WO2025161970A1 - 一种基于深度学习pid优化器的图像分类方法 - Google Patents

一种基于深度学习pid优化器的图像分类方法

Info

Publication number
WO2025161970A1
WO2025161970A1 PCT/CN2025/072620 CN2025072620W WO2025161970A1 WO 2025161970 A1 WO2025161970 A1 WO 2025161970A1 CN 2025072620 W CN2025072620 W CN 2025072620W WO 2025161970 A1 WO2025161970 A1 WO 2025161970A1
Authority
WO
WIPO (PCT)
Prior art keywords
image classification
optimizer
layer
pid
classification 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.)
Pending
Application number
PCT/CN2025/072620
Other languages
English (en)
French (fr)
Inventor
李新德
余乐
张鹏飞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southeast University
Original Assignee
Southeast University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Southeast University filed Critical Southeast University
Publication of WO2025161970A1 publication Critical patent/WO2025161970A1/zh
Anticipated expiration legal-status Critical
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Definitions

  • the present invention belongs to the technical field of computer vision and pattern recognition, and relates to an image classification method based on a deep learning PID optimizer.
  • Deep learning-based image processing technology has achieved significant breakthroughs in machine vision fields (such as image classification), overcoming the challenges of low accuracy and efficiency associated with traditional feature engineering techniques.
  • the powerful performance of deep learning is driven by large datasets, complex neural network structures, and the enormous computational effort and training time required.
  • the use of fast and robust optimizers is a core technology in deep learning models. These optimizers can rapidly update model weights and learn effective empirical knowledge, while also enabling the model to converge to a near global optimum.
  • PID control theory has a well-established application in the field of control science, enabling rapid adjustments to minimize deviations.
  • Deep neural networks are widely used as feature extractors, mapping image data into an embedding space to construct specific feature vectors, which are then used in downstream tasks.
  • a parameter update criterion based on a PID optimizer can quickly reduce deviations in algorithmic predictions.
  • the proportional term P is considered the error between the model's predicted score for a category and the actual label, and is used to adjust the model's importance to each category.
  • the integral term I is used to address long-term error accumulation, helping the model better learn the characteristics of scarce categories.
  • the differential term D balances P and I, preventing the model from overfitting or over-adjusting certain samples.
  • the present invention provides an image classification method based on deep learning PID optimizer.
  • An image classification method based on a deep learning PID optimizer of the present invention comprises the following steps:
  • Step S1 Divide the dataset into training dataset, validation dataset and test dataset
  • Step S2 Construct an image classification model based on deep learning, wherein the image classification model includes an N_layer layer subnetwork, and initialize the weight and maximum number of iterations of the N_layer layer subnetwork;
  • Step S3 using the training data set to train the image classification model and calculate the loss between the model prediction value and the true label;
  • Step S4 Generate K P , KI and K D coefficients of each layer of sub-network based on the generation algorithm
  • Step S5 Based on the K P , KI and K D coefficients generated in step S4 , the inner-loop learnable parameters of the image classification model are updated using a PID optimizer, and the outer-loop learnable parameters of the image classification model are updated using an Adam optimizer.
  • Step S6 testing the generalization performance of the image classification model on the validation dataset, and using the PID optimizer to adjust the weight parameters of the N_layer subnetwork as the initial weight values of the image classification model;
  • Step S7 Repeat steps S3-S6 until the number of iterations reaches the maximum number of iterations set in step S2, and obtain the optimized image classification model.
  • step one also includes image preprocessing: normalizing the mean and variance of each image in the dataset, and then converting the image data into a tensor format.
  • the image classification model in step S2 is composed of a 4-layer convolutional neural network CNN or a ResNet12 structure.
  • the number of convolution kernels of the convolutional neural network CNN is 64, and the convolution kernel size is 3.
  • step S3 the image classification model is trained using the training dataset: a fixed number of batches of data are read in each round of training, and each batch of data includes N_way ⁇ (K support +K query ) sample data, where N_way is the number of categories, K support is the number of support sets for each category, and K query is the number of query sets for each category.
  • step S3 the loss value in step S3 is calculated by the cross entropy loss function.
  • the generated algorithm in step S4 consists of two Linear layers, whose input dimension is N_layer ⁇ 3 and output dimension is N_layer ⁇ 3, where N_layer is the number of layers of the image classification model.
  • step S4 the following formula is used to generate the K P , KI and K D coefficients of each layer of the sub-network:
  • N_layer is the number of layers of the image classification model, and are the mean, variance, and gradient values of each layer parameter of the image classification model.
  • step S5 the PID optimizer is used to update the inner loop learnable parameters of the image classification model using the following formula:
  • K P , K I , K D are the adaptive coefficients generated by the algorithm in step S5
  • ⁇ is the weight decay factor
  • L t and L t-1 are the loss values of the image classification model at the t-th and t-1-th iterations respectively
  • ⁇ tm and ⁇ t -m+1 are the weight attenuation factors at the tm-th and t-m+1-th moments, respectively.
  • the image classification method based on deep learning PID optimizer of the present invention adopts the image classification method based on PID optimizer to solve the problems of slow efficiency and low precision of existing image classification tasks, and adopts a generation algorithm to adaptively generate the K P , KI and K D coefficients of each layer of sub-network of the model to solve the problem of manual adjustment parameters in the PID control algorithm. It has stronger generalization and robustness, and can achieve the effect of quickly adapting to new tasks.
  • FIG1 is a schematic flow chart of an image classification method based on a deep learning PID optimizer according to the present invention
  • Figure 2 is a schematic diagram of the parameter update criteria of the inner loop PID optimizer.
  • the image classification method based on deep learning PID optimizer of the present invention includes the following steps:
  • Step S1 Divide the entire dataset into training dataset, validation dataset, and test dataset according to a certain ratio, and there are no repeated categories among the three.
  • the dataset contains 100 categories, each category contains 600 images, and 70 categories are selected as training datasets, 20 categories as validation datasets, and 10 categories as test datasets; the mean and variance of each image are normalized and preprocessed, and then the image data is converted into a tensor format. A fixed number of batch data is read in each round of training.
  • Each batch data contains N_way ⁇ (K support +K query ) sample data, where N_way is the number of categories, usually set to 5 or 20, K support is the number of support sets for each category, usually set to 5 or 1, and K query is the number of query sets for each category, usually set to 15;
  • Step S2 Construct a small sample image classification learning model and initialize the weights:
  • the constructed model consists of a 4-layer CNN or ResNet12 structure, where the number of convolution kernels in each layer of the CNN model is set to 64 and the convolution kernel size is set to 3; in particular, each sub-network of the constructed model can use its own weights to forward calculate the feature vector of the image, or use specified weights to calculate the feature vector of the image; for the convolution layer of the model, use Gaussian or Xavier initialization to initialize the weights, and for the bias layer of the model, use zero initialization to initialize the weights;
  • Step S3 Pack the images of the training dataset into batch data as described in S1, set the number of iterations, input the packed batch data into the model built in S2 for forward calculation, and map the image features into a feature vector of fixed length;
  • Step S4 Train the model based on the training dataset and calculate the loss between the model-predicted labels and the true label values corresponding to the images.
  • the loss function can be calculated using cross entropy loss.
  • Step S5 Adaptively generate K P , K I , and K D coefficients for each sub-network layer of the model based on a generative algorithm.
  • the generative algorithm consists of two learnable linear layers with an input dimension of N_layer ⁇ 3 and an output dimension of N_layer ⁇ 3, where N_layer is the number of layers in the model constructed in step 2.
  • the input data is the mean and variance of the weights of each layer of the model and the gradient value corresponding to the weights of each layer of the model, that is:
  • N_layer is the number of layers of the model constructed in step 2, and the output is the K P , K I , K D coefficients of each sub-network of the model.
  • Step S6 Based on the loss value calculated in S4, the gradient value of each sub-network is obtained through gradient back propagation operation, and the PID optimizer is used to update the inner loop learnable parameters of the model, and the Adam optimizer is used to update the outer loop learnable parameters of the model.
  • the calculation formula of the PID optimizer parameter update model is:
  • K P , K I , K D are the adaptive coefficients generated by the algorithm in step S5, is the gradient value of the parameter weight of the i-th layer sub-network of the model constructed in step S2 at the t-th iteration, and ⁇ is the weight decay factor;
  • Step S7 In the above S6, after each iteration of the training dataset, the generalization performance of the model is tested on the validation dataset, and the PID optimizer is used to fine-tune the model weight parameters as the initial weight values of the model;
  • Step S8 Use the model weights learned in the above S7 as the initial weight values of the test task, and after a few iterative calculations, test the recognition accuracy of the model on the test dataset.
  • This paper uses a PID optimizer to update the learnable parameters of the model's inner loop, addressing the low learning efficiency problem associated with unequal distributions of training and test tasks in image classification learning algorithms. It also employs a generative algorithm to adaptively generate the K P , KI , and K D coefficients for each subnetwork layer of the model, resolving the manual parameter adjustment challenges of traditional PID controllers and enhancing the model's adaptability. Furthermore, the introduction of a PID control theory algorithm avoids the weak interpretability of traditional image classification learning methods. Compared to previous methods, this method offers greater adaptability and higher recognition accuracy, enabling rapid adaptation to new tasks.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Multimedia (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Molecular Biology (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)

Abstract

本发明公开一种基于深度学习PID优化器的图像分类方法，针对模型内的不同层子网络使用生成算法自适应生成各自的KP,KI和KD系数；在内环使用PID优化器更新模型内环可学习参数，在外环使用Adam优化器更新模型外环可学习参数。本发明的基于深度学习PID优化器的图像分类方法，采用了基于PID优化器的图像分类方法解决了现有图像分类任务效率慢精度低的问题，采用了生成算法自适应生成模型各层子网络的KP,KI和KD系数，解决PID控制算法中的人工调节参数的问题，具有更强的自适应能力和更高的识别精度，能快速适应新的任务。

Description

一种基于深度学习PID优化器的图像分类方法 技术领域
本发明属于计算机视觉和模式识别技术领域，涉及一种基于深度学习PID优化器的图像分类方法。
背景技术
基于深度学习的图像处理技术，在机器视觉领域(如图像分类)有很大的技术突破，解决了传统特征工程技术的精度低效率慢的难题。大规模的数据集、复杂的神经网络结构、庞大的计算量和训练时间是深度学习性能强大的关键原因。为了减少模型的训练时长，加速模型的学习效率，使用快速鲁棒的优化器是深度学习模型的核心技术，一方面优化器可以快速更新模型的权重，学习有效的经验知识；另一方面可以使模型收敛到近似的全局最优点。
现有深度学习中的优化器算法，当训练任务和测试任务分布不一致时其学习效率低下，需要经过几十轮的迭代学习才能收敛，识别精度也大大降低；其次缺乏理论知识研究，适配性较弱，只针对特定任务所涉及的领域有所提升。
PID控制理论在控制学科上有很成熟的应用，针对偏差能快速调节使偏差减小。深度神经网络被广泛用于特征提取器，将图像数据映射到嵌入空间构成特定的特征向量，最后用于下游任务中。针对算法预测存在的偏差值，使用基于PID优化器的参数更新准则能快速减少偏差，将比例项P看作是模型对于某个类别的预测分数与实际标签之间的误差，用于调整模型对每个类别的重要性；使用积分项I处理长期的误差累积，帮助模型更好地学习稀缺类别的特征；微分项D可以平衡P和I，防止模型对于某些样本的过拟合或者过度调整。通过对未来的预测以及历史的积累来调节模型的权重，使算法具备快速适应新任务的能力，面对全新的任务也有很强的识别性能，大幅提升了图像分类的精度。
因此，需要一种基于深度学习PID优化器的图像分类方法以解决上述问题。
发明内容
为了解决上述的缺少具体、完善、成熟的实际实验装置的问题，本发明提供一种基于深度学习PID优化器的图像分类方法。
本发明的一种基于深度学习PID优化器的图像分类方法，包括以下步骤：
步骤S1：将数据集划分为训练数据集、验证数据集和测试数据集；
步骤S2：构建基于深度学习的图像分类模型，所述图像分类模型包括N_layer层子网络，初始化N_layer层子网络的权重和最大迭代次数；
步骤S3：利用训练数据集训练所述图像分类模型并计算模型预测值与真实标签的损失值；
步骤S4：基于生成算法生成每层子网络的KP,KI和KD系数；
步骤S5：基于步骤S4生成的KP,KI和KD系数使用PID优化器来更新图像分类模型的内环可学习参数，使用Adam优化器来更新图像分类模型的外环可学习参数；
步骤S6：在验证数据集上测试所述图像分类模型的泛化性能，并使用PID优化器调整N_layer层子网络的权重参数作为图像分类模型的权重初始值；
步骤S7：重复步骤S3-S6，直到迭代次数达到步骤S2设置的最大迭代次数，得到优化后的图像分类模型。
更进一步的，步骤一中还包括图像预处理：对数据集中每张图像的均值和方差做归一化预处理，然后将图像数据转换成张量格式。
更进一步的，步骤S2中所述图像分类模型由4层卷积神经网络CNN组成或为ResNet12结构。
更进一步的，所述卷积神经网络CNN的卷积核个数为64，卷积核大小为3。
更进一步的，步骤S3中利用训练数据集训练所述图像分类模型：每轮训练时读取固定数量的batch数据，每个batch数据包括N_way×(Ksupport+Kquery)个样本数据，其中，N_way是类别数，Ksupport是每个类别的支持集数量，Kquery是每个类别的查询集数量。
更进一步的，步骤S3中损失值通过交叉熵损失函数计算得到。
更进一步的，步骤S4中生成算法由两个Linear层组成，其输入维度是N_layer×3，输出维度是N_layer×3，其中，N_layer是图像分类模型的层数。
更进一步的，步骤S4中利用下式生成每层子网络的KP,KI和KD系数：
其中，f是生成算法， N_layer是图像分类模型的层数，和分别是图像分类模型每层参数的均值、方差和梯度值。
更进一步的，步骤S5中利用下式使用PID优化器来更新图像分类模型的内环可学习参数：
式中，KP,KI,KD是步骤S5中生成算法生成的自适应系数，是图像分类模型第i层子网络的参数权重在第t轮迭代时的梯度值，α是权重衰减因子，Lt和Lt-1分别是图像分类模型在第t和t-1轮迭代时的损失值，和分别是图像分类模型的第i层子网络在第t和t-1轮迭代时的参数值，αt-m和αt-m+1分别是第t-m和t-m+1时刻的权重衰减因子。
有益效果：本发明的基于深度学习PID优化器的图像分类方法，采用了基于PID优化器的图像分类方法解决了现有图像分类任务效率慢精度低的问题，采用了生成算法自适应生成模型各层子网络的KP,KI和KD系数，解决PID控制算法中的人工调节参数的问题，具有更强的泛化性和鲁棒性，可以达到快速适应新任务的效果。
附图说明
图1为本发明的基于深度学习PID优化器的图像分类方法的流程示意图；
图2为内环PID优化器参数更新准则示意图。
具体实施方式
现在结合附图对本发明作进一步详细的说明。这些附图均为简化的示意图，仅以示意方式说明本发明的基本结构，因此其仅显示与本发明有关的构成。
如图1-2所示，本发明的基于深度学习PID优化器的图像分类方法，包括以下步骤：
步骤S1：将整个数据集按照一定的比例划分为训练数据集、验证数据集和测试数据集，且三者没有重复的类别。以mini-ImageNet数据集为例，该数据集包含100个类别，每个类别包含600张图像，分别选择70类作为训练数据集，20类作为验证数据集，10类作为测试数据集；对每张图像的均值和方差做归一化预处理，然后将图像数据转换成张量格式，每轮训练时读取固定数量的batch数据，每个batch数据包含N_way×(Ksupport+Kquery)个样本数据，其中N_way是类别数，通常设为5或者20，Ksupport是每个类别的支持集数量，通常设为5或者1，Kquery是每个类别的查询集数量，通常设为15；
步骤S2：构建小样本图像分类学习模型并初始化权重：所搭建的模型由4层CNN组成或者是ResNet12结构，其中每层CNN模型的卷积核个数设为64，卷积核大小设为3；特别的，所搭建模型的各个子网络，可使用自身的权重来前向计算得出图像的特征向量，也可使用指定的权重计算得出图像的特征向量；对于模型的卷积层使用高斯或者Xavier初始化方式初始化权重，对于模型的偏置层使用零初始化方式初始化权重；
步骤S3：将训练数据集的图像按照S1中所述打包成batch数据，设置迭代轮数，将打包好的batch数据输入给S2中构建的模型做前向计算，将图像特征映射为固定长度的特征向量；
步骤S4：基于训练数据集训练模型，并计算模型预测的标签和图像对应真实标签值的损失值，其中损失函数可用交叉熵损失计算；
步骤S5：基于生成算法为给模型的每层子网络自适应生成各自的KP,KI,KD系数，其中生成算法是由两层可学习的Linear层组成，其输入维度是N_layer×3，输出维度是N_layer×3，其中N_layer是步骤2构建模型的层数，具体的，输入的数据是模型每层权重的均值和方差以及模型每层权重对应的梯度值，即:
式中，f是生成算法，
N_layer是步骤2所构建模型的层数，输出是模型每层子网络的KP,KI,KD系数。
步骤S6：根据S4中计算得到的损失值，通过梯度反向传播运算得出每层子网络的梯度值，使用PID优化器来更新模型的内环可学习参数，使用Adam优化器来更新模型的外环可学习参数，其中，PID优化器参数更新模型计算公式为：
式中，KP,KI,KD是步骤S5中生成算法生成的自适应系数，是步骤S2中构建的模型第i层子网络的参数权重在第t轮迭代时的梯度值，α是权重衰减因子；
步骤S7：上述S6中，每迭代完一轮训练数据集后，就在验证数据集上测试模型的泛化性能，并使用PID优化器微调模型权重参数作为模型的权重初始值；
步骤S8：将上述S7中学习到的模型权重作为测试任务的初始权重值，经过少量的几步迭代计算，在测试数据集上测试模型的识别精度。
本发明采用PID优化器更新模型内环的可学习参数，解决了图像分类学习算法中训练和测试任务分布不同时学习效率低的问题；采用生成算法自适应生成模型每层子网络的KP,KI,KD系数，解决了传统PID控制器的人工手动调节参数的难题，使模型具备更强的自适应能力；引入PID控制理论算法，避免了传统图像分类学习方法可解释性弱的弊端。相较于以往方法具有更强的自适应能力和更高的识别精度，能快速适应新的任务。
以上述依据本发明的理想实施例为启示，通过上述的说明内容，相关工作人员完全可以在不偏离本项发明技术思想的范围内，进行多样的变更以及修改。本项发明的技术性范围并不局限于说明书上的内容，必须要根据权利要求范围来确定其技术性范围。

Claims (9)

  1. 一种基于深度学习PID优化器的图像分类方法，其特征在于，包括以下步骤：
    步骤S1：将数据集划分为训练数据集、验证数据集和测试数据集；
    步骤S2：构建基于深度学习的图像分类模型，所述图像分类模型包括N_layer层子网络，初始化N_layer层子网络的权重和最大迭代次数；
    步骤S3：利用训练数据集训练所述图像分类模型并计算模型预测值与真实标签的损失值；
    步骤S4：基于生成算法生成每层子网络的KP,KI和KD系数；
    步骤S5：基于步骤S4生成的KP,KI和KD系数使用PID优化器来更新图像分类模型的内环可学习参数，使用Adam优化器来更新图像分类模型的外环可学习参数；
    步骤S6：在验证数据集上测试所述图像分类模型的泛化性能，并使用PID优化器调整N_layer层子网络的权重参数作为图像分类模型的权重初始值；
    步骤S7：重复步骤S3-S6，直到迭代次数达到步骤S2设置的最大迭代次数，得到优化后的图像分类模型。
  2. 根据权利要求1所述的基于深度学习PID优化器的图像分类方法，其特征在于，步骤一中还包括图像预处理：对数据集中每张图像的均值和方差做归一化预处理，然后将图像数据转换成张量格式。
  3. 根据权利要求1所述的基于深度学习PID优化器的图像分类方法，其特征在于，步骤S2中所述图像分类模型由4层卷积神经网络CNN组成或为ResNet12结构。
  4. 根据权利要求3所述的基于深度学习PID优化器的图像分类方法，其特征在于，所述卷积神经网络CNN的卷积核个数为64，卷积核大小为3。
  5. 根据权利要求1所述的基于深度学习PID优化器的图像分类方法，其特征在于，步骤S3中利用训练数据集训练所述图像分类模型：每轮训练时读取固定数量的batch数据，每个batch数据包括N_way×(Ksupport+Kquery)个样本数据，其中，N_way是类别数，Ksupport是每个类别的支持集数量，Kquery是每个类别的查询集数量。
  6. 根据权利要求1所述的基于深度学习PID优化器的图像分类方法，其特征在于，步骤S3中损失值通过交叉熵损失函数计算得到。
  7. 根据权利要求1所述的基于深度学习PID优化器的图像分类方法，其特征在于，步骤S4中生成算法由两个Linear层组成，其输入维度是N_layer×3，输出维度是N_layer×3，其中，N_layer是图像分类模型的层数。
  8. 根据权利要求1所述的基于深度学习PID优化器的图像分类方法，其特征在于，步骤S4中利用下式生成每层子网络的KP,KI和KD系数：
    其中，f是生成算法， N_layer是图像分类模型的层数，和分别是图像分类模型每层参数的均值、方差和梯度值。
  9. 根据权利要求1所述的基于深度学习PID优化器的图像分类方法，其特征在于，步骤S5中利用下式使用PID优化器来更新图像分类模型的内环可学习参数：
    式中，KP,KI,KD是步骤S5中生成算法生成的自适应系数，是图像分类模型第i层子网络的参数权重在第t轮迭代时的梯度值，α是权重衰减因子，Lt和Lt-1分别是图像分类模型在第t和t-1轮迭代时的损失值，和分别是图像分类模型的第i层子网络在第t和t-1轮迭代时的参数值，αt-m和αt-m+1分别是第t-m和t-m+1时刻的权重衰减因子。
PCT/CN2025/072620 2024-01-31 2025-01-16 一种基于深度学习pid优化器的图像分类方法 Pending WO2025161970A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202410135709.6 2024-01-31
CN202410135709.6A CN117975135B (zh) 2024-01-31 2024-01-31 一种基于深度学习pid优化器的图像分类方法

Publications (1)

Publication Number Publication Date
WO2025161970A1 true WO2025161970A1 (zh) 2025-08-07

Family

ID=90845497

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2025/072620 Pending WO2025161970A1 (zh) 2024-01-31 2025-01-16 一种基于深度学习pid优化器的图像分类方法

Country Status (2)

Country Link
CN (1) CN117975135B (zh)
WO (1) WO2025161970A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN120871580A (zh) * 2025-09-24 2025-10-31 国科大杭州高等研究院 基于深度学习的激光锁频系统多级pid参数调控方法
CN121190727A (zh) * 2025-09-24 2025-12-23 杭州第二人生科技有限公司 一种基于图神经网络的移动式多视角图片数据采集方法

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117975135B (zh) * 2024-01-31 2025-07-25 东南大学 一种基于深度学习pid优化器的图像分类方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114047691A (zh) * 2022-01-10 2022-02-15 杭州海康威视数字技术股份有限公司 自适应学习率协同优化的目标识别方法、装置及电子设备
CN115205577A (zh) * 2022-05-13 2022-10-18 北京理工大学 用于图像分类的卷积神经网络的自适应优化训练方法
US20220391706A1 (en) * 2021-06-02 2022-12-08 Google Llc Training neural networks using learned optimizers
CN117975135A (zh) * 2024-01-31 2024-05-03 东南大学 一种基于深度学习pid优化器的图像分类方法

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US12271810B2 (en) * 2018-12-04 2025-04-08 Google Llc Federated learning with adaptive optimization
CN115170879B (zh) * 2022-07-12 2025-04-18 北京大学 基于深度学习优化器的大规模图像分类方法
CN116543229A (zh) * 2023-05-24 2023-08-04 北京理工大学 基于自适应累积系数的深度学习优化器的图像分类方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20220391706A1 (en) * 2021-06-02 2022-12-08 Google Llc Training neural networks using learned optimizers
CN114047691A (zh) * 2022-01-10 2022-02-15 杭州海康威视数字技术股份有限公司 自适应学习率协同优化的目标识别方法、装置及电子设备
CN115205577A (zh) * 2022-05-13 2022-10-18 北京理工大学 用于图像分类的卷积神经网络的自适应优化训练方法
CN117975135A (zh) * 2024-01-31 2024-05-03 东南大学 一种基于深度学习pid优化器的图像分类方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN120871580A (zh) * 2025-09-24 2025-10-31 国科大杭州高等研究院 基于深度学习的激光锁频系统多级pid参数调控方法
CN121190727A (zh) * 2025-09-24 2025-12-23 杭州第二人生科技有限公司 一种基于图神经网络的移动式多视角图片数据采集方法

Also Published As

Publication number Publication date
CN117975135A (zh) 2024-05-03
CN117975135B (zh) 2025-07-25

Similar Documents

Publication Publication Date Title
CN117975135B (zh) 一种基于深度学习pid优化器的图像分类方法
US11049011B2 (en) Neural network classifier
WO2023024577A1 (zh) 面向边缘计算的重参数神经网络架构搜索方法
CN113158582A (zh) 基于复数值前向神经网络的风速预测方法
US20240013048A1 (en) Method and System for Solving QUBO Problems with Hybrid Classical-Quantum Solvers
CN113052261B (zh) 一种基于余弦空间优化的图像分类损失函数的设计方法
CN112766399B (zh) 一种面向图像识别的自适应神经网络训练方法
CN113743474A (zh) 基于协同半监督卷积神经网络的数字图片分类方法与系统
CN111062465A (zh) 一种神经网络结构自调整的图像识别模型及方法
KR102903552B1 (ko) 부스팅 및 행렬 인수분해
US20250284927A1 (en) Method for predicting short-term wind power of newly-built wind power plant based on sample migration
Ni et al. Efficient off-policy reinforcement learning via brain-inspired computing
CN117372807A (zh) 多模态融合的碰撞塑性变形点云序列预测方法
Weng et al. Adapid: An adaptive pid optimizer for training deep neural networks
Amiridi et al. Low-rank characteristic tensor density estimation part II: Compression and latent density estimation
US20250045609A1 (en) Systems and methods for gradient estimation in optimization to improve computational efficiency for quantum systems
WO2025161601A1 (zh) 图像类别的量子识别方法、系统、电子设备及存储介质
Scellier Quantum equilibrium propagation: Gradient-descent training of quantum systems
US20200372363A1 (en) Method of Training Artificial Neural Network Using Sparse Connectivity Learning
CN114299345A (zh) 一种基于跟踪微分器的极限学习机的图像分类方法
CN111914915A (zh) 基于支持向量机的数据分类器集成方法、装置及存储介质
CN115035337B (zh) 一种基于模型融合的交替训练方法
CN118095441A (zh) 一种基于迁移学习的开源大语言模型微调优化方法
CN116912600A (zh) 一种基于变步长admm算法极限学习机的图像分类方法
CN117151998A (zh) 一种基于支持向量回归的图像光照校正方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 25747542

Country of ref document: EP

Kind code of ref document: A1