CN114839789B - Diffraction focusing method and device based on binarization spatial modulation - Google Patents
Diffraction focusing method and device based on binarization spatial modulation Download PDFInfo
- Publication number
- CN114839789B CN114839789B CN202210553808.7A CN202210553808A CN114839789B CN 114839789 B CN114839789 B CN 114839789B CN 202210553808 A CN202210553808 A CN 202210553808A CN 114839789 B CN114839789 B CN 114839789B
- Authority
- CN
- China
- Prior art keywords
- diffraction
- model
- array
- binarization
- evaluation model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000013210 evaluation model Methods 0.000 claims abstract description 62
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000003491 array Methods 0.000 claims abstract description 12
- 238000010276 construction Methods 0.000 claims abstract description 8
- 238000004364 calculation method Methods 0.000 claims abstract description 6
- 238000010801 machine learning Methods 0.000 claims description 7
- 238000003062 neural network model Methods 0.000 claims description 5
- 238000013507 mapping Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 description 8
- 230000008859 change Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 230000003287 optical effect Effects 0.000 description 5
- 230000004913 activation Effects 0.000 description 4
- 238000012937 correction Methods 0.000 description 4
- 238000003384 imaging method Methods 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 3
- 238000001514 detection method Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000000149 argon plasma sintering Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000001066 destructive effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007639 printing Methods 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G02—OPTICS
- G02B—OPTICAL ELEMENTS, SYSTEMS OR APPARATUS
- G02B27/00—Optical systems or apparatus not provided for by any of the groups G02B1/00 - G02B26/00, G02B30/00
- G02B27/42—Diffraction optics, i.e. systems including a diffractive element being designed for providing a diffractive effect
- G02B27/4266—Diffraction theory; Mathematical models
-
- G—PHYSICS
- G02—OPTICS
- G02B—OPTICAL ELEMENTS, SYSTEMS OR APPARATUS
- G02B27/00—Optical systems or apparatus not provided for by any of the groups G02B1/00 - G02B26/00, G02B30/00
- G02B27/0012—Optical design, e.g. procedures, algorithms, optimisation routines
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Optics & Photonics (AREA)
- Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Optical Modulation, Optical Deflection, Nonlinear Optics, Optical Demodulation, Optical Logic Elements (AREA)
Abstract
Description
技术领域Technical field
本发明涉及计算成像技术领域，具体涉及一种基于二值化空间调制的衍射聚焦方法以及装置。The invention relates to the field of computational imaging technology, and in particular to a diffraction focusing method and device based on binary spatial modulation.
背景技术Background technique
利用空间调制的方式将空间中传播的波面分为多个独立单元，对每个单元施加振幅或者相位调制，可以调控后续波面的衍射传播，进而实现在指定位置的聚焦。相比连续相位或者连续振幅调制，二值化调制更容易实现，且调制速度更快。例如对于极紫外光或者X射线都难以进行连续相位调制，但可以通过控制空间中不同位置的“透过”与“不透过”来实现二值化调制，典型的例子是波带片。在可见光波段，基于数字微镜器件的二值化空间调制器可以达到数万赫兹的调制频率，远高于目前基于液晶的连续相位或者连续振幅空间光调制器。Spatial modulation is used to divide the wave surface propagating in space into multiple independent units. Applying amplitude or phase modulation to each unit can control the diffraction propagation of subsequent wave surfaces, thereby achieving focusing at a designated position. Compared with continuous phase or continuous amplitude modulation, binary modulation is easier to implement and the modulation speed is faster. For example, it is difficult to achieve continuous phase modulation for extreme ultraviolet light or X-rays, but binary modulation can be achieved by controlling the "transmission" and "opacity" at different locations in space. A typical example is a zone plate. In the visible light band, binary spatial modulators based on digital micromirror devices can reach modulation frequencies of tens of thousands of Hz, which is much higher than current continuous phase or continuous amplitude spatial light modulators based on liquid crystals.
当波面具有未知的振幅及相位空间分布，或者传播过程中经历散射等非自由传播的过程时，则需要结合波前校正技术来实现聚焦。例如在光学领域，基于逐点尝试或者遗传算法的二值化波前校正技术可以利用数字微镜器件实现光透过散射体的聚焦，但无法灵活改变聚焦位置；如果需要改变聚焦位置则需要重新进行整个波前校正过程。基于传输矩阵测量的二值化波前校正技术在多次测量获得散射传输特性矩阵后可以实现在多个位置的聚焦，但存在计算耗时较长、易受噪声影响等缺点。本发明可以有效实现基于二值化空间调制的聚焦，且实现方式简单、模型构建方法可扩展性强、计算时间短，受噪声影响较小。When the wave surface has unknown amplitude and phase spatial distribution, or experiences non-free propagation processes such as scattering during propagation, wavefront correction technology needs to be combined to achieve focusing. For example, in the field of optics, binary wavefront correction technology based on point-by-point attempts or genetic algorithms can use digital micromirror devices to focus light through scatterers, but it cannot flexibly change the focus position; if the focus position needs to be changed, it needs to be re- Perform the entire wavefront correction process. The binary wavefront correction technology based on transmission matrix measurement can achieve focusing at multiple positions after obtaining the scattering transmission characteristic matrix through multiple measurements, but it has shortcomings such as long calculation time and vulnerability to noise. The present invention can effectively realize focusing based on binary spatial modulation, and has a simple implementation method, strong scalability of the model construction method, short calculation time, and is less affected by noise.
发明内容Contents of the invention
本发明的目的在于提供一种基于二值化空间调制的衍射聚焦方法及装置，具备实现方式简单、模型构建方法可扩展性强、计算时间短，受噪声影响较小等优点。The purpose of the present invention is to provide a diffraction focusing method and device based on binary spatial modulation, which has the advantages of simple implementation, strong scalability of the model construction method, short calculation time, and less influence by noise.
为达到上述技术目的，本发明采取了以下技术方案：In order to achieve the above technical objectives, the present invention adopts the following technical solutions:
第一方面，本发明提供了一种衍射聚焦方法，包括如下步骤：In a first aspect, the present invention provides a diffraction focusing method, including the following steps:
随机生成若干二值化数组并配置二值化空间调制器的调制状态，对入射波波面进行空间调制并获取若干衍射波强度,其中各衍射波强度均在预设的衍射波强度采集范围内获取；Randomly generate several binary arrays and configure the modulation state of the binary spatial modulator, perform spatial modulation on the incident wave surface and obtain several diffraction wave intensities, where each diffraction wave intensity is obtained within the preset diffraction wave intensity collection range ;
构建评价模型，根据选择的待聚焦位置、所述随机二值化数组以及对应的衍射波强度，对所述评价模型进行训练；其中，所述评价模型用于反映二值化数组与衍射波强度中待聚焦位置的衍射波强度的映射关系；Construct an evaluation model, and train the evaluation model according to the selected position to be focused, the random binarized array and the corresponding diffraction wave intensity; wherein the evaluation model is used to reflect the binarized array and diffraction wave intensity The mapping relationship of the diffraction wave intensity at the position to be focused;
构建策略模型，采用训练完备的评价模型对所述策略模型进行训练，其中，策略模型用于产生一个最优数组，所述最优数组作为所述评价模型的输入时可以使评价模型的输出最大化；Construct a strategy model, and use a fully trained evaluation model to train the strategy model. The strategy model is used to generate an optimal array. When the optimal array is used as the input of the evaluation model, the output of the evaluation model can be maximized. change;
对所述最优数组进行二值化并配置二值化空间调制器的调制状态，以实现在待聚焦位置的聚焦。The optimal array is binarized and the modulation state of the binarized spatial modulator is configured to achieve focusing at the position to be focused.
在其中一些实施例中，所述随机生成若干二值化数组并配置二值化空间调制器的调制状态，对入射波波面进行空间调制并获取若干衍射波强度，包括：In some embodiments, the method randomly generates several binarized arrays and configures the modulation state of the binarized spatial modulator, spatially modulates the incident wave surface and obtains several diffraction wave intensities, including:
步骤一、随机生成二值化数组；Step 1: Randomly generate a binary array;
步骤二、根据所述二值化数组改变二值化空间调制器的调制状态；Step 2: Change the modulation state of the binarized spatial modulator according to the binarized array;
步骤三、获取传感器采集的空间调制器之后有效衍射区域内、与所述调制状态对应的入射波波面的衍射波强度；Step 3: Obtain the diffraction wave intensity of the incident wave surface corresponding to the modulation state in the effective diffraction area after the spatial modulator collected by the sensor;
步骤四、重复步骤一至步骤三，得到若干个衍射波强度以及与所述衍射波强度对应的二值化数组。Step 4: Repeat steps 1 to 3 to obtain several diffraction wave intensities and binarized arrays corresponding to the diffraction wave intensities.
在其中一些实施例中，所述构建评价模型，根据选择的待聚焦位置、所述随机二值化数组以及对应的衍射波强度，对所述评价模型进行训练，包括：In some embodiments, the construction of an evaluation model and training of the evaluation model based on the selected position to be focused, the random binarized array and the corresponding diffraction wave intensity include:
选择待聚焦位置，分别获取每一个衍射波强度中待聚焦位置的衍射波强度，以得到若干个训练对，所述训练对为二值化数组与待聚焦位置衍射波强度；Select the position to be focused, and obtain the diffraction wave intensity of the position to be focused in each diffraction wave intensity to obtain several training pairs, where the training pairs are the binarized array and the diffraction wave intensity of the position to be focused;
采用若干个训练对对构建的评价模型进行训练，其中，所述评价模型的输入为二值化数组，所述评价模型的输出为衍射波强度中待聚焦位置的衍射波强度。Several training sessions are used to train the constructed evaluation model, wherein the input of the evaluation model is a binary array, and the output of the evaluation model is the diffraction wave intensity at the position to be focused in the diffraction wave intensity.
在其中一些实施例中，所述构建策略模型，采用训练完备的评价模型对所述策略模型进行训练，包括：In some embodiments, the method of constructing a policy model and using a fully trained evaluation model to train the policy model includes:
构建策略模型，并随机生成一个输入数组；Build a strategy model and randomly generate an input array;
以所述输入数组作为策略模型的输入，以所述策略模型的输出作为评价模型的输入，以所述评价模型的输出结果最大化作为训练目标，根据所述评价模型的输出更新所述策略模型，直至策略模型输出一个最优数组，所述最优数组作为所述评价模型的输入时可以使评价模型的输出最大化。The input array is used as the input of the strategy model, the output of the strategy model is used as the input of the evaluation model, the maximization of the output result of the evaluation model is used as the training goal, and the strategy model is updated according to the output of the evaluation model. , until the policy model outputs an optimal array, which can maximize the output of the evaluation model when used as the input of the evaluation model.
在其中一些实施例中，所述对所述最优数组进行二值化并配置二值化空间调制器的调制状态，以实现在待聚焦位置的聚焦，包括：In some embodiments, binarizing the optimal array and configuring the modulation state of the binarized spatial modulator to achieve focusing at the position to be focused includes:
获取最优数组，并将所述最优数组进行二值化处理，根据二值化处理后的最优数组配置二值化空间调制器的调制状态，以实现待聚焦位置的聚焦。Obtain the optimal array, perform binarization processing on the optimal array, and configure the modulation state of the binarized spatial modulator according to the binarized optimal array to achieve focusing on the position to be focused.
在其中一些实施例中，所述评价模型和所述策略模型为机器学习模型，所述机器学习模型包括但不限于神经网络模型。In some embodiments, the evaluation model and the policy model are machine learning models, and the machine learning models include but are not limited to neural network models.
在其中一些实施例中，当所述选择的待聚焦位置变化时，重新进行评价模型的训练、策略模型的训练、模型输出最优数组的二值化以及二值化空间调制器的配置，即可。In some embodiments, when the selected focus position changes, the training of the evaluation model, the training of the policy model, the binarization of the optimal array of model outputs, and the configuration of the binarized spatial modulator are re-carried out, that is, Can.
第二方面，本发明还提供一种基于二值化空间调制的衍射聚焦装置，包括：In a second aspect, the present invention also provides a diffraction focusing device based on binary spatial modulation, including:
波的产生源，二值化空间调制器、波的强度探测装置，以及执行如上所述的基于二值化空间调制的衍射聚焦方法所需要的计算及存储设备，其中，计算及存储设备执行数组的生成以及模型的构建及训练，并读取强度探测装置的强度以及控制二值化空间调制设备的调制状态，所述波的产生源用于产生入射波。The wave generation source, the binarized spatial modulator, the wave intensity detection device, and the computing and storage equipment required to perform the diffraction focusing method based on the binarized spatial modulation as described above, wherein the computing and storage equipment executes the array Generation and model construction and training, and reading the intensity of the intensity detection device and controlling the modulation state of the binary spatial modulation device, the wave generation source is used to generate the incident wave.
与现有技术相比，本发明提供的衍射聚焦方法以及装置，通过构建评价模型和策略模型，利用评价模型和策略模型可以对二值化空间调制器的调制状态进行优化，可以对难以进行连续相位或者振幅调制的波进行控制，进而完成在指定位置的有效聚焦，此外，随机衍射强度信号采集时不需要指定具体聚焦位置，因此改变不同的聚焦位置时，不需要重新进行随机衍射强度信号采集，节省了流程，而且实现方式简单，模型构建方法可扩展性强，在空间调制单元数较多时仍具有较高的效率。Compared with the existing technology, the diffraction focusing method and device provided by the present invention can optimize the modulation state of the binary spatial modulator by constructing an evaluation model and a strategy model, and can optimize the modulation state of the binary spatial modulator, which is difficult to carry out continuously. Phase or amplitude modulated waves are controlled to achieve effective focusing at a designated position. In addition, there is no need to specify a specific focus position when collecting random diffraction intensity signals. Therefore, when changing different focus positions, there is no need to re-acquire random diffraction intensity signals. , which saves the process and is simple to implement. The model construction method is highly scalable and still has high efficiency when the number of spatial modulation units is large.
附图说明Description of drawings
图1是本发明提供的衍射聚焦方法的一实施例的流程图；Figure 1 is a flow chart of an embodiment of the diffraction focusing method provided by the present invention;
图2是实施例中的装置示意图；Figure 2 is a schematic diagram of the device in the embodiment;
图3是实施例中散射光在所选择聚焦位置的聚焦效果图。Figure 3 is a diagram showing the focusing effect of scattered light at a selected focus position in the embodiment.
具体实施方式Detailed ways
为了使本发明的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本发明进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention and are not intended to limit the present invention.
本发明所涉及的基于二值化空间调制的衍射聚焦方法及装置可用于光学散射聚焦系统中。在光学散射聚焦系统中，由于散射体对入射光进行了复杂且未知的振幅以及相位调制。出射光波经衍射后变为散斑的形式，无法形成有效聚焦。因此，需要把入射光在空间上划分为独立单元，保留能在所选择的聚焦位置上发生相长干涉的单元，去除发生相消干涉的单元，这种单元划分以及是否保留的选择可以由空间光调制器实现,结合本发明的方法可以对每个单元的保留状态进行优化，以实现在所选择聚焦位置的聚焦。The diffraction focusing method and device based on binary spatial modulation involved in the present invention can be used in optical scattering focusing systems. In the optical scattering focusing system, the scatterers perform complex and unknown amplitude and phase modulation of the incident light. The emitted light waves become speckles after diffraction and cannot be effectively focused. Therefore, it is necessary to divide the incident light into independent units in space, retain the units that can constructively interfere at the selected focus position, and remove the units that cause destructive interference. This unit division and the choice of whether to retain it can be determined by the space. Optical modulator implementation, combined with the method of the present invention, allows the retention state of each unit to be optimized to achieve focus at the selected focus position.
根据本发明内容，可以提供一种光的散射聚焦方法，请参阅图1，包括如下步骤：According to the present invention, a light scattering and focusing method can be provided. Please refer to Figure 1, which includes the following steps:
S100、随机生成若干二值化数组并配置二值化空间调制器的调制状态，对入射波波面进行空间调制并获取若干衍射波强度,其中各衍射波强度均在预设的衍射波强度采集范围内获取。S100. Randomly generate several binary arrays and configure the modulation state of the binary spatial modulator, spatially modulate the incident wave surface and obtain several diffraction wave intensities, where each diffraction wave intensity is within the preset diffraction wave intensity collection range. obtained within.
在具体实施时，所述步骤S100具体包括：In specific implementation, step S100 specifically includes:
步骤一、随机生成二值化数组，其中，本实施例中，所述二值化数组为-1，1随机组成的二值化数组，需要理解的是，二值化数组不限于-1,1组成，在其它的实施例中，还可由例如-0.1,0.1组成的方式，本发明实施例对此不做限定；Step 1. Randomly generate a binary array. In this embodiment, the binary array is a binary array randomly composed of -1,1. It should be understood that the binary array is not limited to -1, 1. In other embodiments, it can also be composed of, for example, -0.1, 0.1, which is not limited in the embodiments of the present invention;
步骤二、根据所述二值化数组改变二值化空间调制器的调制状态；Step 2: Change the modulation state of the binarized spatial modulator according to the binarized array;
步骤三、获取传感器采集的位于所述二值化空间调制器之后有效区域内、与所述调制状态对应的入射波波面的衍射波强度；Step 3: Obtain the diffraction wave intensity of the incident wave surface corresponding to the modulation state in the effective area behind the binary spatial modulator and collected by the sensor;
步骤四、重复步骤一至步骤三，得到若干个衍射波强度以及与所述衍射波强度对应的二值化数组。Step 4: Repeat steps 1 to 3 to obtain several diffraction wave intensities and binarized arrays corresponding to the diffraction wave intensities.
具体的，请参阅图2，半导体激光器产生中心波长为635nm的激光，入射至DMD空间光调制器1表面。空间光调制器1照明区被划分为100×100个独立调制单元，每个调制单元可以使照射至其上的光按光路前进或者偏转出光路。然后经聚焦透镜2聚焦后，入射至普通A4打印纸表面3并发生散射，聚焦区域直径约1.5mm。图像传感器4的中心距离聚焦区域约6cm，成像区域大小为5.7mm×4.8mm。计算机5产生2200个由-1，1随机组成的二值数组S,数组大小为100×100，分别控制DMD的各调制单元的调制状态；使用图像传感器4采集散射光强度I，并对散射光强度I进行归一化。Specifically, please refer to Figure 2. The semiconductor laser generates laser light with a central wavelength of 635 nm, which is incident on the surface of the DMD spatial light modulator 1. The illumination area of the spatial light modulator 1 is divided into 100×100 independent modulation units. Each modulation unit can make the light irradiated on it advance according to the optical path or deflect out of the optical path. Then, after being focused by the focusing lens 2, it is incident on the surface 3 of ordinary A4 printing paper and scattered. The diameter of the focused area is about 1.5mm. The center of the image sensor 4 is about 6cm away from the focus area, and the imaging area size is 5.7mm×4.8mm. The computer 5 generates 2200 binary arrays S randomly composed of -1, 1, with an array size of 100×100, to control the modulation state of each modulation unit of the DMD respectively; the image sensor 4 is used to collect the scattered light intensity I, and measure the scattered light Intensity I is normalized.
S200、构建评价模型，根据选择的待聚焦位置、所述随机二值化数组以及对应的衍射波强度，对所述评价模型进行训练；其中，所述评价模型用于反映二值化数组与衍射波强度中待聚焦位置的衍射波强度的映射关系。S200. Construct an evaluation model, and train the evaluation model according to the selected position to be focused, the random binarized array and the corresponding diffraction wave intensity; wherein the evaluation model is used to reflect the relationship between the binarized array and diffraction The mapping relationship of the diffracted wave intensity at the position to be focused in the wave intensity.
在一些实施例中，步骤S200具体包括：In some embodiments, step S200 specifically includes:
选择待聚焦位置，分别获取每一个衍射波强度中待聚焦位置的衍射波强度，以得到若干个训练对，所述训练对为二值化数组与待聚焦位置的衍射波强度；Select the position to be focused, and obtain the diffraction wave intensity of the position to be focused in each diffraction wave intensity to obtain several training pairs, where the training pairs are the binarized array and the diffraction wave intensity of the position to be focused;
采用若干个训练对对构建的评价模型进行训练，其中，所述评价模型的输入为二值化数组，所述评价模型的输出为衍射波强度中待聚焦位置的衍射波强度。Several training sessions are used to train the constructed evaluation model, wherein the input of the evaluation model is a binary array, and the output of the evaluation model is the diffraction wave intensity at the position to be focused in the diffraction wave intensity.
其中，所述评价模型为机器学习模型，所述机器学习模型包括但不限于神经网络模型。Wherein, the evaluation model is a machine learning model, and the machine learning model includes but is not limited to a neural network model.
具体的，在图像传感器成像区选择待聚焦位置，分别计算每一个I中待聚焦位置的衍射信号强度之和Ic，并与其对应的S组成训练数据对；待聚焦位置可以选择图像传感器成像平面任意位置，例如图3所示位置，包含5×5个相邻像素。评价模型采用全连接神经网络作为评价模型，包含输入层、2层隐藏层以及输出层，输出层神经元数量为1，其余各层神经元数量均为64。所述输入层、隐藏层的激活函数为ReLU，输出层不使用激活函数。初始化评价神经网络，以S为输入，计算评价神经网络的输出与Ic之间的误差，使用平方和误差作为误差函数，采用梯度下降方法进行训练。评价神经网络参数采用均方根传播方法进行优化，学习率设置为2×10-3。Specifically, the position to be focused is selected in the imaging area of the image sensor, the sum of the diffraction signal intensities Ic of the position to be focused in each I is calculated, and the training data pair is formed with its corresponding S; the position to be focused can be selected from any imaging plane of the image sensor. A location, such as the location shown in Figure 3, contains 5 × 5 adjacent pixels. The evaluation model uses a fully connected neural network as the evaluation model, including an input layer, two hidden layers and an output layer. The number of neurons in the output layer is 1, and the number of neurons in the other layers is 64. The activation functions of the input layer and hidden layer are ReLU, and the output layer does not use activation functions. Initialize the evaluation neural network, take S as input, calculate the error between the output of the evaluation neural network and Ic, use the sum of square errors as the error function, and use the gradient descent method for training. The evaluation neural network parameters were optimized using the root mean square propagation method, and the learning rate was set to 2×10 -3 .
S300、构建策略模型，采用训练完备的评价模型对所述策略模型进行训练，其中，策略模型用于产生一个最优数组，所述最优数组作为所述评价模型的输入时可以使评价模型的输出最大化。S300. Construct a strategy model, and use a fully trained evaluation model to train the strategy model. The strategy model is used to generate an optimal array. When the optimal array is used as the input of the evaluation model, it can make the evaluation model Maximize output.
在一些实施例中，步骤S300具体包括：In some embodiments, step S300 specifically includes:
构建策略模型，并随机生成一个输入数组；Build a strategy model and randomly generate an input array;
以所述输入数组作为策略模型的输入，以所述策略模型的输出作为评价模型的输入，以所述评价模型的输出结果最大化作为训练目标，根据所述评价模型的输出更新所述策略模型，直至策略模型输出一个最优数组，所述最优数组作为所述训练完备的评价模型的输入时可以使评价模型的输出最大化。The input array is used as the input of the strategy model, the output of the strategy model is used as the input of the evaluation model, the maximization of the output result of the evaluation model is used as the training goal, and the strategy model is updated according to the output of the evaluation model. , until the policy model outputs an optimal array, which can maximize the output of the evaluation model when used as the input of the fully trained evaluation model.
其中，所述策略模型为机器学习模型，所述机器学习模型包括但不限于神经网络模型。在策略模型为神经网络模型时，可采用梯度上升的方法来实现模型的更新，当然，在其它的实施例中，还可采用其它可实现策略模型更新的方法，本发明实施例对此不做限定。Wherein, the policy model is a machine learning model, and the machine learning model includes but is not limited to a neural network model. When the policy model is a neural network model, the gradient ascent method can be used to update the model. Of course, in other embodiments, other methods for updating the policy model can also be used, which is not the case in this embodiment of the present invention. limited.
具体的，使用全连接神经网络为策略模型并进行初始化，包含输入层、2层隐藏层以及输出层，所述输入层、隐藏层的激活函数为ReLU，输出层不使用激活函数；随机生成一个数组为输入，策略神经网络的输出作为评价模型的输入，采用梯度上升方法，以最大化评价模型输出作为训练目标；策略神经网络采用均方根传播方法进行训练，学习率设置为2×10-5。训练完成后策略神经网络的输出数组为最优数组，最优数组可以最大化评价模型的输出。Specifically, a fully connected neural network is used as the policy model and initialized, including an input layer, two hidden layers, and an output layer. The activation functions of the input layer and hidden layer are ReLU, and the output layer does not use an activation function; a randomly generated The array is the input, and the output of the policy neural network is used as the input of the evaluation model. The gradient ascent method is used to maximize the output of the evaluation model as the training goal; the policy neural network is trained using the root mean square propagation method, and the learning rate is set to 2×10 - 5 . After the training is completed, the output array of the policy neural network is the optimal array, and the optimal array can maximize the output of the evaluation model.
S400、对所述最优数组进行二值化并配置二值化空间调制器的调制状态，以实现在待聚焦位置的聚焦。S400: Binarize the optimal array and configure the modulation state of the binarized spatial modulator to achieve focusing at the position to be focused.
具体的，步骤S400具体包括：Specifically, step S400 specifically includes:
获取最优数组，并将所述最优数组进行二值化处理，根据二值化处理后的最优数组配置二值化空间调制器的调制状态，以实现待聚焦位置的聚焦。Obtain the optimal array, perform binarization processing on the optimal array, and configure the modulation state of the binarized spatial modulator according to the binarized optimal array to achieve focusing on the position to be focused.
其中，所述二值化方法为，将最优数组中的正值变为1，负值变为-1。Wherein, the binarization method is to change the positive value in the optimal array to 1 and the negative value to -1.
基于上述基于二值化空间调制的衍射聚焦方法，本发明还相应的提供一种基于二值化空间调制的衍射聚焦装置，包括波的产生源，二值化空间调制器、波的强度探测装置，以及执行如上述各实施例所述的基于二值化空间调制的衍射聚焦方法所需要的计算及存储设备，其中，计算及存储设备执行数组的生成以及模型的构建及训练，并读取强度探测装置的强度以及控制二值化空间调制设备的调制状态，所述波的产生源用于产生入射波。Based on the above diffraction focusing method based on binary spatial modulation, the present invention also provides a diffraction focusing device based on binary spatial modulation, including a wave generation source, a binary spatial modulator, and a wave intensity detection device. , and the computing and storage devices required to perform the diffraction focusing method based on binary spatial modulation as described in the above embodiments, wherein the computing and storage devices perform the generation of arrays and the construction and training of models, and read the intensity Detecting the intensity of the device and controlling the modulation state of the binarized spatial modulation device, said wave generation source is used to generate the incident wave.
由于上文已对基于二值化空间调制的光散射聚焦方法以及装置进行详细描述，在此不再赘述。Since the light scattering focusing method and device based on binary spatial modulation have been described in detail above, they will not be described again here.
以上已将本发明做一详细说明，以上所述，仅为本发明之实施例而已，当不能限定本发明实施范围，即凡依本申请范围所作均等变化与修饰，皆应仍属本发明涵盖范围内。The present invention has been described in detail above. The above are only examples of the present invention. They cannot limit the scope of the present invention. That is, all equivalent changes and modifications made according to the scope of the present application should still be covered by the present invention. within the range.
Claims (8)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210553808.7A CN114839789B (en) | 2022-05-20 | 2022-05-20 | Diffraction focusing method and device based on binarization spatial modulation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210553808.7A CN114839789B (en) | 2022-05-20 | 2022-05-20 | Diffraction focusing method and device based on binarization spatial modulation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114839789A CN114839789A (en) | 2022-08-02 |
| CN114839789B true CN114839789B (en) | 2023-09-12 |
Family
ID=82572664
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210553808.7A Active CN114839789B (en) | 2022-05-20 | 2022-05-20 | Diffraction focusing method and device based on binarization spatial modulation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114839789B (en) |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103299166A (en) * | 2010-10-27 | 2013-09-11 | 康奈尔大学 | Light field image sensor, method and applications |
| CN109459852A (en) * | 2019-01-07 | 2019-03-12 | 史丽 | A kind of scattering medium optical imaging method neural network based |
| CN109528161A (en) * | 2018-11-06 | 2019-03-29 | 浙江大学 | A kind of high speed and high resoltuion scanning microscopy imaging System and method for based on machine learning |
| DE102018108628A1 (en) * | 2018-04-11 | 2019-10-17 | Stiftung Caesar Center Of Advanced European Studies And Research | Method and system for generating a desired light distribution when passing through a diffuser |
| CN111354055A (en) * | 2020-03-13 | 2020-06-30 | 西南科技大学 | Optical coherence tomography depth reconstruction method based on depth learning |
| CN111552069A (en) * | 2020-04-21 | 2020-08-18 | 中国人民解放军国防科技大学 | A method and system for automatic focusing of microscopic images based on deep reinforcement learning |
| WO2020244952A1 (en) * | 2019-06-07 | 2020-12-10 | Ecole Polytechnique Federale De Lausanne (Epfl) | System and method for projecting images through scattering media |
| CN113296258A (en) * | 2021-04-29 | 2021-08-24 | 中国科学院西安光学精密机械研究所 | Focusing light spot scanning endoscopic imaging system and method based on multimode fiber coupler |
| CN113574438A (en) * | 2019-02-03 | 2021-10-29 | 巴伊兰大学 | System and method for imaging through scattering media |
| CN113985566A (en) * | 2021-09-10 | 2022-01-28 | 西南科技大学 | Scattered light focusing method based on spatial light modulation and neural network |
| CN114118361A (en) * | 2021-10-29 | 2022-03-01 | 北京宇航系统工程研究所 | Situation assessment method based on deep learning parameter anchoring |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6545790B2 (en) * | 1999-11-08 | 2003-04-08 | Ralph W. Gerchberg | System and method for recovering phase information of a wave front |
| US8930145B2 (en) * | 2010-07-28 | 2015-01-06 | Covidien Lp | Light focusing continuous wave photoacoustic spectroscopy and its applications to patient monitoring |
| EP3797384A4 (en) * | 2018-05-22 | 2022-03-16 | Ramot at Tel-Aviv University Ltd. | Method and system for imaging and image processing |
| FR3114155B1 (en) * | 2020-09-15 | 2022-07-29 | Supersonic Imagine | METHOD AND SYSTEM FOR ULTRASOUND CHARACTERIZATION OF A MEDIUM |
-
2022
- 2022-05-20 CN CN202210553808.7A patent/CN114839789B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103299166A (en) * | 2010-10-27 | 2013-09-11 | 康奈尔大学 | Light field image sensor, method and applications |
| DE102018108628A1 (en) * | 2018-04-11 | 2019-10-17 | Stiftung Caesar Center Of Advanced European Studies And Research | Method and system for generating a desired light distribution when passing through a diffuser |
| CN109528161A (en) * | 2018-11-06 | 2019-03-29 | 浙江大学 | A kind of high speed and high resoltuion scanning microscopy imaging System and method for based on machine learning |
| CN109459852A (en) * | 2019-01-07 | 2019-03-12 | 史丽 | A kind of scattering medium optical imaging method neural network based |
| CN113574438A (en) * | 2019-02-03 | 2021-10-29 | 巴伊兰大学 | System and method for imaging through scattering media |
| WO2020244952A1 (en) * | 2019-06-07 | 2020-12-10 | Ecole Polytechnique Federale De Lausanne (Epfl) | System and method for projecting images through scattering media |
| CN111354055A (en) * | 2020-03-13 | 2020-06-30 | 西南科技大学 | Optical coherence tomography depth reconstruction method based on depth learning |
| CN111552069A (en) * | 2020-04-21 | 2020-08-18 | 中国人民解放军国防科技大学 | A method and system for automatic focusing of microscopic images based on deep reinforcement learning |
| CN113296258A (en) * | 2021-04-29 | 2021-08-24 | 中国科学院西安光学精密机械研究所 | Focusing light spot scanning endoscopic imaging system and method based on multimode fiber coupler |
| CN113985566A (en) * | 2021-09-10 | 2022-01-28 | 西南科技大学 | Scattered light focusing method based on spatial light modulation and neural network |
| CN114118361A (en) * | 2021-10-29 | 2022-03-01 | 北京宇航系统工程研究所 | Situation assessment method based on deep learning parameter anchoring |
Non-Patent Citations (1)
| Title |
|---|
| 方兆翔.光场波前调控及其在散射介质中聚焦特性的研究.《中国博士学位论文全文数据库 基础科学辑》.2019,(第5期),第1-91页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114839789A (en) | 2022-08-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Luo et al. | Focusing light through scattering media by reinforced hybrid algorithms | |
| CN111386538B (en) | Photonic Neural Network System | |
| CN111582435B (en) | Diffraction depth neural network system based on residual error network | |
| US7512573B2 (en) | Optical processor for an artificial neural network | |
| CN109459852B (en) | Scattering medium optical imaging method based on neural network | |
| CN111832721A (en) | All-optical neural network | |
| Maktoobi et al. | Diffractive coupling for photonic networks: how big can we go? | |
| CN103674264A (en) | Image fusion device and method based on period diffraction correlated imaging | |
| CN109900355B (en) | Imaging method and device | |
| Fang et al. | Particle swarm optimization to focus coherent light through disordered media | |
| CN111812672A (en) | An image reconstruction method and reflection terahertz ghost imaging system | |
| CN114815961A (en) | On-chip optical diffraction computing processor and all-optical image classification device | |
| Goel et al. | Referenceless characterization of complex media using physics-informed neural networks | |
| CN114839789B (en) | Diffraction focusing method and device based on binarization spatial modulation | |
| EP4494038A1 (en) | Method and apparatus for training a neural network | |
| CN113985566B (en) | Scattered light focusing method based on spatial light modulation and neural network | |
| Shi et al. | Imaging consecutive targets through scattering medium and around corners beyond the optical memory effect using untrained network | |
| CN111652372B (en) | Wavefront restoration method and system based on diffractive optical neural network | |
| US20230058418A1 (en) | Sensor-display device | |
| CN116858809A (en) | Method and device for measuring scattering medium transmission matrix based on phase mask enhancement | |
| JP2023527135A (en) | Super Ising emulator with many-body interactions and all-to-all connectivity | |
| CN117114066A (en) | Optical computing topological structure, optical computing topological system and regulating and controlling method of optical computing topological system | |
| CN118364662B (en) | Wavefront sensor construction method based on codec joint optimization | |
| CN114924410B (en) | Focusing method and device based on small phase modulation and phase compensation | |
| Yang et al. | Binary Optical Machine Learning: Million-Scale Physical Neural Networks with Nano Neurons |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |