual bucket signal, and restoring a high-quality target object image. The formula is as follows:

[0131] ;

[0132] ;

[0133] In this embodiment, the principle diagram of the associated imaging is as follows: Figure 3 As shown in the figure, the signal source can select a laser for outputting a light beam; a digital micromirror device is used to modulate the light beam; lenses 1 and 2 are used to converge the light beam; the target object is the object to be imaged; the barrel detector is used to collect the light intensity signal passing through the target object; a computer is used to connect the digital micromirror device and the barrel detector, record the modulated speckle of the digital micromirror device and the signal collected by the barrel detector, and then perform imaging through an algorithm program.

[0134] A schematic diagram of an association imaging method based on transfer learning in this application is shown in Figure 4 As shown in the figure, the specific process is to obtain optimized speckle based on differential correlation imaging. By migrating the optimized speckle and the differential correlation imaging results based on the optimized speckle to unsupervised correlation imaging, it is only necessary to update the pixel matrix to obtain a high-quality target image, without updating millions or even tens of millions of network parameters, which greatly improves the computing efficiency and reduces the requirements for computing resources of terminal devices.

[0135] It should be noted that the present application is not limited to the above-mentioned embodiments. The above-mentioned embodiments are only examples, and the embodiments having the same structure as the technical idea and exerting the same effect within the scope of the technical solution of the present application are all included in the technical scope of the present application. In addition, without departing from the scope of the main purpose of the present application, various modifications that can be thought of by those skilled in the art to the embodiments and other methods of combining some of the constituent elements in the embodiments are also included in the scope of the present application.

Claims

1. A correlation imaging method based on transfer learning, characterized in that: The following steps are involved: Step 1: Construct a trainable speckle network based on differential correlation imaging and train the network to obtain optimized speckles; Step 2: Use the optimized speckle modulated light source to illuminate the target object to obtain the bucket detector value; Step 3: Perform differential correlation imaging on the collected bucket signal and the optimized speckle to obtain the initial target pixel matrix; Step 4: Convolve the target pixel matrix with the optimized speckle to obtain the estimated bucket signal. The formula is as follows: ； in: is the estimated bucket signal; is the optimized speckle at the i-th detection; is the convolution operation; is the target pixel matrix; Step 5: By minimizing the error between the estimated bucket signal and the actual bucket signal, update the value of each pixel in the target pixel matrix and quickly restore the high-quality target object image. The formula is as follows: ； ;in: is the restored high-quality target object image; is the target image that needs to be reconstructed; The parameter value for which the function obtains the minimum value within the domain; The estimated bucket signal when the error between the estimated bucket signal and the real bucket signal is the smallest; is the estimated bucket signal; is the real barrel signal; is the optimized speckle at the i-th detection; is the convolution operation.

2. The associative imaging method based on transfer learning according to claim 1, characterized in that: The step 1 specifically includes: Step 1.1: Use a trainable convolutional layer to transform the speckle Indicates that the batch size of the input images is set and the differential correlation imaging algorithm is added to the convolutional network; Step 1.2: Prepare a dataset consisting of identical images As the input of the network, and divided into training set and test set; Step 1.3: Make a copy of the dataset to calculate the loss function. The formula is as follows: ； in: is the loss function; It is the result of differential correlation imaging after convolution operation on the input image and the convolution layer; For input picture; is the weight of the loss term; is the structural similarity index; is the mean square error; Step 1.4: Use the Adam optimizer to train the speckle using the dataset. When the performance does not improve significantly in multiple rounds of testing on the test set, terminate the training and obtain the trained speckle. Step 1.5: Binarize the trained speckle pattern using the following formula: ； in: is the speckle after binarization; is the sign function, used for positive and negative binarization; is the speckle after training; Step 1.6: Replace the convolutional layer with the binarized speckle, add an additive binarized convolution, and retrain to obtain the optimized speckle.

3. The associative imaging method based on transfer learning according to claim 1, characterized in that: The step 2 specifically includes: The optimized speckle pattern is loaded into the DMD, and the light source is modulated by the DMD and irradiated to the object. The light passing through the object is received by a bucket detector without spatial resolution to obtain the real bucket signal value.

4. The associative imaging method based on transfer learning according to claim 1, characterized in that: The step three specifically includes: The collected bucket signal value and the optimized speckle are differentially correlated to obtain the initial target pixel matrix. The formula is as follows: ； ； ； ； in: is the target pixel matrix; is the total number of sampling times; is the optimized speckle at the i-th detection; is the total light intensity detected by the bucket detector for the i-th time; is the ensemble average of the optimized speckle at the i-th detection; is the ensemble average of the total light intensity detected by the bucket detector for the i-th time; For the i-th The sum of the light intensities; For the i-th The ensemble average of the sum of light intensities.

Images