技术领域technical field
本发明涉及计算机视觉技术领域，尤其涉及一种细胞图像分割方法和装置。The invention relates to the technical field of computer vision, in particular to a cell image segmentation method and device.
背景技术Background technique
细胞分割是医学图像分析处理中最基础、同时也是最为重要的内容之一，是细胞进行识别、计数等后续工作的一个基本前提。细胞图像的分割对定量分析及处理细胞信息、研究细胞变异、以及实现细胞显微或超显微结构的三维重建等具有不可替代的现实意义。Cell segmentation is one of the most basic and important contents in medical image analysis and processing, and it is a basic premise for subsequent work such as cell identification and counting. The segmentation of cell images has irreplaceable practical significance for quantitative analysis and processing of cell information, study of cell variation, and realization of three-dimensional reconstruction of cell microscopic or ultramicroscopic structures.
传统的细胞图像分割方法大致分为两类：基于区域的分割方法以及基于边缘的分割方法。其中，基于区域的分割方法的基本原理是通过把具有相似特征的相邻区域归为一类来实现分割，具有代表性的分割方法有阈值法、区域生长法、以及聚类法等。基于边缘的分割方法一般通过把灰度级或者结构具有突变的地方作为边缘来进行分割，具有代表性的方法有微分算子法、模型法等。Traditional cell image segmentation methods can be roughly divided into two categories: region-based segmentation methods and edge-based segmentation methods. Among them, the basic principle of the region-based segmentation method is to achieve segmentation by grouping adjacent regions with similar characteristics into one category. Representative segmentation methods include threshold method, region growing method, and clustering method. The edge-based segmentation method generally uses the gray level or the place where the structure has a sudden change as the edge to segment, and the representative methods include differential operator method and model method.
然而，上述细胞分割方法对于重叠细胞和粘连细胞的分割效果差，步骤复杂，运算量大，准确率低。However, the above-mentioned cell segmentation methods have poor segmentation effects on overlapping cells and cohesive cells, and the steps are complicated, with a large amount of calculation and low accuracy.
发明内容Contents of the invention
本发明实施例提供一种细胞图像分割方法和装置，用以解决现有的细胞图像分割对于重叠细胞和粘连细胞的分割效果差、步骤复杂、运算量大、准确率低的问题。Embodiments of the present invention provide a method and device for cell image segmentation, which are used to solve the problems of poor segmentation effect, complicated steps, large amount of computation, and low accuracy for overlapping cells and cohesive cells in existing cell image segmentation.
第一方面，本发明实施例提供一种细胞图像分割方法，包括：In a first aspect, an embodiment of the present invention provides a cell image segmentation method, including:
获取待分割细胞图像；Obtain the image of the cell to be segmented;
将所述待分割细胞图像输入至细胞分割模型中，获取所述细胞分割模型输出的细胞分割结果；其中，所述细胞分割模型是基于样本细胞图像和所述样本细胞图像对应的真实标签图像，对增强U-Net网络进行训练得到的；所述增强U-Net网络是在初始U-Net网络中增加网络层数，并在网络中加入BN层的神经网络。The cell image to be segmented is input into the cell segmentation model, and the cell segmentation result output by the cell segmentation model is obtained; wherein, the cell segmentation model is based on the sample cell image and the real label image corresponding to the sample cell image, It is obtained by training the enhanced U-Net network; the enhanced U-Net network is a neural network in which the number of network layers is increased in the initial U-Net network and a BN layer is added to the network.
第二方面，本发明实施例提供一种细胞图像分割装置，包括：In a second aspect, an embodiment of the present invention provides a cell image segmentation device, including:
图像获取单元，用于获取待分割细胞图像；an image acquisition unit, configured to acquire images of cells to be segmented;
图像分割单元，用于将所述待分割细胞图像输入至细胞分割模型中，获取所述细胞分割模型输出的细胞分割结果；其中，所述细胞分割模型是基于样本细胞图像和所述样本细胞图像对应的真实标签图像，对增强U-Net网络进行训练得到的；所述增强U-Net网络是在初始U-Net网络中增加网络层数，并在网络中加入BN层的神经网络。An image segmentation unit, configured to input the image of the cell to be segmented into the cell segmentation model, and obtain the cell segmentation result output by the cell segmentation model; wherein, the cell segmentation model is based on the sample cell image and the sample cell image The corresponding real label image is obtained by training the enhanced U-Net network; the enhanced U-Net network is to increase the number of network layers in the initial U-Net network, and add a BN layer neural network to the network.
第三方面，本发明实施例提供一种电子设备，包括处理器、通信接口、存储器和总线，其中，处理器，通信接口，存储器通过总线完成相互间的通信，处理器可以调用存储器中的逻辑指令，以执行如第一方面所提供的方法的步骤。In a third aspect, an embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a bus, wherein the processor, the communication interface, and the memory complete communication with each other through the bus, and the processor can call the logic in the memory Instructions to execute the steps of the method provided in the first aspect.
第四方面，本发明实施例提供一种非暂态计算机可读存储介质，其上存储有计算机程序，该计算机程序被处理器执行时实现如第一方面所提供的方法的步骤。In a fourth aspect, an embodiment of the present invention provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method provided in the first aspect are implemented.
本发明实施例提供的一种细胞图像分割方法和装置，通过对初始U-Net网络中增加了网络层数，并在网络中加入BN层的神经网络进行训练得到用于细胞分割的细胞分割模型，在基于初始U-Net网络提高重叠细胞和粘连细胞的分割效果，优化细胞分割准确率的同时，基于BN算法加速模型训练、优化模型性能。将待分割细胞图像输入至由此训练得到的细胞分割模型，即可实现快速、准确、简便的细胞分割。A cell image segmentation method and device provided by the embodiments of the present invention obtain a cell segmentation model for cell segmentation by increasing the number of network layers in the initial U-Net network and adding a BN layer neural network to the network for training , while improving the segmentation effect of overlapping cells and cohesive cells based on the initial U-Net network and optimizing the accuracy of cell segmentation, the BN algorithm is used to accelerate model training and optimize model performance. By inputting the image of the cell to be segmented into the cell segmentation model trained in this way, fast, accurate and simple cell segmentation can be realized.
附图说明Description of drawings
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作一简单地介绍，显而易见地，下面描述中的附图是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description These are some embodiments of the present invention. Those skilled in the art can also obtain other drawings based on these drawings without creative work.
图1为本发明实施例提供的细胞图像分割方法的流程示意图；FIG. 1 is a schematic flow chart of a cell image segmentation method provided by an embodiment of the present invention;
图2为本发明另一实施例提供的细胞图像分割方法的流程示意图；Fig. 2 is a schematic flowchart of a cell image segmentation method provided by another embodiment of the present invention;
图3为本发明实施例提供的待分割细胞图像；Fig. 3 is the cell image to be segmented provided by the embodiment of the present invention;
图4为本发明实施例提供的细胞分割结果示意图；Fig. 4 is a schematic diagram of cell segmentation results provided by the embodiment of the present invention;
图5为本发明实施例提供的细胞图像分割装置的结构示意图；5 is a schematic structural diagram of a cell image segmentation device provided by an embodiment of the present invention;
图6为本发明实施例提供的电子设备的结构示意图。FIG. 6 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention.
具体实施方式Detailed ways
为使本发明实施例的目的、技术方案和优点更加清楚，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments It is a part of embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.
针对现有的细胞分割方法对于重叠细胞和粘连细胞的分割效果差、步骤复杂、运算量大和准确率低的问题，本发明实施例提供一种细胞图像分割方法。图1为本发明实施例提供的细胞图像分割方法的流程示意图，如图1所示，该方法包括：Aiming at the problems that the existing cell segmentation methods have poor segmentation effect on overlapping cells and cohesive cells, complex steps, heavy computation and low accuracy, the embodiment of the present invention provides a cell image segmentation method. Fig. 1 is a schematic flow chart of the cell image segmentation method provided by the embodiment of the present invention, as shown in Fig. 1, the method includes:
步骤110，获取待分割细胞图像。Step 110, acquiring the image of the cell to be segmented.
此处，待分割细胞图像即需要进行分割的细胞图像。Here, the cell image to be segmented is the cell image that needs to be segmented.
步骤120，将待分割细胞图像输入至细胞分割模型中，获取细胞分割模型输出的细胞分割结果；其中，细胞分割模型是基于样本细胞图像和样本细胞图像对应的真实标签图像，对增强U-Net网络进行训练得到的；增强U-Net网络是在初始U-Net网络中增加网络层数，并在网络中加入BN层的神经网络。Step 120, input the image of the cell to be segmented into the cell segmentation model, and obtain the cell segmentation result output by the cell segmentation model; wherein, the cell segmentation model is based on the sample cell image and the real label image corresponding to the sample cell image, and the enhanced U-Net The network is trained; the enhanced U-Net network is to increase the number of network layers in the initial U-Net network, and add a BN layer neural network to the network.
具体地，用于对输入的待分割细胞图像进行分割，生成细胞分割结果并输出。此处，细胞分割结果是细胞分割模型对待分割细胞图像进行分割得到的结果。Specifically, it is used to segment the input cell image to be segmented, generate and output the cell segmentation result. Here, the cell segmentation result is the result obtained by segmenting the image of the cell to be segmented by the cell segmentation model.
此处，细胞分割模型是基于样本细胞图像和样本细胞图像对应的真实标签图像，对增强U-Net网络进行训练得到的。样本细胞图像与真实标签图像一一对应，真实标签图像是对样本细胞图像进行分割得到的。Here, the cell segmentation model is obtained by training the enhanced U-Net network based on the sample cell image and the real label image corresponding to the sample cell image. There is a one-to-one correspondence between the sample cell image and the real label image, and the real label image is obtained by segmenting the sample cell image.
增强U-Net网络是结合了BN(Batch Normalization)算法的U-Net网络，即在初始U-Net网络的基础上，扩增了网络层数，并通过BN算法标准化处理初始U-Net网络的部分隐层的输入数据，即在初始U-Net网络的部分隐层之前加设一层BN层。The enhanced U-Net network is a U-Net network combined with the BN (Batch Normalization) algorithm, that is, on the basis of the initial U-Net network, the number of network layers is expanded, and the initial U-Net network is standardized through the BN algorithm. The input data of the partial hidden layer, that is, add a BN layer before the partial hidden layer of the initial U-Net network.
其中，初始U-Net网络即预先选取的常规U-Net网络。U-Net网络是卷积神经网络的一种变形，主要由两部分组成：收缩路径(contracting path)和扩展路径(expandingpath)。收缩路径主要是用来捕捉图片中的上下文信息(context information)，而与之相对称的扩展路径则是为了对图片中所需要分割出来的部分进行精准定位(localization)。U-Net网络为实现精准的定位，收缩路径上提取出来的高像素特征会在升采样(upsampling)过程中与新的特征图(feature map)进行结合，以最大程度的保留前面降采样(downsampling)过程一些重要的特征信息；而为了实现网络结构的高效的运行，U-Net网络中去除了全连接层，以很大程度上减少需要训练的参数。此外，得益于特殊的U形结构，U-Net网络可以很好的保留图片中的所有信息。Among them, the initial U-Net network is a pre-selected conventional U-Net network. The U-Net network is a variant of the convolutional neural network, which is mainly composed of two parts: contracting path and expanding path. The contraction path is mainly used to capture the context information in the picture, and the corresponding expansion path is to accurately localize the part of the picture that needs to be segmented. In order to achieve precise positioning in the U-Net network, the high-pixel features extracted on the contraction path will be combined with the new feature map in the upsampling process to preserve the previous downsampling (downsampling) to the greatest extent. ) process some important feature information; and in order to achieve efficient operation of the network structure, the fully connected layer is removed from the U-Net network to greatly reduce the parameters that need to be trained. In addition, thanks to the special U-shaped structure, the U-Net network can well retain all the information in the picture.
为了缓解U-Net网络训练时，网络中间层的数据分布发生改变的问题，本发明实施例应用BN算法标准化处理初始U-Net网络的部分隐层的输入数据，并将处理后的数据输入至下一层中，以提高了网络训练的速度及收敛性能。In order to alleviate the problem that the data distribution of the middle layer of the network changes during U-Net network training, the embodiment of the present invention applies the BN algorithm to standardize the input data of some hidden layers of the initial U-Net network, and inputs the processed data to In the next layer, the speed and convergence performance of network training are improved.
本发明实施例提供的方法，通过对初始U-Net网络中增加了网络层数，并在网络中加入BN层的神经网络进行训练得到用于细胞分割的细胞分割模型，在基于初始U-Net网络提高重叠细胞和粘连细胞的分割效果，优化细胞分割准确率的同时，基于BN算法加速模型训练、优化模型性能。将待分割细胞图像输入至由此训练得到的细胞分割模型，即可实现快速、准确、简便的细胞分割。In the method provided by the embodiment of the present invention, the cell segmentation model for cell segmentation is obtained by increasing the number of network layers in the initial U-Net network and adding the neural network of the BN layer to the network. Based on the initial U-Net The network improves the segmentation effect of overlapping cells and cohesive cells, optimizes the accuracy of cell segmentation, and at the same time accelerates model training and optimizes model performance based on the BN algorithm. By inputting the image of the cell to be segmented into the cell segmentation model trained in this way, fast, accurate and simple cell segmentation can be realized.
基于上述实施例，该方法中，步骤120之前还包括：步骤100，基于样本细胞图像和样本细胞图像对应的真实标签图像，对增强U-Net网络进行训练。Based on the above embodiment, before step 120, the method further includes: step 100, based on the sample cell image and the real label image corresponding to the sample cell image, the enhanced U-Net network is trained.
细胞分割模型具体可通过如下方式训练得到：首先，收集大量样本细胞图像和样本细胞图像对应的真实标签图像；其中，样本细胞图像和真实标签图像一一对应，真实标签图像可以是由专家手工分割得到的。随即基于样本语料标注任务和样本标注结果对在增加了网络层数、并在网络中的部分隐层前加设了BN层的初始U-Net网络进行训练，从而得到细胞分割模型。本发明实施例不对初始U-Net网络的规模作具体限定。The cell segmentation model can be trained in the following way: First, collect a large number of sample cell images and real label images corresponding to the sample cell images; among them, the sample cell images correspond to the real label images one by one, and the real label images can be manually segmented by experts owned. Then, based on the sample corpus labeling task and sample labeling results, the initial U-Net network with increased network layers and a BN layer added before some hidden layers in the network was trained to obtain a cell segmentation model. The embodiment of the present invention does not specifically limit the scale of the initial U-Net network.
基于上述任一实施例，该方法中，步骤100之前还包括：获取样本细胞图像和样本细胞图像的对应的真实标签图像：对任一数据集细胞图像和该数据集细胞图像的数据集标签图像进行合并，获取数据集合并图像；对数据集合并图像进行数据增强，得到数据集增强图像；数据增强包括几何变换、随机弹性形变、随机缩放、逆时针剪切变换中的至少一种；将数据集增强图像进行图像分离，得到样本细胞图像和样本细胞图像的真实标签图像。Based on any of the above-mentioned embodiments, in the method, before step 100, it also includes: acquiring the sample cell image and the corresponding real label image of the sample cell image: for any dataset cell image and the dataset label image of the dataset cell image Perform merging to obtain the merged image of the data set; perform data enhancement on the merged image of the data set to obtain an enhanced image of the data set; data enhancement includes at least one of geometric transformation, random elastic deformation, random scaling, and counterclockwise shear transformation; the data The enhanced image is collected for image separation, and the sample cell image and the real label image of the sample cell image are obtained.
具体地，用于细胞图像分割训练的训练集与其余领域的训练集相比，其获取成本无论在时间上还是在资源的消耗上都更大。为了解决训练集过小的问题，需要针对已获取的数据集细胞图像和数据集标签图像进行扩增。Specifically, the acquisition cost of the training set used for cell image segmentation training is greater in terms of time and resource consumption compared with the training set of other fields. In order to solve the problem that the training set is too small, it is necessary to amplify the acquired dataset cell images and dataset label images.
此处，数据集细胞图像是从当前公开的数据集中获取的细胞图像，数据集标签图像为该数据集细胞图像的分割结果。Here, the dataset cell image is the cell image obtained from the currently public dataset, and the dataset label image is the segmentation result of the dataset cell image.
针对任一数据集细胞图像及其数据集标签图像，首先对任一数据集细胞图像及其数据集标签图像进行合并，获取合并后的图像即数据集合并图像。数据集合并图像中既包含有数据集细胞图像的特征，也包含有数据集标签图像的特征。随即对数据集合并图像进行数据增强，得到数据增强后的数据集合并图像，即数据集增强图像。数据集增强图像后，包含经过数据增强的数据集细胞图像和数据集标签图像，且数据集细胞图像和数据集标签图像经历的数据增强操作完全一致。在完成数据增强后，对数据集增强图像进行图像分离，得到数据集增强图像分离后的细胞图像和标签图像，并将分离后的细胞图像和标签图像作为样本细胞图像和真实标签图像。即，针对任一数据集细胞图像及其数据集分割结果进行合并、数据增强和分离，即可得到新的样本细胞图像和真实标签图像，进而实现训练集的扩增。For any data set cell image and its data set label image, firstly, any data set cell image and its data set label image are merged, and the merged image is obtained as the data set merged image. The merged image of the data set contains not only the features of the cell image of the data set, but also the features of the label image of the data set. Immediately, data augmentation is performed on the merged image of the dataset to obtain a merged image of the dataset after data augmentation, that is, the augmented image of the dataset. After the dataset image is enhanced, it contains the dataset cell image and the dataset label image after data enhancement, and the data enhancement operation undergone by the dataset cell image and the dataset label image is exactly the same. After the data enhancement is completed, image separation is performed on the enhanced image of the dataset to obtain the separated cell image and label image of the enhanced image of the dataset, and the separated cell image and label image are used as the sample cell image and the real label image. That is, by merging, data enhancement and separation for any data set cell image and its data set segmentation results, a new sample cell image and a real label image can be obtained, thereby realizing the expansion of the training set.
基于上述任一实施例，该方法中，步骤100之前还包括：对样本细胞图像进行归一化处理。Based on any of the above embodiments, in the method, before step 100, the method further includes: performing normalization processing on the sample cell image.
具体地，在将大量样本细胞图像应用于细胞分割模型训练之前，需要对每一样本细胞图像进行0-1归一化处理，以便于加速模型收敛。Specifically, before applying a large number of sample cell images to cell segmentation model training, it is necessary to perform 0-1 normalization processing on each sample cell image in order to accelerate model convergence.
基于上述任一实施例，该方法中，步骤100之前还包括：基于高斯分布初始化方法对进行初始化。Based on any of the above-mentioned embodiments, in the method, before step 100, the method further includes: initializing based on a Gaussian distribution initialization method.
具体地，在训练细胞分割模型之前，需要对增强U-Net网络参数进行初始化。高斯分布(Gaussian)初始化需要根据预先设定的高斯函数的均值和标准差生成的高斯分布进行参数初始化配置。本发明实施例提供的方法中，参数由高斯分布产生，其中，n是权重张量的扇入。此外，迭代轮次epoch＝30，批量样本batchsize＝4，学习率η＝1e-4。Specifically, before training the cell segmentation model, the network parameters of the enhanced U-Net need to be initialized. Gaussian distribution (Gaussian) initialization requires parameter initialization configuration based on the Gaussian distribution generated by the pre-set Gaussian function mean and standard deviation. In the method provided by the embodiment of the present invention, the parameter is distributed by Gaussian yields, where n is the fan-in of the weight tensor. In addition, the iteration round epoch=30, the batch sample batchsize=4, and the learning rate η=1e-4.
基于上述任一实施例，该方法中，步骤100具体包括：Based on any of the above-mentioned embodiments, in this method, step 100 specifically includes:
步骤101，将样本细胞图像和样本细胞图像对应的真实标签图像输入至增强U-Net网络，获取增强U-Net网络输出的训练细胞分割结果。Step 101, input the sample cell image and the real label image corresponding to the sample cell image to the enhanced U-Net network, and obtain the training cell segmentation result output by the enhanced U-Net network.
此处，训练细胞分割结果是训练过程中的增强U-Net网络对输入的样本细胞图像进行分割得到的分割结果。Here, the training cell segmentation result is the segmentation result obtained by segmenting the input sample cell image by the enhanced U-Net network during the training process.
步骤102，将真实标签图像和训练细胞分割结果输入至对数损失函数，获取损失函数值。Step 102, input the real label image and training cell segmentation results into the logarithmic loss function to obtain the loss function value.
此处，对数损失函数为binary_cross_entropy，binary_cross_entropy是二分类的交叉熵，此处的二分类是指是否为细胞边缘。Here, the logarithmic loss function is binary_cross_entropy, binary_cross_entropy is the cross entropy of the binary classification, and the binary classification here refers to whether it is the edge of the cell.
步骤103，基于损失函数值，通过Adam优化算法对增强U-Net网络进行参数调节。Step 103, based on the value of the loss function, adjust the parameters of the enhanced U-Net network through the Adam optimization algorithm.
Adam是一种可以替代传统随机梯度下降过程的一阶优化算法，它能基于训练数据迭代地更新神经网络权重。使用Adam优化算法最小化损失函数值，以更新增强U-Net网络权重，直至迭代次数超过预先设置的迭代轮次epoch。Adam is a first-order optimization algorithm that can replace the traditional stochastic gradient descent process, which can iteratively update the neural network weights based on the training data. Use the Adam optimization algorithm to minimize the loss function value to update the enhanced U-Net network weights until the number of iterations exceeds the preset iteration round epoch.
基于上述任一实施例，该方法中，步骤120之前还包括：将测试细胞图像输入至细胞分割模型中，获取细胞分割模型输出的测试细胞分割结果；若测试细胞图像对应的基准细胞分割结果与测试细胞分割结果不同，则基于测试细胞图像和基准细胞分割结果训练细胞分割模型。Based on any of the above-mentioned embodiments, in the method, before step 120, the method further includes: input the test cell image into the cell segmentation model, and obtain the test cell segmentation result output by the cell segmentation model; if the reference cell segmentation result corresponding to the test cell image is the same as If the test cell segmentation results are different, the cell segmentation model is trained based on the test cell image and the benchmark cell segmentation results.
具体地，在应用细胞分割模型进行待分割细胞图像的分割之前，需要对训练好的细胞分割模型进行测试验证。此处，测试细胞图像是用于对细胞分割模型进行测试验证的细胞图像，基准细胞分割结果是预先通过专家对测试细胞图像进行分割得到的分割结果。测试细胞分割结果是细胞分割模型输出的分割结果。在基准细胞分割结果与测试细胞分割结果不同时，基于测试细胞图像和基准细胞分割结果训练细胞分割模型，以进一步优化模型性能。Specifically, before applying the cell segmentation model to segment the image of the cell to be segmented, it is necessary to test and verify the trained cell segmentation model. Here, the test cell image is a cell image used to test and verify the cell segmentation model, and the benchmark cell segmentation result is a segmentation result obtained by segmenting the test cell image by an expert in advance. The test cell segmentation result is the segmentation result output by the cell segmentation model. When the baseline cell segmentation results are different from the test cell segmentation results, the cell segmentation model is trained based on the test cell images and the baseline cell segmentation results to further optimize the model performance.
基于上述任一实施例，图2为本发明另一实施例提供的细胞图像分割方法的流程示意图，如图2所示，该方法包括：Based on any of the above embodiments, Fig. 2 is a schematic flowchart of a cell image segmentation method provided by another embodiment of the present invention, as shown in Fig. 2, the method includes:
步骤210，获取ISBI2012竞赛组委会提供的一个公开数据集。该数据集中显微细胞图像来源于果蝇第一龄幼虫的腹神经细胞。该数据集中的显微细胞图像即数据集细胞图像。将该数据集中的26幅图像作为训练集中的样本细胞图像，并针对上述26幅图像分别进行数据扩增，以扩大训练集的规模。此处，数据扩增需要满足：(1)扩增所得的样本细胞图像于原始的数据集细胞图像独立同分布或近似独立同分布；(2)扩增所得的样本细胞图像保留原图像的重要特征。Step 210, obtaining a public data set provided by the organizing committee of the ISBI2012 competition. The microscopic cell images in this data set are derived from the ventral nerve cells of the first instar larvae of Drosophila. The microscopic cell images in this data set are the data set cell images. The 26 images in the data set were used as sample cell images in the training set, and data amplification was performed on the above 26 images to expand the size of the training set. Here, the data amplification needs to meet: (1) the amplified sample cell image is independent and identically distributed or approximately independent and identically distributed with the original data set cell image; (2) the amplified sample cell image retains the importance of the original image feature.
针对任一数据集细胞图像，数据扩增步骤如下：For any data set cell image, the data amplification steps are as follows:
对任一数据集细胞图像和该数据集细胞图像对应的数据集标签图像进行合并，获取数据集合并图像；对数据集合并图像进行数据增强，得到数据集增强图像；将数据集增强图像进行图像分离，得到样本细胞图像和样本细胞图像对应的真实标签图像。Merge the cell image of any dataset and the dataset tag image corresponding to the cell image of the dataset to obtain the merged image of the dataset; perform data enhancement on the merged image of the dataset to obtain the enhanced image of the dataset; perform image processing on the enhanced image of the dataset Separated to obtain the sample cell image and the real label image corresponding to the sample cell image.
步骤220，对步骤210数据扩增所得的训练集中的每一样本细胞图像进行预处理，通过0-1归一化处理将每一样本细胞图像转换为大小相同的矩阵输出。Step 220 , perform preprocessing on each sample cell image in the training set obtained from the data amplification in step 210 , and convert each sample cell image into a matrix output with the same size through 0-1 normalization processing.
步骤230，构建增强U-Net网络。增强U-Net网络是在初始U-Net网络中增加网络层数，并在网络中加入BN层的神经网络。在完成的构建后，基于高斯分布初始化方法对增强U-Net网络进行初始化。Step 230, constructing an enhanced U-Net network. The enhanced U-Net network is a neural network that increases the number of network layers in the initial U-Net network and adds a BN layer to the network. After the completed construction, the enhanced U-Net network is initialized based on the Gaussian distribution initialization method.
步骤240，基于样本细胞图像和样本细胞图像对应的真实标签图像，对增强U-Net网络进行训练。此处，训练所用的损失函数为对数损失函数binary_cross_entropy，使用Adam优化算法最小化损失函数，更新模型权重，将训练完成的增强U-Net网络作为细胞分割模型。Step 240, based on the sample cell image and the real label image corresponding to the sample cell image, the enhanced U-Net network is trained. Here, the loss function used for training is the logarithmic loss function binary_cross_entropy, the Adam optimization algorithm is used to minimize the loss function, the model weight is updated, and the trained enhanced U-Net network is used as the cell segmentation model.
步骤250，将如图3所示的待分割细胞图像输入至细胞分割模型，获取细胞分割模型输出的如图4所示的细胞分割结果。Step 250, input the image of the cell to be segmented as shown in FIG. 3 into the cell segmentation model, and obtain the cell segmentation result output by the cell segmentation model as shown in FIG. 4 .
本发明实施例提供的方法，通过对初始U-Net网络中增加了网络层数，并在网络中加入BN层的神经网络进行训练得到用于细胞分割的细胞分割模型，在基于初始U-Net网络提高重叠细胞和粘连细胞的分割效果，优化细胞分割准确率的同时，基于BN算法加速模型训练、优化模型性能。将待分割细胞图像输入至由此训练得到的细胞分割模型，即可实现快速、准确、简便的细胞分割。In the method provided by the embodiment of the present invention, the cell segmentation model for cell segmentation is obtained by increasing the number of network layers in the initial U-Net network and adding the neural network of the BN layer to the network. Based on the initial U-Net The network improves the segmentation effect of overlapping cells and cohesive cells, optimizes the accuracy of cell segmentation, and at the same time accelerates model training and optimizes model performance based on the BN algorithm. By inputting the image of the cell to be segmented into the cell segmentation model trained in this way, fast, accurate and simple cell segmentation can be realized.
基于上述任一实施例，图5为本发明实施例提供的细胞图像分割装置的结构示意图，如图5所示，该装置包括图像获取单元510和图像分割单元520；Based on any of the above-mentioned embodiments, FIG. 5 is a schematic structural diagram of a cell image segmentation device provided by an embodiment of the present invention. As shown in FIG. 5 , the device includes an image acquisition unit 510 and an image segmentation unit 520;
其中，图像获取单元510用于获取待分割细胞图像；Wherein, the image acquisition unit 510 is used to acquire the image of the cell to be segmented;
图像分割单元520用于将所述待分割细胞图像输入至细胞分割模型中，获取所述细胞分割模型输出的细胞分割结果；其中，所述细胞分割模型是基于样本细胞图像和所述样本细胞图像对应的真实标签图像，对增强U-Net网络进行训练得到的；所述增强U-Net网络是在初始U-Net网络中增加了网络层数，并在网络中加入BN层的神经网络。The image segmentation unit 520 is configured to input the image of the cell to be segmented into the cell segmentation model, and obtain the cell segmentation result output by the cell segmentation model; wherein, the cell segmentation model is based on the sample cell image and the sample cell image The corresponding real label image is obtained by training the enhanced U-Net network; the enhanced U-Net network is a neural network in which the number of network layers is increased in the initial U-Net network and a BN layer is added to the network.
本发明实施例提供的装置，通过对初始U-Net网络中增加了网络层数，并在网络中加入BN层的神经网络进行训练得到用于细胞分割的细胞分割模型，在基于初始U-Net网络提高重叠细胞和粘连细胞的分割效果，优化细胞分割准确率的同时，基于BN算法加速模型训练、优化模型性能。将待分割细胞图像输入至由此训练得到的细胞分割模型，即可实现快速、准确、简便的细胞分割。The device provided by the embodiment of the present invention obtains a cell segmentation model for cell segmentation by increasing the number of network layers in the initial U-Net network and adding a BN layer neural network to the network. The network improves the segmentation effect of overlapping cells and cohesive cells, optimizes the accuracy of cell segmentation, and at the same time accelerates model training and optimizes model performance based on the BN algorithm. By inputting the image of the cell to be segmented into the cell segmentation model trained in this way, fast, accurate and simple cell segmentation can be realized.
基于上述任一实施例，该装置还包括训练单元；Based on any of the above embodiments, the device further includes a training unit;
训练单元用于基于所述样本细胞图像和所述样本细胞图像对应的真实标签图像，对所述增强U-Net网络进行训练。The training unit is used to train the enhanced U-Net network based on the sample cell image and the real label image corresponding to the sample cell image.
基于上述任一实施例，该装置还包括数据扩增单元；数据扩增单元用于：Based on any of the above embodiments, the device also includes a data amplification unit; the data amplification unit is used for:
对任一数据集细胞图像和所述任一数据集细胞图像对应的数据集标签图像进行合并，获取数据集合并图像；Merge the cell image of any data set and the data set label image corresponding to the cell image of any data set, and obtain the merged image of the data set;
对所述数据集合并图像进行数据增强，得到数据集增强图像；所述数据增强包括几何变换、随机弹性形变、随机缩放、逆时针剪切变换中的至少一种；Performing data enhancement on the merged image of the data set to obtain an enhanced image of the data set; the data enhancement includes at least one of geometric transformation, random elastic deformation, random scaling, and counterclockwise shear transformation;
将所述数据集增强图像进行图像分离，得到所述样本细胞图像和所述样本细胞图像对应的真实标签图像。Image separation is performed on the enhanced image of the data set to obtain the sample cell image and the real label image corresponding to the sample cell image.
基于上述任一实施例，该装置还包括归一化单元；Based on any of the above embodiments, the device further includes a normalization unit;
归一化单元用于对所述样本细胞图像进行归一化处理。The normalization unit is used for normalizing the sample cell image.
基于上述任一实施例，该装置还包括初始化单元；Based on any of the above embodiments, the device further includes an initialization unit;
初始化单元用于基于高斯分布初始化方法对所述增强U-Net网络进行初始化。The initialization unit is used for initializing the enhanced U-Net network based on a Gaussian distribution initialization method.
基于上述任一实施例，该装置中，训练单元具体用于：Based on any of the above embodiments, in the device, the training unit is specifically used for:
将所述样本细胞图像输入至所述增强U-Net网络，获取所述增强U-Net网络输出的训练细胞分割结果；The sample cell image is input to the enhanced U-Net network, and the training cell segmentation result output by the enhanced U-Net network is obtained;
将所述真实标签图像和所述训练细胞分割结果输入至对数损失函数，获取损失函数值；Inputting the real label image and the training cell segmentation result into a logarithmic loss function to obtain a loss function value;
基于所述损失函数值，通过Adam优化算法对所述增强U-Net网络进行参数调节。Based on the loss function value, the parameters of the enhanced U-Net network are adjusted through the Adam optimization algorithm.
基于上述任一实施例，该装置还包括测试单元；测试单元用于：Based on any of the above embodiments, the device also includes a test unit; the test unit is used for:
将测试细胞图像输入至所述细胞分割模型中，获取所述细胞分割模型输出的测试细胞分割结果；Input the test cell image into the cell segmentation model, and obtain the test cell segmentation result output by the cell segmentation model;
若所述测试细胞图像对应的基准细胞分割结果与所述测试细胞分割结果不同，则基于所述测试细胞图像和所述基准细胞分割结果训练所述细胞分割模型。If the reference cell segmentation result corresponding to the test cell image is different from the test cell segmentation result, the cell segmentation model is trained based on the test cell image and the reference cell segmentation result.
图6为本发明实施例提供的电子设备的实体结构示意图，如图6所示，该电子设备可以包括：处理器(processor)601、通信接口(Communications Interface)602、存储器(memory)603和通信总线604，其中，处理器601，通信接口602，存储器603通过通信总线604完成相互间的通信。处理器601可以调用存储在存储器603上并可在处理器601上运行的计算机程序，以执行上述各实施例提供的细胞图像分割方法，例如包括：获取待分割细胞图像；将所述待分割细胞图像输入至细胞分割模型中，获取所述细胞分割模型输出的细胞分割结果；其中，所述细胞分割模型是基于样本细胞图像和所述样本细胞图像对应的真实标签图像，对增强U-Net网络进行训练得到的；所述增强U-Net网络是在初始U-Net网络中增加网络层数，并在网络中加入BN层的神经网络。FIG. 6 is a schematic diagram of the physical structure of an electronic device provided by an embodiment of the present invention. As shown in FIG. The bus 604 , wherein the processor 601 , the communication interface 602 , and the memory 603 communicate with each other through the communication bus 604 . The processor 601 can call the computer program stored in the memory 603 and run on the processor 601 to execute the cell image segmentation method provided by the above-mentioned embodiments, for example, including: acquiring the image of the cell to be segmented; The image is input into the cell segmentation model, and the cell segmentation result output by the cell segmentation model is obtained; wherein, the cell segmentation model is based on the sample cell image and the real label image corresponding to the sample cell image, and can enhance the U-Net network Obtained by training; the enhanced U-Net network is to increase the number of network layers in the initial U-Net network, and add the neural network of the BN layer in the network.
此外，上述的存储器603中的逻辑指令可以通过软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本发明实施例的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本发明各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。In addition, the above logic instructions in the memory 603 may be implemented in the form of software functional units and when sold or used as an independent product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program codes. .
本发明实施例还提供一种非暂态计算机可读存储介质，其上存储有计算机程序，该计算机程序被处理器执行时实现以执行上述各实施例提供的细胞图像分割方法，例如包括：获取待分割细胞图像；将所述待分割细胞图像输入至细胞分割模型中，获取所述细胞分割模型输出的细胞分割结果；其中，所述细胞分割模型是基于样本细胞图像和所述样本细胞图像对应的真实标签图像，对增强U-Net网络进行训练得到的；所述增强U-Net网络是在初始U-Net网络中增加网络层数，并在网络中加入BN层的神经网络。An embodiment of the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it is implemented to perform the cell image segmentation method provided by the above-mentioned embodiments, for example, including: obtaining The image of the cell to be segmented; input the image of the cell to be segmented into the cell segmentation model, and obtain the cell segmentation result output by the cell segmentation model; wherein, the cell segmentation model is based on the correspondence between the sample cell image and the sample cell image The real label image of is obtained by training the enhanced U-Net network; the enhanced U-Net network is to increase the number of network layers in the initial U-Net network, and add the neural network of the BN layer in the network.
以上所描述的系统实施例仅仅是示意性的，其中所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。本领域普通技术人员在不付出创造性的劳动的情况下，即可以理解并实施。The system embodiments described above are only illustrative, and the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in One place, or it can be distributed to multiple network elements. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. It can be understood and implemented by those skilled in the art without any creative efforts.
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到各实施方式可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件。基于这样的理解，上述技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品可以存储在计算机可读存储介质中，如ROM/RAM、磁碟、光盘等，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行各个实施例或者实施例的某些部分所述的方法。Through the above description of the implementations, those skilled in the art can clearly understand that each implementation can be implemented by means of software plus a necessary general hardware platform, and of course also by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the prior art can be embodied in the form of software products, and the computer software products can be stored in computer-readable storage media, such as ROM/RAM, magnetic discs, optical discs, etc., including several instructions to make a computer device (which may be a personal computer, server, or network device, etc.) execute the methods described in various embodiments or some parts of the embodiments.
最后应说明的是：以上实施例仅用以说明本发明的技术方案，而非对其限制；尽管参照前述实施例对本发明进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本发明各实施例技术方案的精神和范围。Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still be Modifications are made to the technical solutions described in the foregoing embodiments, or equivalent replacements are made to some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910504393.2ACN110288605A (en) | 2019-06-12 | 2019-06-12 | Cell Image Segmentation Method and Device |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910504393.2ACN110288605A (en) | 2019-06-12 | 2019-06-12 | Cell Image Segmentation Method and Device |
| Publication Number | Publication Date |
|---|---|
| CN110288605Atrue CN110288605A (en) | 2019-09-27 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910504393.2APendingCN110288605A (en) | 2019-06-12 | 2019-06-12 | Cell Image Segmentation Method and Device |
| Country | Link |
|---|---|
| CN (1) | CN110288605A (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112102323A (en)* | 2020-09-17 | 2020-12-18 | 陕西师范大学 | Adherent nucleus segmentation method based on generation of countermeasure network and Caps-Unet network |
| CN112446427A (en)* | 2020-11-27 | 2021-03-05 | 王伟佳 | Method and device for identifying myeloid blood cells, storage medium and electronic equipment |
| CN112446892A (en)* | 2020-11-18 | 2021-03-05 | 黑龙江机智通智能科技有限公司 | Cell nucleus segmentation method based on attention learning |
| CN112561897A (en)* | 2020-12-22 | 2021-03-26 | 电子科技大学中山学院 | Photonic crystal fiber end face structure extraction method based on U-Net |
| CN113128455A (en)* | 2021-04-30 | 2021-07-16 | 上海睿钰生物科技有限公司 | Cell image reconstruction model training method and system |
| CN113139973A (en)* | 2021-04-01 | 2021-07-20 | 武汉市疾病预防控制中心 | Artificial intelligence-based plasmodium identification method and equipment |
| CN113689395A (en)* | 2021-08-20 | 2021-11-23 | 深圳先进技术研究院 | An automated stem cell detection method, system, terminal and storage medium |
| CN114612738A (en)* | 2022-02-16 | 2022-06-10 | 中国科学院生物物理研究所 | Training method of cell electron microscope image segmentation model and organelle interaction analysis method |
| US11410303B2 (en) | 2019-04-11 | 2022-08-09 | Agilent Technologies Inc. | Deep learning based instance segmentation via multiple regression layers |
| US11803963B2 (en) | 2019-02-01 | 2023-10-31 | Sartorius Bioanalytical Instruments, Inc. | Computational model for analyzing images of a biological specimen |
| CN119339371A (en)* | 2024-12-18 | 2025-01-21 | 浙江大学 | Blastocyst image instance segmentation method, device, computer equipment and storage medium |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180108139A1 (en)* | 2016-10-19 | 2018-04-19 | U.S. Department Of Veterans Affairs | System And Method For N-Dimensional Image Segmentation Using Convolutional Neural Networks |
| CN109064443A (en)* | 2018-06-22 | 2018-12-21 | 哈尔滨工业大学 | A multi-model organ segmentation method and system based on abdominal ultrasound images |
| CN109191472A (en)* | 2018-08-28 | 2019-01-11 | 杭州电子科技大学 | Based on the thymocyte image partition method for improving U-Net network |
| CN109685813A (en)* | 2018-12-27 | 2019-04-26 | 江西理工大学 | A kind of U-shaped Segmentation Method of Retinal Blood Vessels of adaptive scale information |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180108139A1 (en)* | 2016-10-19 | 2018-04-19 | U.S. Department Of Veterans Affairs | System And Method For N-Dimensional Image Segmentation Using Convolutional Neural Networks |
| CN109064443A (en)* | 2018-06-22 | 2018-12-21 | 哈尔滨工业大学 | A multi-model organ segmentation method and system based on abdominal ultrasound images |
| CN109191472A (en)* | 2018-08-28 | 2019-01-11 | 杭州电子科技大学 | Based on the thymocyte image partition method for improving U-Net network |
| CN109685813A (en)* | 2018-12-27 | 2019-04-26 | 江西理工大学 | A kind of U-shaped Segmentation Method of Retinal Blood Vessels of adaptive scale information |
| Title |
|---|
| 贝琛圆 等: "基于改进U-Net网络的腺体细胞图像分割算法", 《电子科技》* |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11803963B2 (en) | 2019-02-01 | 2023-10-31 | Sartorius Bioanalytical Instruments, Inc. | Computational model for analyzing images of a biological specimen |
| US11748881B2 (en) | 2019-04-11 | 2023-09-05 | Agilent Technologies, Inc. | Deep learning based instance segmentation via multiple regression layers |
| US11410303B2 (en) | 2019-04-11 | 2022-08-09 | Agilent Technologies Inc. | Deep learning based instance segmentation via multiple regression layers |
| CN112102323A (en)* | 2020-09-17 | 2020-12-18 | 陕西师范大学 | Adherent nucleus segmentation method based on generation of countermeasure network and Caps-Unet network |
| CN112102323B (en)* | 2020-09-17 | 2023-07-07 | 陕西师范大学 | Adhesion cell nucleus segmentation method based on generation of countermeasure network and Caps-Unet network |
| CN112446892A (en)* | 2020-11-18 | 2021-03-05 | 黑龙江机智通智能科技有限公司 | Cell nucleus segmentation method based on attention learning |
| CN112446427B (en)* | 2020-11-27 | 2021-07-27 | 王伟佳 | Method and device for identifying myeloid blood cells, storage medium and electronic equipment |
| CN112446427A (en)* | 2020-11-27 | 2021-03-05 | 王伟佳 | Method and device for identifying myeloid blood cells, storage medium and electronic equipment |
| CN112561897A (en)* | 2020-12-22 | 2021-03-26 | 电子科技大学中山学院 | Photonic crystal fiber end face structure extraction method based on U-Net |
| CN113139973A (en)* | 2021-04-01 | 2021-07-20 | 武汉市疾病预防控制中心 | Artificial intelligence-based plasmodium identification method and equipment |
| CN113128455A (en)* | 2021-04-30 | 2021-07-16 | 上海睿钰生物科技有限公司 | Cell image reconstruction model training method and system |
| CN113689395A (en)* | 2021-08-20 | 2021-11-23 | 深圳先进技术研究院 | An automated stem cell detection method, system, terminal and storage medium |
| CN113689395B (en)* | 2021-08-20 | 2025-05-16 | 深圳先进技术研究院 | An automated stem cell detection method, system, terminal and storage medium |
| CN114612738A (en)* | 2022-02-16 | 2022-06-10 | 中国科学院生物物理研究所 | Training method of cell electron microscope image segmentation model and organelle interaction analysis method |
| CN119339371A (en)* | 2024-12-18 | 2025-01-21 | 浙江大学 | Blastocyst image instance segmentation method, device, computer equipment and storage medium |
| Publication | Publication Date | Title |
|---|---|---|
| CN110288605A (en) | Cell Image Segmentation Method and Device | |
| CN107506761B (en) | Brain image segmentation method and system based on saliency learning convolutional neural network | |
| CN107633522B (en) | Brain Image Segmentation Method and System Based on Local Similarity Active Contour Model | |
| CN107480707B (en) | A Deep Neural Network Method Based on Information Lossless Pooling | |
| CN112052754B (en) | Polarization SAR image ground object classification method based on self-supervision characterization learning | |
| CN107506822B (en) | Deep neural network method based on space fusion pooling | |
| CN111652892A (en) | Remote sensing image building vector extraction and optimization method based on deep learning | |
| CN109492674B (en) | Generation method and device of SSD (solid State disk) framework for target detection | |
| CN110992351A (en) | sMRI image classification method and device based on multi-input convolutional neural network | |
| CN105095902B (en) | Picture feature extracting method and device | |
| CN110443222A (en) | Method and apparatus for training face's critical point detection model | |
| CN107169504A (en) | A kind of hand-written character recognition method based on extension Non-linear Kernel residual error network | |
| CN112132827A (en) | Pathological image processing method and device, electronic equipment and readable storage medium | |
| CN113095333B (en) | Unsupervised feature point detection method and unsupervised feature point detection device | |
| CN106971198A (en) | A kind of pneumoconiosis grade decision method and system based on deep learning | |
| CN103985112B (en) | Image segmentation method based on improved multi-objective particle swarm optimization and clustering | |
| CN110866938B (en) | A fully automatic video moving object segmentation method | |
| CN114972753A (en) | A lightweight semantic segmentation method and system based on contextual information aggregation and assisted learning | |
| CN116630971B (en) | Spore segmentation method of wheat scab based on CRF_ResUnet++ network | |
| CN112101364A (en) | A Semantic Segmentation Method Based on Incremental Learning of Parameter Importance | |
| CN107609589A (en) | A kind of feature learning method of complex behavior sequence data | |
| CN113139556B (en) | Manifold multi-view image clustering method and system based on adaptive composition | |
| Patil et al. | Segmentation of cotton leaf images using a modified chan vese method | |
| CN114743042B (en) | Longjing tea quality identification method based on depth characteristics and TrAdaBoost | |
| CN110084247A (en) | A kind of multiple dimensioned conspicuousness detection method and device based on fuzzy characteristics |
| 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 | ||
| RJ01 | Rejection of invention patent application after publication | Application publication date:20190927 | |
| RJ01 | Rejection of invention patent application after publication |