CN117237703A - Zero-sample day-night domain self-adaptive training method and image classification method based on spectrum analysis - Google Patents
Zero-sample day-night domain self-adaptive training method and image classification method based on spectrum analysis Download PDFInfo
- Publication number
- CN117237703A CN117237703A CN202311062407.2A CN202311062407A CN117237703A CN 117237703 A CN117237703 A CN 117237703A CN 202311062407 A CN202311062407 A CN 202311062407A CN 117237703 A CN117237703 A CN 117237703A
- Authority
- CN
- China
- Prior art keywords
- feature map
- training
- determining
- spectrum
- sample
- 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
Links
- 238000012549 training Methods 0.000 title claims abstract description 101
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000010183 spectrum analysis Methods 0.000 title claims abstract description 24
- 238000010586 diagram Methods 0.000 claims abstract description 24
- 238000013145 classification model Methods 0.000 claims abstract description 11
- 230000006870 function Effects 0.000 claims description 25
- 238000001228 spectrum Methods 0.000 claims description 25
- 230000003044 adaptive effect Effects 0.000 claims description 15
- 230000003595 spectral effect Effects 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 2
- 230000006978 adaptation Effects 0.000 abstract description 8
- 238000001914 filtration Methods 0.000 description 10
- 230000009286 beneficial effect Effects 0.000 description 6
- 238000012800 visualization Methods 0.000 description 5
- 238000001514 detection method Methods 0.000 description 4
- 230000002708 enhancing effect Effects 0.000 description 3
- 238000005286 illumination Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000007493 shaping process Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Classifications
-
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a zero sample day-night domain self-adaptive training method based on spectrum analysis, which comprises the following steps: receiving a daytime image training set; acquiring an original training feature diagram according to the daytime image training set; determining an enhanced feature map according to the original training feature map; determining a supervision contrast loss function according to the original training feature map and the enhancement feature map; calculating an output descriptor according to the original training feature map; determining a cross entropy loss function according to the enhancement feature map and the output descriptor; and updating network parameters according to the supervision comparison loss function and the cross entropy loss function to obtain the self-adaptive classification model. The invention also provides a zero-sample day-night domain self-adaptive image classification method based on spectrum analysis, which can remove the influence of generalization irrelevant factors and retain generalization relevant object characteristics so as to improve the adaptation performance of the model in a night domain.
Description
Technical Field
The invention belongs to the technical field of depth image recognition, and particularly relates to a zero sample day-night domain self-adaptive training method and an image classification method based on spectrum analysis.
Background
In practical applications, the depth image recognition model is susceptible to illumination variations. For example, when a model trained on daytime scenes is applied to nighttime scenes, the performance of the model is often significantly degraded due to the influence of domain transfer, and thus, zero-sample day-night domain adaptation remains an important research area. The goal of domain adaptation is to train a model on a source domain dataset, making the model perform well on a different but similar target domain dataset, and domain adaptation can save expensive data collection and annotation costs. The existing depth domain adaptive methods can be mainly divided into two types: a method based on differential optimization and a method based on resistance learning. The zero sample day-night domain adaptation task aims to generalize well the model trained on day data only to the night domain. The zero sample day-night domain adaptation task is more challenging than traditional domain adaptation because it cannot access the target domain (night scene) data during training. The existing zero sample day-night domain self-adaptation method focuses on extracting a color-unchanged representation to relieve the low light influence of a night scene, and ignores the influence of other factors (such as textures and styles) on generalization capability, so that generalization performance of a model is poor, and adaptation performance of the model in a night domain is poor.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a zero-sample day-night domain self-adaptive training method and an image classification method based on spectrum analysis. The technical problems to be solved by the invention are realized by the following technical scheme:
a zero sample day-night domain self-adaptive training method based on spectrum analysis comprises the following steps:
receiving a daytime image training set;
acquiring an original training feature map according to the daytime image training set;
determining an enhanced feature map according to the original training feature map;
determining a supervision contrast loss function according to the original training feature map and the enhancement feature map;
calculating an output descriptor according to the original training feature map;
determining a cross entropy loss function from the enhancement feature map and the output descriptor;
and training and updating network parameters according to the supervision comparison loss function and the cross entropy loss function to obtain a self-adaptive classification model.
In one embodiment of the present invention, the determining an enhancement feature map according to the original training feature map includes:
determining the spectrum characteristics of the feature map according to the original training feature map;
determining a spectrum high-frequency component according to the spectrum characteristics of the characteristic diagram;
determining a high-pass feature map according to the frequency spectrum high-frequency component;
and determining an enhancement feature map according to the original training feature map and the high-pass feature map.
In one embodiment of the present invention, the determining the feature map spectral feature according to the original training feature map includes:
and carrying out Fourier transform on the original training feature map to obtain the spectrum features of the feature map.
In one embodiment of the present invention, the determining the spectral high-frequency component according to the feature map spectral feature includes:
and removing the low-frequency component of the spectral characteristics of the characteristic map through a high-pass filter to obtain a spectral high-frequency component.
In one embodiment of the present invention, the determining a high-pass feature map according to the spectral high-frequency component includes:
and carrying out inverse Fourier transform on the frequency spectrum high-frequency component to obtain a high-pass characteristic diagram.
In one embodiment of the invention, determining an enhancement feature map from the original training feature map and the high-pass feature map comprises:
performing batch splicing processing on the original training feature map and the high-pass feature map to obtain a spliced feature map;
and inputting the spliced characteristic map into a residual error network for conversion, and outputting an enhanced characteristic map.
In one embodiment of the present invention, the calculating the output descriptor according to the original training feature map includes:
determining a dimension reduction feature map according to the original training feature map;
determining a prototype set according to the dimension reduction feature map;
and calculating an output descriptor according to the prototype set.
In one embodiment of the invention, the expression of the supervised contrast loss function is:
wherein y is i And y j Labels representing anchor samples i and samples j, respectively, B represents the batch size,indicating the label in the batch as y j Number of samples of (1) i≠j And->Representing similar index functions s i，j ＝v i T v j /||v i ||||v j The i represents cosine similarity between anchor sample i and sample j, v i And v j The high-level characteristic vectors of the anchor point sample i and the anchor point sample j are respectively represented, and t represents a temperature super-parameter; the original training feature map is used as an anchor point sample, the enhanced feature map is used as a positive sample, and a feature map of a different category from the current input is used as a negative sample.
In one embodiment of the present invention, the expression of the high-pass filter is:
where α and β denote cut-off thresholds for removing low frequency components, and (u, v) denote coordinates of a center spectrum.
The second aspect of the embodiment of the invention provides a zero-sample day-night domain self-adaptive image classification method based on spectrum analysis, which comprises the following steps:
the pre-trained feature extractor performs feature extraction on the data set to be classified and outputs a feature map to be classified;
the feature map to be classified is input into a prototype compensation module after training to calculate a prototype, and an output descriptor to be classified is obtained according to the prototype calculation;
the feature images to be classified and the output descriptors to be classified are spliced, and the spliced feature images to be classified are output;
inputting the spliced characteristic diagrams to be classified into a trained classification network, and outputting classification results;
the pre-trained feature extractor, the trained prototype compensation module and the trained classification network form the self-adaptive classification model obtained by training the training method provided by the first aspect of the embodiment of the invention.
The invention has the beneficial effects that:
the invention is beneficial to extracting generalized representation by comparing the original training feature map with the enhancement feature map so as to reduce the influence of low-frequency components (such as the texture, style, color and other factors of the image). The prototype is calculated from the bottom-layer feature map so as to promote the high layer to contain rich semantic information, thereby being beneficial to enhancing the generalization capability of the model, and improving the classification and detection capability of the model in the night domain under the condition that the night image cannot be accessed in the training process.
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Drawings
Fig. 1 is a flow chart of a zero sample day-night domain adaptive training method based on spectrum analysis according to an embodiment of the present invention;
fig. 2 is an overall network frame diagram of a zero-sample day-night domain adaptive training method based on spectrum analysis according to an embodiment of the present invention;
fig. 3 is a schematic diagram of a knowledge filtering network of a zero sample day-night domain adaptive training method based on spectrum analysis according to an embodiment of the present invention;
FIG. 4 is a visual image of the night domain detection results of the image classification method of the present invention and the prior art;
FIG. 5a is a schematic diagram of the image classification method of the present invention and the prior art feature visualization using t-SNE on a CODaN dataset;
FIG. 5b is a schematic diagram of the image classification method of the present invention and prior art feature visualization using t-SNE on a ShapeNet dataset;
fig. 6 is a schematic diagram of the image classification method of the present invention and the feature map visualization result of the prior art in the coadn dataset.
Detailed Description
The present invention will be described in further detail with reference to specific examples, but embodiments of the present invention are not limited thereto.
Example 1
As shown in fig. 1, 2 and 3, the present embodiment provides a zero-sample day-night domain adaptive training method based on spectrum analysis, which includes the following steps:
step 10, receiving a daytime image training set.
And step 20, obtaining an original training feature map according to the daytime image training set.
In the step, the input daytime image training set is subjected to feature extraction by a pre-trained feature extractor to output an original training feature mapWherein H, W and C respectively represent the height of the original training feature map,Width and number of channels. To reduce the impact of generalization independent factors and extract the generalization representation, frequency transforms are subsequently required to remove the low frequency components of the extracted original training feature map.
And step 30, determining an enhancement feature map according to the original training feature map. Specifically, an enhanced feature map is determined from the original training feature map. Both this step and step 40 are performed by the contrast filter module.
Specifically, step 30 includes steps 31-34:
and step 31, determining the spectrum characteristics of the feature map according to the original training feature map. Specifically, in this step, the original training feature map F b Performing Fourier transform to obtain a characteristic map spectrum characteristic, wherein the obtained characteristic map spectrum characteristic can be expressed as follows:
wherein,i-th feature map representing the original training feature map, i=1,..>Representing fourier transform, (u, v) representing the coordinates of the center spectrum.
And step 32, determining a frequency spectrum high-frequency component according to the frequency spectrum characteristic of the characteristic diagram. Specifically, a high-pass filter removes low-frequency components of spectral features of the feature map to obtain high-frequency components of the spectrum.
In this step, after fourier transform, the low frequency component of the image contains domain specific information, and the high frequency component contains domain invariant information. Next, in order to mitigate the influence of the low frequency component, a high pass filter is defined to remove the low frequency component, the high pass filter being:
where α and β denote cutoff thresholds for removing low frequency components. Then, the high-pass filter removes low-frequency components of the spectral features of the feature map by performing element-by-element multiplication operation with the spectral features of the feature map to obtain output spectral high-frequency components:
wherein, as indicated by the element-wise product,representing the spectral high frequency components of the high pass filter output.
And step 33, determining a high-pass characteristic diagram according to the frequency spectrum high-frequency components. Specifically, the spectrum high-frequency component is subjected to inverse fourier transform to obtain a high-pass characteristic diagram:
wherein,is a filtered high-pass feature map.
After the high-pass filter, the output characteristic diagram contains less background information irrelevant to generalization and contains more abundant object related contents than the original input characteristic diagram, which is beneficial to enhancing the capability of the model for extracting robust characteristics and improving the generalization performance of the model.
And step 34, determining an enhanced feature map according to the original training feature map and the high-pass feature map. This step includes step 341 and step 342:
and 341, performing batch splicing processing on the original training feature map and the high-pass feature map to obtain a spliced feature map.
And 342, inputting the spliced characteristic diagram into a residual network for conversion, and outputting an enhanced characteristic diagram F.
F＝Φ([F a ，F b ] B )；
Wherein the residual network Φ consists of four residual network blocks for converting the connection result, [,] B representing a splice operation for the batch size dimension.
And step 40, determining a supervision contrast loss function according to the original training characteristic diagram and the enhancement characteristic diagram.
To encourage the filtered features to contain more subject-related information, supervised contrast learning (Prannay Khosla et al. "Supervised contrastive learning". In: advances In neural information processing systems (2020), pp.18661-18673.) is employed to further enhance the filtered features. Specifically, taking the initially extracted features of the current input and the corresponding filtered features as anchor points and positive samples, respectively, and the features of a different class than the current input as negative samples, a supervised contrast loss function may be determined:
wherein y is i And y j Labels representing anchor samples i and samples j, respectively, B represents the batch size,indicating the label in the batch as y j Is a number of samples of (a). l (L) i≠j And->Is a similar index function, e.g. if i+.j, l i≠j E {0,1} = 1, otherwise l i≠j ∈{0，1}＝0。s i，j ＝v i T v j /||v i ||||v j I is the cosine similarity between the anchor sample i and sample j, where v i And v j Representing the high-level eigenvectors representing anchor samples i and j, respectively. t is a temperature super parameter and is set to 0.07. The original training feature map is used as an anchor point sample, the enhanced feature map is used as a positive sample, and the class of the enhanced feature map is different from that of the current inputThe feature map serves as a negative sample.
The semantic gap between the original features and the corresponding filtering features is reduced by using the supervision contrast loss, so that the filtering features are promoted to contain rich object related information, and the influence of night scene low illumination is relieved.
Step 50, calculating an output descriptor according to the original training feature map. After the original training feature map is acquired, the step 30 and the step 40 are performed simultaneously in parallel.
After the processing by the comparison filtering module (after the processing of the step 30 and the step 40), the extracted features contain rich object related information, so that the generalization capability of the model is improved. However, the contrast filter module may filter out some discriminant content that is critical to identifying the corresponding object. To this end, multiple prototypes may be computed from the bottom level feature map by a prototype compensation module to enhance semantic information of the high level features. Specifically, step 50 is performed by a prototype-compensation module, comprising steps 51-53:
and step 51, determining a dimension reduction feature map according to the original training feature map.
And step 52, determining a prototype set according to the dimension reduction feature map.
Step 53, calculating the output descriptor from the prototype set.
Specifically, the original training feature map is first subjected to maximum pooling operationPerforming dimension reduction to obtain dimension reduction feature map +.>Then the feature diagram F after dimension reduction bm A set of prototypes was calculated by inputting into a prototype compensation module (Aming Wu et al, "Universal-prototype enhancing for few-shot object detection". In: proceedings of the IEEE/CVF International Conference on Computer vision.2021, pp.9567-9576.). These prototypes constitute the prototype set->Wherein p is i Is the prototype and D is the number of prototypes. Next, a descriptor representing image level information is calculated based on the prototype set P:
wherein,and->Is a convolution parameter +.>Representing the output descriptor->Representing a residual operation of assigning visual features to corresponding prototypes.
Here, the contrast filter module and the prototype compensation module form a knowledge filter network.
Step 60, determining a cross entropy loss function according to the enhancement feature map and the output descriptor.
In this step, specifically, F andthe splicing result of the cross entropy loss function is input into a classification network to obtain the prediction probability, so that the cross entropy loss function is obtained through calculation.
Wherein,representing the prediction probability:
psi consists of two 1 x 1 convolutional layers with ReLU activation function for pair Ji WeiDegree.Is->Is a shaping result of (a). />And->Is a parameter of the full connectivity layer. [,] Ｃ representing the concatenation operation of the channel dimensions. By means of the concatenation operation, descriptors containing the underlying discriminant information can be added>Fused into the feature F, the discrimination capability of the F is enhanced. Cls stands for classification network, which is a fully connected layer, where N is the number of categories in the daytime image training set. The ability of the model to extract domain invariant features can be further enhanced by using a prototype compensation module.
Step 70, training according to the supervision contrast loss function and the cross entropy loss function to update network parameters, and obtaining the self-adaptive classification model after training.
Specifically, training is performed according to the supervision contrast loss function and the cross entropy loss function to update parameters of the contrast filtering module, the prototype compensation module and the classification network, and training is completed to obtain a trained contrast filtering module, a trained prototype compensation module and a trained classification network. The trained prototype compensation module, the trained classification network and the pre-trained feature extractor form an adaptive classification model.
Specifically, the overall training objective (overall loss function) By cross entropy loss function->And supervised contrast loss function->Composition is prepared.
The loss function of the model is:
where λ represents a superparameter, setting λ to 1 balances the two loss terms.
The training method of the embodiment can effectively improve the image classification and target detection capability of the model in the night domain. The contrast filtering module can effectively remove influence of irrelevant factors, reduce influence of low-frequency components (such as styles) and extract domain invariant features. The prototype compensation module is beneficial to the high-level to contain rich semantic information and improves the generalization capability of the model.
It should be noted that, the comparison filtering module and the prototype compensation module need to be used simultaneously during training. The structure of the network model in the training process comprises the following steps: a pre-trained feature extractor, a contrast filter module, a prototype compensation module, and a classification network. However, in the actual reasoning process, the prototype compensation module is reserved by discarding the contrast filtering module, the model is generalized to an unknown target domain, so that the image is classified, that is, after training is completed, the network from which the trained contrast filtering module is removed is used as the self-adaptive classification model, and the self-adaptive classification model obtained after training is completed comprises: a pre-trained feature extractor, a trained prototype compensation module, and a trained classification network.
Example two
The embodiment provides a spectrum analysis-based zero-sample day-night domain self-adaptive image classification method, which is applied to a self-adaptive classification model obtained by training in the embodiment one, wherein the self-adaptive classification model comprises: a pre-trained feature extractor, a trained prototype compensation module, and a trained classification network;
specifically, the zero sample day-night domain self-adaptive image classification method based on spectrum analysis comprises the following steps:
step one, a pre-trained feature extractor performs feature extraction on a data set to be classified, and outputs a feature map to be classified.
And step two, inputting the feature map to be classified into a prototype compensation module after training to calculate a prototype, and obtaining an output descriptor to be classified according to the prototype calculation.
Step three, splicing the feature images to be classified and the output descriptors to be classified, and outputting spliced feature images to be classified;
and step four, inputting the spliced feature images to be classified into a trained classification network for classification, and outputting classification results.
As shown in fig. 4, the image classification method of the present invention (shown in the second row) accurately detects objects in the night image, compared to the result of the fast R-CNN (shown in the first row).
Fig. 5a is a schematic diagram showing the results of feature visualization by using t-SNE on the coan dataset by the present invention and the prior art CIConv, and fig. 5b is a schematic diagram showing the results of feature visualization by using t-SNE on the ShapeNet dataset by the present invention and the prior art CIConv, wherein color points represent different object categories, and it can be seen that more discriminating features can be extracted by the method of the present invention, so that the classification performance is significantly improved.
As shown in fig. 6, compared with the feature map from CIConv (shown in the second column), the features extracted by the method of the present invention (shown in the third column) contain more object-related information and less background information, which is beneficial to improving generalization performance.
Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.
In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Further, one skilled in the art can engage and combine the different embodiments or examples described in this specification.
The foregoing is a further detailed description of the invention in connection with the preferred embodiments, and it is not intended that the invention be limited to the specific embodiments described. It will be apparent to those skilled in the art that several simple deductions or substitutions may be made without departing from the spirit of the invention, and these should be considered to be within the scope of the invention.
Claims (10)
1. A zero sample day-night domain self-adaptive training method based on spectrum analysis is characterized by comprising the following steps:
receiving a daytime image training set;
acquiring an original training feature map according to the daytime image training set;
determining an enhanced feature map according to the original training feature map;
determining a supervision contrast loss function according to the original training feature map and the enhancement feature map;
calculating an output descriptor according to the original training feature map;
determining a cross entropy loss function from the enhancement feature map and the output descriptor;
and training and updating network parameters according to the supervision comparison loss function and the cross entropy loss function to obtain a self-adaptive classification model.
2. The spectrum analysis-based zero-sample day-night domain adaptive training method according to claim 1, wherein said determining an enhanced feature map from said original training feature map comprises:
determining the spectrum characteristics of the feature map according to the original training feature map;
determining a spectrum high-frequency component according to the spectrum characteristics of the characteristic diagram;
determining a high-pass feature map according to the frequency spectrum high-frequency component;
and determining an enhancement feature map according to the original training feature map and the high-pass feature map.
3. The spectrum analysis-based zero-sample day-night domain adaptive training method according to claim 2, wherein said determining feature map spectrum features from the original training feature map comprises:
and carrying out Fourier transform on the original training feature map to obtain the spectrum features of the feature map.
4. The spectrum analysis-based zero-sample day-night domain adaptive training method according to claim 2, wherein said determining spectral high-frequency components according to the spectral features of the feature map comprises:
and removing the low-frequency component of the spectral characteristics of the characteristic map through a high-pass filter to obtain a spectral high-frequency component.
5. The spectrum analysis-based zero-sample day-night domain adaptive training method according to claim 2, wherein the determining a high-pass feature map according to the spectrum high-frequency component comprises:
and carrying out inverse Fourier transform on the frequency spectrum high-frequency component to obtain a high-pass characteristic diagram.
6. The spectral analysis-based zero-sample day-night domain adaptive training method of claim 5, wherein determining an enhanced feature map from the original training feature map and the high-pass feature map comprises:
performing batch splicing processing on the original training feature map and the high-pass feature map to obtain a spliced feature map;
and inputting the spliced characteristic map into a residual error network for conversion, and outputting an enhanced characteristic map.
7. The spectrum analysis-based zero-sample day-night domain adaptive training method according to claim 1, wherein said calculating an output descriptor from said original training feature map comprises:
determining a dimension reduction feature map according to the original training feature map;
determining a prototype set according to the dimension reduction feature map;
and calculating an output descriptor according to the prototype set.
8. The spectrum analysis-based zero-sample day-night domain adaptive training method according to claim 1, wherein the expression of the supervision contrast loss function is:
wherein y is i And y j Labels representing anchor samples i and samples j, respectively, B represents the batch size,indicating the label in the batch as y j Number of samples of (1) i≠j And->Representing similar index functions s i，j ＝v i T v j /||v i ||||v j The i represents cosine similarity between anchor sample i and sample j, v i And v j The high-level characteristic vectors of the anchor point sample i and the anchor point sample j are respectively represented, and t represents a temperature super-parameter; the original trainingThe training feature map is used as an anchor point sample, the enhancement feature map is used as a positive sample, and a feature map of a different category from the current input is used as a negative sample.
9. The spectrum analysis-based zero-sample day-night domain adaptive training method according to claim 1, wherein the expression of the high-pass filter is:
where α and β denote cut-off thresholds for removing low frequency components, and (u, v) denote coordinates of the center spectrum.
10. A zero sample day-night domain self-adaptive image classification method based on spectrum analysis is characterized by comprising the following steps:
the pre-trained feature extractor performs feature extraction on the data set to be classified and outputs a feature map to be classified;
the feature map to be classified is input into a prototype compensation module after training to calculate a prototype, and an output descriptor to be classified is obtained according to the prototype calculation;
the feature images to be classified and the output descriptors to be classified are spliced, and the spliced feature images to be classified are output;
inputting the spliced characteristic diagrams to be classified into a trained classification network, and outputting classification results;
the pre-trained feature extractor, the trained prototype compensation module and the trained classification network form the training method according to any one of claims 1-9, and an adaptive classification model is obtained by training.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311062407.2A CN117237703A (en) | 2023-08-22 | 2023-08-22 | Zero-sample day-night domain self-adaptive training method and image classification method based on spectrum analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311062407.2A CN117237703A (en) | 2023-08-22 | 2023-08-22 | Zero-sample day-night domain self-adaptive training method and image classification method based on spectrum analysis |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117237703A true CN117237703A (en) | 2023-12-15 |
Family
ID=89083465
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311062407.2A Pending CN117237703A (en) | 2023-08-22 | 2023-08-22 | Zero-sample day-night domain self-adaptive training method and image classification method based on spectrum analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117237703A (en) |
-
2023
- 2023-08-22 CN CN202311062407.2A patent/CN117237703A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108830855B (en) | Full convolution network semantic segmentation method based on multi-scale low-level feature fusion | |
| Raja et al. | Color object detection based image retrieval using ROI segmentation with multi-feature method | |
| CN109190752B (en) | Image semantic segmentation method based on global features and local features of deep learning | |
| CN106919920B (en) | Scene recognition method based on convolution characteristics and space vision bag-of-words model | |
| CN111126386B (en) | Sequence domain adaptation method based on countermeasure learning in scene text recognition | |
| Wang et al. | Multifocus image fusion using convolutional neural networks in the discrete wavelet transform domain | |
| CN110211127B (en) | Image partition method based on bicoherence network | |
| CN112990282B (en) | Classification method and device for fine-granularity small sample images | |
| Wan et al. | Generative adversarial multi-task learning for face sketch synthesis and recognition | |
| CN111275732B (en) | Foreground object image segmentation method based on depth convolution neural network | |
| CN114333062B (en) | Pedestrian re-recognition model training method based on heterogeneous dual networks and feature consistency | |
| CN113496221B (en) | Point supervision remote sensing image semantic segmentation method and system based on depth bilateral filtering | |
| López-Cifuentes et al. | Attention-based knowledge distillation in scene recognition: the impact of a dct-driven loss | |
| CN104573701A (en) | Automatic detection method of corn tassel traits | |
| CN116843715B (en) | Multi-view collaborative image segmentation method and system based on deep learning | |
| CN108664968A (en) | A kind of unsupervised text positioning method based on text selection model | |
| CN113378620A (en) | Cross-camera pedestrian re-identification method in surveillance video noise environment | |
| Li et al. | Incremental learning of infrared vehicle detection method based on SSD | |
| CN110555406B (en) | Video moving target identification method based on Haar-like characteristics and CNN matching | |
| Özyurt et al. | A new method for classification of images using convolutional neural network based on Dwt-Svd perceptual hash function | |
| CN117523626A (en) | Pseudo RGB-D face recognition method | |
| CN117237703A (en) | Zero-sample day-night domain self-adaptive training method and image classification method based on spectrum analysis | |
| Ma et al. | Proposing regions from histopathological whole slide image for retrieval using selective search | |
| CN110210311B (en) | Face recognition method based on channel feature fusion sparse representation | |
| Wiling | Locust Genetic Image Processing Classification Model-Based Brain Tumor Classification in MRI Images for Early Diagnosis |
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 |