CN111539306B - A method for building recognition in remote sensing images based on the replaceability of activation expressions - Google Patents
A method for building recognition in remote sensing images based on the replaceability of activation expressions Download PDFInfo
- Publication number
- CN111539306B CN111539306B CN202010314628.4A CN202010314628A CN111539306B CN 111539306 B CN111539306 B CN 111539306B CN 202010314628 A CN202010314628 A CN 202010314628A CN 111539306 B CN111539306 B CN 111539306B
- Authority
- CN
- China
- Prior art keywords
- replaceability
- convolution kernel
- expression
- activation
- remote sensing
- 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
- 230000014509 gene expression Effects 0.000 title claims abstract description 76
- 230000004913 activation Effects 0.000 title claims abstract description 59
- 238000000034 method Methods 0.000 title claims abstract description 31
- 230000004044 response Effects 0.000 claims abstract description 37
- 238000012549 training Methods 0.000 claims abstract description 11
- 238000003062 neural network model Methods 0.000 claims abstract description 10
- 230000008569 process Effects 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 abstract description 3
- 238000012360 testing method Methods 0.000 description 8
- 238000013136 deep learning model Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 3
- 238000013138 pruning Methods 0.000 description 3
- 230000000630 rising effect Effects 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000003252 repetitive effect Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000008094 contradictory effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 238000009440 infrastructure construction Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/176—Urban or other man-made structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
本发明公开了基于激活表达可替换性的遥感图像建筑物识别方法，包括以下步骤：获取遥感图像建筑物数据集；训练普通深度神经网络模型；计算识别所述模型中每个卷积核的独立最大响应图；计算每个卷积核的激活表达可替换性；根据每个卷积核的激活表达可替换性对模型卷积核进行修剪，保留小的激活表达可替换性；使用修剪后的深度神经网络模型进行遥感图像建筑物识别。本发明方法针对基于深度学习的建筑物识别模型提出激活的表达可替换性指标方法，它可以量化同层上每个卷积核在特征空间表达的可替换性，卷积核的激活的表达可替换性值越低，代表它在特征空间越不可替换，进而针对卷积核进行选择性修剪，从而有效提高遥感图像建筑物识别模型的识别精度。
The invention discloses a method for recognizing buildings in remote sensing images based on the replaceability of activation expressions, comprising the following steps: acquiring a data set of buildings in remote sensing images; training a common deep neural network model; Maximum response map; calculate the activation expression replaceability of each convolution kernel; prune the model convolution kernel according to the activation expression replaceability of each convolution kernel, retaining the small activation expression replaceability; use the pruned Deep neural network model for building recognition in remote sensing images. The method of the invention proposes an activated expression replaceability index method for the building recognition model based on deep learning, which can quantify the replaceability of the expression of each convolution kernel on the same layer in the feature space, and the activation expression of the convolution kernel can be The lower the replacement value is, the more irreplaceable it is in the feature space, and the convolution kernel is selectively pruned to effectively improve the recognition accuracy of the remote sensing image building recognition model.
Description
Technical Field
The invention belongs to the technical field of remote sensing image identification, and relates to a remote sensing image building identification method based on activation expression replaceability.
Background
In recent years, a great number of remote sensing satellites are lifted off and simultaneously bring a great number of remote sensing images. Remote-sensing image data is increasing dramatically, including a variety of remote-sensing images with different spectral and spatial resolutions. These remote sensing images bring great economic value. The method can be used for rapidly extracting targets such as buildings in the remote sensing image, and can effectively help urban planning, infrastructure construction, illegal building detection and the like. At present, a large number of building identification algorithms based on deep learning are developed at present, but the whole identification precision of the model is difficult to meet the practical application due to the fact that the generalization of the remote sensing image building identification model is not known.
At present, there are two main methods for measuring the generalization ability of the deep learning model. The first scheme mainly depends on a traditional statistical learning theory method, such as VC dimension, Rademacher complexity and other algorithms to explore the relationship between the robustness, complexity and generalization of the model. These theories suggest that models containing a large number of parameters tend to over-fit on the data, but at the same time reduce the generalization ability on the test data. However, this conclusion is contrary to the performance of the current deep learning model, and the traditional statistical learning theory cannot reasonably explain the generalization ability of the deep learning model. The second scheme mainly explains and evaluates the generalization capability of the model from the change of the parameter space in the deep learning model optimization process. Schmidhuber considers that the generalization ability of the model is related to the straightness of minima and the straightness of Bayesian boundaries. However, Dinh indicates that a non-smooth min-depth learning model may actually have a better generalization. Wang links the smoothness of the solution and the generalization ability of the model under a Bayesian framework, and theoretically proves that the generalization ability of the model is not only related to Hessian spectrum, but also related to the smoothness of the solution, the scale of parameters and the number of training samples. In addition, the random gradient descent algorithm for training the deep learning model can also improve the generalization capability of the model. Many of the conclusions in the second approach are also contradictory.
In practical application, a rich and well-distributed remote sensing image building test set is usually used for evaluating the generalization capability of the model. In this approach, however, Recht suggests that obtaining a well-behaved model on a particular test set may not embody the model's own generalization ability, and that accuracy based on the test set is fragile and subject to changes due to subtle changes in data distribution. The evaluation on the test set also has the problem of inaccuracy. Therefore, at present, no reasonable algorithm can correctly measure the generalization of the remote sensing image recognition model.
Disclosure of Invention
In view of the above, the present invention provides a method for identifying a building based on a remote sensing image with an alternative activation expression, and the present invention provides an index for effectively evaluating the generalization of a remote sensing image identification model, and the remote sensing image identification model is pruned through the index, so as to improve the accuracy of the model and further improve the identification capability of the model.
The invention aims to realize the method for identifying the buildings based on the remote sensing images with the alternative activation expression, which comprises the following steps:
step 1, obtaining a building data set of a remote sensing image;
step 2, training a common deep neural network model;
step 3, calculating and identifying an independent maximum response graph of each convolution kernel in the model;
step 4, calculating the activation expression replaceability of each convolution kernel;
step 5, pruning the model convolution kernels according to the activation expression replaceability of each convolution kernel, and keeping small activation expression replaceability;
and 6, using the trimmed deep neural network model to identify the remote sensing image building.
Specifically, in the training process of the deep neural network model described in step 2, the training set is represented as D ═ { X, y }, X represents the nth remote sensing image, y represents the building label corresponding to the nth remote sensing image, Θ ═ W, b } represents the weight of the deep neural network to be trained, W represents the ith convolutional layer, b represents the offset on the ith layer, and the trained weight Θ is obtained by defining the loss function of the recognition task and using the BP algorithm*So that the model can be made static with small error values while achieving a high recognition accuracy on the data set D.
Further, in the calculation process of the independent maximum response graph in step 3, the objective function is
Wherein, X of the initial input is random noise theta*For the trained weights, J represents the number of all convolution kernels in the l layers, and the output of each convolution kernel is hl，i(X，Θ*) Which represents the output of the ith convolution kernel at the l-th layer, hl，-i(X，Θ*) Other feature maps representing the activation values of the outputs excluding the target i convolution kernels, argmax (×) representing the maximum response map of the outputs, X*Then the independent maximum response graph of the final output is represented;
fixing corresponding weight value, using gradient rising algorithm to iteratively update X*So that it can make the output activation value of ith convolution kernel of l layer be maximum and X obtained by target function*The target convolution kernel output can be enabled to be as large as possible, and simultaneously, the integral output of other convolution kernels is ensured to be small, and the final X is obtained*Then the independent maximum response map in feature space for the corresponding convolution kernel.
Further, the alternative calculation of the activation expression comprises the following steps:
step 401, calculating the expression replaceability, wherein the formula of the expression replaceability is as follows
RS (l, i) represents the property that the unwrapping feature of the ith convolution kernel of the ith layer can be replaced by other convolution kernels on the same layer, wherein | { xlL represents the total number of the ith layer of convolution kernels, IAM (l, i) represents the characteristic representation of an independent maximum response graph generated by the ith layer of convolution kernels, and f (IAM (l, i)) represents the activation value of the ith layer of convolution kernels obtained by forward propagation of the characteristic representation of the generated independent maximum response graph; i { xl，j：xl，j＞xl，iThe I represents the number of convolution kernels which are larger than the activation value of the ith convolution kernel in the l layer; expression replaceability quantifies the replaceability of the convolution kernel expression on the same layer, and the metric value ranges from 0 to 1]；
Step 402, calculating the expression replaceability of activation, wherein the expression replaceability of activation is defined as:
AR (l, i) represents the ratio of the activation values of the corresponding convolution kernels, wherein the ratio is not a 0 value, and the expression replaceability of activation represents the replaceability of the expression of the effective activation value output by the target convolution kernel in the feature space.
For the existing remote sensing image building identification model, the generalization upper bound obtained by the traditional theoretical method is limited to the evaluation of the model. They do not specifically quantify the generalization ability of deep learning models. The method for setting the test set needs more skills to ensure the balance of data distribution in the test set, and is difficult to truly reflect the performance of the model on unseen data. There is therefore a need for an index that quantifies model generalization that can directly compare the generalization ability of a model without using a test set. The aim of identifying the building is improved. In this regard, the generalization capability of the model is directly quantified using the main structural convolution kernel of the building identification model. Each convolution kernel has the ability to extract features. And the final recognizer recognizes the extracted features to finish reasoning the unseen data, namely embodying the generalization. Numerous analyses in parameter space also verify the importance of the convolution kernel to model generalization. The performance of the convolution kernel in the feature space also belongs to the performance of the generalization capability of the model. Naturally, as the convolution kernel can present a richer representation of features in the feature space, these rich features will be more favorable for predicting new unseen data. The measure of richness of the model acquisition features has important significance for evaluating generalization. The method can further quantize the generalization capability of the model by measuring the richness of the characteristics obtained by the convolution kernel, so that the model is pruned, and the identification precision of the remote sensing image building by the model is effectively improved.
Drawings
FIG. 1 is a schematic flow diagram of the process of the present invention;
FIG. 2 is an alternative schematic representation of the expression of activation of an embodiment of the method of the present invention.
Detailed Description
The present invention is further illustrated by the following examples and the accompanying drawings, but the present invention is not limited thereto in any way, and any modifications or alterations based on the teaching of the present invention are within the scope of the present invention.
As shown in fig. 1, the method for identifying buildings based on remote sensing images with alternative activation expression comprises the following steps:
step 1, obtaining a building data set of a remote sensing image;
step 2, training a common deep neural network model;
step 3, calculating and identifying an independent maximum response graph of each convolution kernel in the model;
step 4, calculating the activation expression replaceability of each convolution kernel;
step 5, pruning the model convolution kernels according to the activation expression replaceability of each convolution kernel, and keeping small activation expression replaceability;
and 6, using the trimmed deep neural network model to identify the remote sensing image building.
In a remote sensing image building identification task, the training set is assumed to be D ═ X, y. X represents the nth remote sensing image, and y represents a building label corresponding to the nth remote sensing image. And the weight Θ of the deep neural network that needs training is { W, b }. W denotes the first convolutional layer, and similarly b denotes the offset on the first layer. Theta is obtained by defining a loss function of an identification task and using a BP algorithm*So that the model can achieve a high recognition accuracy on the data set D while still maintaining a small error value. After the model is converged, for each new input image, the image can be converted into corresponding feature vectors through the processing of a convolution kernel. The last recognizer of the model can correctly recognize the image.
Each convolution kernel will produce a different response to a different feature. The output of each convolution kernel is defined as h (X, θ), which represents the output of the ith convolution kernel on the l-th layer. The maximum response graph algorithm proposed by Erhan et al, which is intended to obtain the features that result in the maximum response of the convolution kernel output, is defined as follows,
X*＝argmaxhl，i(X，Θ*)
the X of the initial input is random noise. For a trained weight Θ*. Fixing corresponding weight value, using gradient rising algorithm to iteratively update X*So that it can maximize the output activation value of the ith convolution kernel of the l-th layer. Finally obtained X*Then a visual representation that can cause the convolution kernel to produce the maximum response is represented.
However, in the maximum response map algorithm, only the activation value of the target convolution kernel is considered to be maximized. We have found that the resulting visual representation also causes other convolution kernels in the same layer to produce a high response, i.e. the resulting visual representation is characteristically entangled with the representations of the other convolution kernels. The resulting image does not represent the representation of the corresponding convolution kernel in the feature space well. Therefore, in order to better obtain the unwrapping characteristic of the convolution kernel in the characteristic space, the embodiment modifies the target of the maximum response map algorithm. The activation value output by the target convolution kernel is maximum, and meanwhile, other convolution kernels are guaranteed to obtain less input as far as possible.
Further, in the calculation process of the independent maximum response graph in step 3, the objective function is
Wherein, X of the initial input is random noise theta*For the trained weights, J represents the number of all convolution kernels in the l layers, and the output of each convolution kernel is hl，i(X，Θ*) Which represents the output of the ith convolution kernel at the ith layer,hl，-i(X，Θ*) Other characteristic graphs representing the output activation values excluding the target i convolution kernels, argmax (X) representing the maximum response graph of the output, and X representing the independent maximum response graph of the final output;
fixing corresponding weight value, using gradient rising algorithm to iteratively update X*So that it can make the output activation value of ith convolution kernel of l layer be maximum and X obtained by target function*The target convolution kernel output can be enabled to be as large as possible, and simultaneously, the integral output of other convolution kernels is ensured to be small, and the final X is obtained*Then the independent maximum response map in feature space for the corresponding convolution kernel.
For remote sensing image building identification, the model can predict new data by relying on various characteristics. When the new remote sensing image has a good prediction result, the model has strong generalization capability. We can help us measure the model generalization ability by measuring the characteristics of the model's representation in the feature space.
Currently, the unwrapping characteristic of each convolution kernel can be obtained through an independent maximum response graph algorithm. But different convolution kernels may produce a repetitive expression, there being other convolution kernels that produce a higher response for the unwrapped feature. The method of the invention provides a method for expressing an alternative Repeatable Stabilization (RS) to quantify the repeatability of expressions of other convolution kernels, namely the replaceability of the expressions of the convolution kernels on a feature space.
And (4) obtaining the unwrapping characteristic of the target convolution kernel by using an independent maximum response graph algorithm, and propagating the unwrapping characteristic of the target convolution kernel to the layer where the target convolution kernel is located in the forward direction. If the activation value of the target convolution kernel at the corresponding layer is maximum, then the feature indicating the convolution kernel is that other convolution kernels cannot be replaced. Conversely, when there are other convolution kernels in the same layer whose activation values are greater than the activation value of the target convolution kernel, then the feature representing the target convolution kernel may be replaced. The feature that can be replaced is missing as well as other convolution kernels that express the feature, so the convolution kernel is not important.
Further, the alternative calculation of the activation expression includes steps 401 and 402.
Step 401, calculating the expression replaceability, wherein the formula of the expression replaceability is as follows
RS (l, i) represents the property that the unwrapping feature of the ith convolution kernel of the ith layer can be replaced by other convolution kernels on the same layer, wherein | { xlL represents the total number of the ith layer of convolution kernels, IAM (l, i) represents the characteristic representation of an independent maximum response graph generated by the ith layer of convolution kernels, and f (IAM (l, i)) represents the activation value of the ith layer of convolution kernels obtained by forward propagation of the characteristic representation of the generated independent maximum response graph; i { xl，j：xl，j＞xl，iThe I represents the number of convolution kernels which are larger than the activation value of the ith convolution kernel in the l layer; expression replaceability quantifies the replaceability of the convolution kernel expression on the same layer, and the metric value ranges from 0 to 1]。
A convolution kernel with low expression replaceability may indicate that the expression is not easily replaced. And when the expression replaceability is high, two meanings are given. One is that the expression of the convolution kernel is easily replaced by other convolution kernels. The other indicates that the convolution kernel does not learn any features. Since f (x) is close to 0 as known from the independent maximum response graph formula when the target convolution kernel does not respond to any feature, the result is that the average response value of the other convolution kernels is minimal. These convolution kernels will produce nearly similar results. But the convolution kernel does not learn any features, and can also cause higher response of other convolution kernels and make the expression replaceability value of the convolution kernel higher. The alternative expression also means that the convolution kernel does not learn the features.
The invention further proposes an Activated expression alternative Activated Representational Substistion (ARS) that can uniformly represent the replaceability of convolution kernel expressions. According to the independent maximum response graph formula, the output is 0 when the target convolution kernel does not learn the characteristics. And (4) alternatively representing the activation condition of the convolution kernel by using an expression, and representing the proportion of non-zero values in the output characteristic diagram.
Step 402, combining the expression replaceability and the activation response value of the convolution kernel, calculating the activated expression replaceability, wherein the activated expression replaceability is defined as:
AR (l, i) represents the ratio of the activation values of the corresponding convolution kernels, wherein the ratio is not a 0 value, and the expression replaceability of activation represents the replaceability of the expression of the effective activation value output by the target convolution kernel in the feature space.
A convolution kernel with low replaceability of the activated expression means that the activated expression on the same layer is less easily replaced, which is important for generalization of the model. As the value of ARS becomes larger, it means that the convolution kernel changes from repetitive expression to meaningless expression. Convolution kernels with high ARS are not important for model generalization.
The realization principle of activated expression replaceability (ARS) is that the feature richness of a building model of a remote sensing image is measured, so that the generalization characteristic of the model is measured, and the recognition result of the recognition model is improved in a targeted manner. The method can also excellently improve the identification precision of the building model of the remote sensing image. An activated Alternate Representation (ARS) principle diagram is shown in fig. 2, where the gray scale and shape represent different features learned by the remote sensing image deep learning model. Because the number of model convolution kernels is fixed, there may be richer features for the entire model when the representation of each convolution kernel is not easily replaced by the representations of the other convolution kernels. Therefore, the expression replaceability of the activation of the convolution kernel has a strong relationship with the generalization of the model.
As shown in fig. 2, the input image is a button and a face composed of different features. The model 2 shows expressions which are not easily replaced by other convolution kernels in three convolution kernels, namely, the activated expressions of each convolution kernel are low in replaceability, so that the input image can be correctly identified. And similar expressions exist in the convolution kernels of the model 1, the activation expressions of partial convolution kernels are large in replaceability, and a face cannot be correctly identified.
According to the invention content and the embodiment, the method aims at the expression replaceability index method for the deep learning-based building identification model, which can quantify the replaceability of the expression of each convolution kernel on the same layer in the feature space. The more important the expression replaceability value of the activation of the convolution kernel is lower, the expression replaceability value represents that the convolution kernel is irreplaceable in the feature space, and the method can be used for selectively pruning the convolution kernel, so that the identification accuracy of the remote sensing image building identification model is effectively improved.
Claims (2)
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010314628.4A CN111539306B (en) | 2020-04-21 | 2020-04-21 | A method for building recognition in remote sensing images based on the replaceability of activation expressions |
| AU2021101713A AU2021101713A4 (en) | 2020-04-21 | 2021-04-03 | Remote sensing image building recognition method based on activated representational substitution |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010314628.4A CN111539306B (en) | 2020-04-21 | 2020-04-21 | A method for building recognition in remote sensing images based on the replaceability of activation expressions |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111539306A CN111539306A (en) | 2020-08-14 |
| CN111539306B true CN111539306B (en) | 2021-07-06 |
Family
ID=71972984
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010314628.4A Active CN111539306B (en) | 2020-04-21 | 2020-04-21 | A method for building recognition in remote sensing images based on the replaceability of activation expressions |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN111539306B (en) |
| AU (1) | AU2021101713A4 (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106326886A (en) * | 2016-11-07 | 2017-01-11 | 重庆工商大学 | Finger-vein image quality evaluation method and system based on convolutional neural network |
| CN106446150A (en) * | 2016-09-21 | 2017-02-22 | 北京数字智通科技有限公司 | Method and device for precise vehicle retrieval |
| CN108022647A (en) * | 2017-11-30 | 2018-05-11 | 东北大学 | The good pernicious Forecasting Methodology of Lung neoplasm based on ResNet-Inception models |
| CN109919098A (en) * | 2019-03-08 | 2019-06-21 | 广州视源电子科技股份有限公司 | target object identification method and device |
| CN110580450A (en) * | 2019-08-12 | 2019-12-17 | 西安理工大学 | A traffic sign recognition method based on convolutional neural network |
| CN110633646A (en) * | 2019-08-21 | 2019-12-31 | 数字广东网络建设有限公司 | Method and device for detecting image sensitive information, computer equipment and storage medium |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9870609B2 (en) * | 2016-06-03 | 2018-01-16 | Conduent Business Services, Llc | System and method for assessing usability of captured images |
| CN108492200B (en) * | 2018-02-07 | 2022-06-17 | 中国科学院信息工程研究所 | User attribute inference method and device based on convolutional neural network |
| CN109284779A (en) * | 2018-09-04 | 2019-01-29 | 中国人民解放军陆军工程大学 | Object detection method based on deep full convolution network |
| US11726950B2 (en) * | 2019-09-28 | 2023-08-15 | Intel Corporation | Compute near memory convolution accelerator |
-
2020
- 2020-04-21 CN CN202010314628.4A patent/CN111539306B/en active Active
-
2021
- 2021-04-03 AU AU2021101713A patent/AU2021101713A4/en not_active Ceased
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106446150A (en) * | 2016-09-21 | 2017-02-22 | 北京数字智通科技有限公司 | Method and device for precise vehicle retrieval |
| CN106326886A (en) * | 2016-11-07 | 2017-01-11 | 重庆工商大学 | Finger-vein image quality evaluation method and system based on convolutional neural network |
| CN108022647A (en) * | 2017-11-30 | 2018-05-11 | 东北大学 | The good pernicious Forecasting Methodology of Lung neoplasm based on ResNet-Inception models |
| CN109919098A (en) * | 2019-03-08 | 2019-06-21 | 广州视源电子科技股份有限公司 | target object identification method and device |
| CN110580450A (en) * | 2019-08-12 | 2019-12-17 | 西安理工大学 | A traffic sign recognition method based on convolutional neural network |
| CN110633646A (en) * | 2019-08-21 | 2019-12-31 | 数字广东网络建设有限公司 | Method and device for detecting image sensitive information, computer equipment and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| Visualizing Higher-Layer Features of a Deep Network;Dumitru Erhan，Y. Bengio，Aaron Courville;《ResearchGate》;20090131;第1-13页 * |
| 在反卷积网络中引入数值解可视化卷积神经网络;俞海宝，沈琦，冯国灿;《计算机科学》;20170630;第146-150页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111539306A (en) | 2020-08-14 |
| AU2021101713A4 (en) | 2021-05-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| KR102589303B1 (en) | Method and apparatus for generating fixed point type neural network | |
| CN108095716B (en) | Electrocardiosignal detection method based on confidence rule base and deep neural network | |
| CN107194336B (en) | Polarization SAR Image Classification Method Based on Semi-supervised Deep Distance Metric Network | |
| CN107247989A (en) | A kind of neural network training method and device | |
| CN111160176B (en) | Fusion feature-based ground radar target classification method for one-dimensional convolutional neural network | |
| CN111199270B (en) | Regional wave height forecasting method and terminal based on deep learning | |
| WO2020048389A1 (en) | Method for compressing neural network model, device, and computer apparatus | |
| CN106951825A (en) | A kind of quality of human face image assessment system and implementation method | |
| JP2019197355A (en) | Clustering device, clustering method, and program | |
| CN109002792B (en) | SAR image change detection method based on hierarchical multi-model metric learning | |
| CN110222925A (en) | Performance quantization wire examination method, device and computer readable storage medium | |
| CN109460815A (en) | A kind of monocular depth estimation method | |
| CN115393634A (en) | A real-time detection method for few-shot targets based on transfer learning strategy | |
| CN112766496B (en) | Deep learning model safety guarantee compression method and device based on reinforcement learning | |
| CN110363163A (en) | A SAR target image generation method with controllable azimuth angle | |
| CN111539306B (en) | A method for building recognition in remote sensing images based on the replaceability of activation expressions | |
| CN113035363B (en) | Probability density weighted genetic metabolic disease screening data mixed sampling method | |
| CN114444654A (en) | NAS-oriented training-free neural network performance evaluation method, device and equipment | |
| CN107437112B (en) | A kind of mixing RVM model prediction methods based on the multiple dimensioned kernel function of improvement | |
| CN117095188B (en) | Electric power safety strengthening method and system based on image processing | |
| CN108960406B (en) | MEMS gyroscope random error prediction method based on BFO wavelet neural network | |
| CN115620147B (en) | Differentiable architecture search method and device for deep convolutional neural network | |
| CN116978499A (en) | GRA-WOA-GRU-based glass horseshoe kiln temperature prediction method | |
| CN105740815A (en) | Human body behavior identification method based on deep recursive and hierarchical condition random fields | |
| CN116796821A (en) | Efficient neural network architecture searching method and device for 3D target detection algorithm |
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 |