Distributed aligned unmanned aerial vehicle remote sensing small sample generalization enhancement method and system

Abstract

The invention discloses a distributed alignment unmanned aerial vehicle remote sensing small sample generalization enhancement method and system, belongs to the technical field of remote sensing detection, and is applied to the aspect of unmanned aerial vehicle remote sensing image recognition. The implementation method comprises the following steps: 1, performing optical detection on remote sensing images of the unmanned aerial vehicle to obtain a remote sensing image data set; further, the remote sensing image data set is composed of a remote sensing image marked data set and a remote sensing image unmarked data set; 2, performing remote sensing image feature extraction on the remote sensing image data set by using a deep neural network; 3, calibrating the feature distribution extracted by the marked sample by using the feature distribution statistical magnitude extracted by the unmarked sample; and 4, sampling from the calibrated distribution to expand marked samples, and training a robust classifier. Compared with the prior art, the generalization ability of small sample annotation data is improved in unmanned aerial vehicle remote sensing image recognition.

Description

Technical Field

[0001] The present invention relates to a distribution-aligned UAV remote sensing small sample generalization enhancement method and system, belongs to the field of remote sensing detection technology, and is applied to UAV remote sensing image recognition. Background Art

[0002] UAV remote sensing image recognition technology is crucial in a variety of fields, including agricultural monitoring, environmental protection, and disaster management. By analyzing remote sensing images captured by drones, crop growth can be assessed in real time, environmental changes can be monitored, and the impact of natural disasters can be detected, providing data support for decision-making. Despite significant technological advances in recent years, the complex and changing geographical environment (such as terrain, weather conditions, and varying lighting) still poses challenges to the stability of recognition results.

[0003] Current drone remote sensing image recognition primarily relies on deep learning methods. These methods utilize multi-layered network structures to extract image features, gradually capturing information from low-level details to high-level semantics. Through downsampling and feature fusion, these models achieve excellent computational efficiency and recognition accuracy. However, these methods often require extensive training with labeled data to fully realize their performance.

[0004] To address the problem of insufficient data, few-shot learning technology is gaining increasing attention in drone remote sensing image recognition. Few-shot learning trains models using very few labeled samples, giving them strong generalization capabilities and reducing reliance on large-scale datasets. Although deep learning technology has made progress in this field, complex backgrounds, such as vegetation, buildings, or other environmental interference, still hinder the model's effective extraction of target features. Especially under small sample conditions, background interference has a more significant impact on the model learning process, resulting in reduced robustness and effectiveness of feature extraction. This background interference challenge limits further improvement in drone remote sensing image recognition performance, and there is an urgent need to improve the stability and accuracy of the model through optimization of few-shot learning.

[0005] Therefore, how to improve the generalization ability of small sample labeled data in UAV remote sensing image recognition has become an urgent problem to be solved. Summary of the Invention

[0006] The purpose of this invention is to solve the technical problem of improving the generalization ability of small sample annotated data in UAV remote sensing image recognition, and propose a distribution-aligned UAV remote sensing small sample generalization enhancement method and system.

[0007] The purpose of the present invention is achieved through the following technical solutions:

[0008] The present invention achieves the above-mentioned purpose by adopting the following technical solutions.

[0009] A distribution-aligned small-sample generalization enhancement method for UAV remote sensing includes the following steps:

[0010] Step 1: Perform optical detection on the UAV remote sensing image to obtain a remote sensing image dataset; the remote sensing image dataset is further composed of a remote sensing image labeled dataset as shown in formula (1) and a remote sensing image unlabeled dataset as shown in formula (2);

[0011]

[0012] in, Represents the acquired labeled remote sensing image, y i Represents the category label of the remote sensing image; represents a real number set of remote sensing image data; w, h, c represent the length, width and number of channels of the input remote sensing image respectively, and N represents the number of labeled remote sensing images;

[0013]

[0014] in, represents the acquired unlabeled remote sensing image, and M represents the number of unlabeled remote sensing images;

[0015] Step 2: Use the deep neural network to extract remote sensing image features from the remote sensing image dataset using the method shown in formula (3);

[0016]

[0017] Among them, f represents feature extraction of the acquired remote sensing image, represents the set of real numbers, n c Represents the channel dimension of the output, w, h, c represent the length, width and number of channels of the input remote sensing image respectively;

[0018] Step 3: Use the feature distribution statistics extracted from the unlabeled samples to calibrate the feature distribution extracted from the labeled samples;

[0019] Step 3.1: Perform Gaussian transformation on the remote sensing image features using the method shown in formula (4);

[0020]

[0021] Among them, x' represents the image features after transformation, f(x i ) represents the input image feature. γ represents the transformation amplitude; i represents the sample index;

[0022] Step 3.2: Perform K-means clustering on the features extracted from the unlabeled samples to form classification result clusters;

[0023] Step 3.3: Perform Gaussian distribution statistics on the classification result clusters;

[0024] Step 3.3.1: Perform mean statistics on the classification result clusters using the method shown in formula (5);

[0025]

[0026] Among them, q′ j is the jth sample of the Gaussian-like distribution feature in the N×M query set, It is represented as a referential function that uses the method shown in formula (6) to distinguish the classes; c n Represented as the cluster of the nth category;

[0027]

[0028] Step 3.3.2: Perform covariance statistics on the nth cluster of the classification result cluster using the method shown in formula (7);

[0029]

[0030] Where T is the transpose operation of the matrix;

[0031] Step 3.4: Match the feature distribution of the labeled samples with the feature distribution of the unlabeled samples according to the similarity of the distribution means;

[0032] Calibration 3.5: Calibrate the distribution of the marked feature, and then obtain the mean and covariance of the calibrated distribution as shown in formula (8);

[0033] μ′=βN(μ n )+(1-β)s′ i ,

[0034]

[0035] Where μ′ is the calibrated mean, ∑′ is the calibrated covariance, N is defined as the nearest neighbor function, β determines the weight of the Gaussian statistics of the nearest cluster in the Gaussian statistics of the calibration distribution, and s′ i is the Gaussian-like distribution feature of the support set, α and β are hyperparameters, α determines the degree of dispersion of the features sampled from the calibrated new category distribution;

[0036] Step 4: Augment the labeled samples by sampling from the calibrated distribution and train a robust classifier.

[0037] Step 4.1: Expand the labeled samples to obtain a sampled dataset;

[0038] Step 4.1.1: Construct the mapping relationship between samples and labels to form sample-label pairs (si ,y i );

[0039] Step 4.1.2: For the label y i The labeled sample features are expanded to obtain the sampling data set shown in formula (9);

[0040]

[0041] Step 4.2: Use full connection to train the classifier in the manner shown in formula (10);

[0042]

[0043] Among them, x′ is the sampled data set x generated and randomly sampled samples from the original dataset x, It represents the indicator function, when y i When the label and k categories are consistent The result is 1, and 0 if it is inconsistent. N′ is the data set x after sampling. generated and the total number of samples in the original data set x, K is the total number of categories of samples Pr(y i |x;θ) is the classifier θ that classifies x into y i probability;

[0044] A distributed aligned UAV remote sensing small sample generalization enhancement system is used to implement the above method. The invention discloses a distributed aligned UAV remote sensing small sample generalization enhancement system comprising a remote sensing image acquisition device and a data processor.

[0045] The remote sensing image acquisition device is used to detect the target area and obtain the remote sensing image data set; which is used as the input of the data processor;

[0046] The data processor includes an image feature extraction module, a distribution calibration module, a feature sampling module, and a category decision module;

[0047] The image feature extraction module is used to extract image features from the remote sensing image dataset and use it as input to the distribution calibration module;

[0048] The distribution calibration module is used to calibrate the feature distribution of the extracted small sample by using the feature calibration method of Gaussian distribution statistics, thereby alleviating the distribution deviation caused by sample noise in the small sample distribution and obtaining a robust feature distribution; and using it as the input of the feature sampling module;

[0049] The feature sampling module is used to sample the calibrated distribution, and the sampled features and features extracted from real data are used together for classifier training, which is used as input to the category decision module;

[0050] The category decision module is used to identify and classify the target feature semantics and output the classification result of the target category.

[0051] Beneficial effects:

[0052] Compared with the existing technology, it has the following beneficial effects:

[0053] By using a feature calibration method based on the Gaussian distribution statistics of unlabeled samples, distorted sample distributions can be effectively corrected. Compared with traditional remote sensing target recognition methods, this invention calibrates the sample distribution and expands the sample by sampling from the calibrated distribution. This alleviates the negative impact of insufficient labeled samples on feature extraction in deep learning models, allowing the model to focus more on samples that conform to the true distribution, thereby improving the discriminative ability of extracted features and enhancing the accuracy and robustness of target recognition. In addition, the data processor can use an embedded motherboard and can be integrated into a variety of intelligent monitoring devices, with the advantages of simple implementation, low complexity, strong adaptability, and plug-and-play. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a schematic flow diagram of the present invention; DETAILED DESCRIPTION

[0055] In order to better illustrate the purpose and advantages of the present invention, the following is a further description of the invention in conjunction with the accompanying drawings and examples. It should be noted that the implementation of the present invention is not limited to the following embodiments, and any form of modification or change made to the present invention will fall within the scope of protection of the present invention.

[0056] Example

[0057] like Figure 1 As shown, the present embodiment provides a distribution-aligned UAV remote sensing small sample generalization enhancement method, and the specific implementation steps are as follows:

[0058] Step 1: Perform optical detection on the UAV remote sensing image to obtain a remote sensing image dataset; the remote sensing image dataset is further composed of a remote sensing image labeled dataset as shown in formula (1) and a remote sensing image unlabeled dataset as shown in formula (2);

[0059]

[0060] in, Represents the acquired labeled remote sensing image, y i Represents the category label of the remote sensing image; represents a real number set of remote sensing image data; w, h, c represent the length, width and number of channels of the input remote sensing image respectively, and N represents the number of labeled remote sensing images;

[0061]

[0062] in, represents the acquired unlabeled remote sensing image, and M represents the number of unlabeled remote sensing images;

[0063] In the embodiment, the remote sensing image acquisition device performs optical detection on the target area to obtain remote sensing optical images and construct a remote sensing image data set. The remote sensing image acquisition device is a camera onboard a drone.

[0064] Step 2: Use the deep neural network to extract remote sensing image features from the remote sensing image dataset using the method shown in formula (3);

[0065]

[0066] Among them, f represents feature extraction of the acquired remote sensing image, represents the set of real numbers, n c Represents the channel dimension of the output, w, h, c represent the length, width and number of channels of the input remote sensing image respectively;

[0067] In an embodiment, the image feature extraction module is a WideResNet28 pre-trained model based on a deep learning method.

[0068] Step 3: Use the feature distribution statistics extracted from the unlabeled samples to calibrate the feature distribution extracted from the labeled samples;

[0069] Step 3.1: Perform Gaussian transformation on the remote sensing image features using the method shown in formula (4);

[0070]

[0071] Among them, x' represents the image features after transformation, f(x i ) represents the input image feature. γ represents the transformation amplitude; i represents the sample index;

[0072] Step 3.2: Perform K-means clustering on the features extracted from the unlabeled samples to form classification result clusters;

[0073] Step 3.3: Perform Gaussian distribution statistics on the classification result clusters;

[0074] Step 3.3.1: Perform mean statistics on the classification result clusters using the method shown in formula (5);

[0075]

[0076] Among them, q′ j is the jth sample of the Gaussian-like distribution feature in the N×M query set, It is represented as a referential function that uses the method shown in formula (6) to distinguish the classes; c n Represented as the cluster of the nth category;

[0077]

[0078] Step 3.3.2: Perform covariance statistics on the classification result clusters using the method shown in formula (7);

[0079]

[0080] Where T is the transpose operation of the matrix;

[0081] Step 3.4: Match the feature distribution of the labeled samples with the feature distribution of the unlabeled samples according to the similarity of the distribution means;

[0082] Calibration 3.5: Calibrate the distribution of the marked feature, and then obtain the mean and covariance of the calibrated distribution as shown in formula (8);

[0083] μ′=βN(μ n )+(1-β)s′ i ,

[0084]

[0085] Where N′ is defined as the nearest neighbor function, β determines the weight of the Gaussian statistics of the nearest cluster in the Gaussian statistics of the calibration distribution, and s′ i is the Gaussian-like distribution feature of the support set, α and β are hyperparameters, α determines the degree of dispersion of the features sampled from the calibrated new category distribution;

[0086] In an embodiment, the similarity of distribution means is measured using Euclidean distance.

[0087] Step 4: Augment the labeled samples by sampling from the calibrated distribution and train a robust classifier.

[0088] Step 4.1: Expand the labeled samples to obtain a sampled dataset;

[0089] Step 4.1.1: Construct the mapping relationship between samples and labels to form sample-label pairs (s i ,y i );

[0090] Step 4.1.2: For the label y i The labeled sample features are expanded to obtain the sampling data set shown in formula (9) where x generated is the sample collected from the calibrated distribution;

[0091]

[0092] Step 4.2: Use the fully connected classifier to train the classifier using the loss function shown in formula (10);

[0093]

[0094] Among them, x′ is the sampled data set x generated and randomly sampled samples from the original dataset x, It represents the indicator function, when y i When the label and k categories are consistent The result is 1, and 0 if it is inconsistent. N′ is the data set x after sampling. generated and the total number of samples in the original data set x, K is the total number of categories of samples Pr(y i |x;θ) is the classifier θ that classifies x into y i probability;

[0095] In the embodiment, the sampling method directly adopts the Gaussian distribution sampling method after calibration of the mean and variance, the classifier can also be replaced by other classifiers, and the loss function adopts the commonly used classification loss function, cross entropy loss.

[0096] A distributed aligned UAV remote sensing small sample generalization enhancement system is used to implement the above method. The invention discloses a distributed aligned UAV remote sensing small sample generalization enhancement system comprising a remote sensing image acquisition device and a data processor.

[0097] The remote sensing image acquisition device is used to detect the target area and obtain the remote sensing image data set; which is used as the input of the data processor;

[0098] The data processor includes an image feature extraction module, a distribution calibration module, a feature sampling module, and a category decision module;

[0099] The image feature extraction module is used to extract image features from the remote sensing image dataset and use it as input to the distribution calibration module;

[0100] The distribution calibration module is used to calibrate the feature distribution of the extracted small sample by using the feature calibration method of Gaussian distribution statistics, thereby alleviating the distribution deviation caused by sample noise in the small sample distribution and obtaining a robust feature distribution; and using it as the input of the feature sampling module;

[0101] The feature sampling module is used to sample the calibrated distribution, and the sampled features and features extracted from real data are used together for classifier training, which is used as input to the category decision module;

[0102] The category decision module is used to identify and classify the target feature semantics and output the classification result of the target category.

[0103] In an embodiment, the remote sensing image acquisition device is a camera mounted on a drone. The data processor is a computer or an embedded motherboard. The image feature extraction module is a WideResNet28 pre-trained model based on deep learning methods.

[0104] The above specific description further illustrates the purpose, technical solutions and beneficial effects of the invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A distribution-aligned small sample generalization enhancement method for UAV remote sensing, characterized by: The following steps are included: Step 1: Perform optical detection on the UAV remote sensing image to obtain a remote sensing image dataset; the remote sensing image dataset is further composed of a remote sensing image labeled dataset as shown in formula (1) and a remote sensing image unlabeled dataset as shown in formula (2); in, Represents the acquired labeled remote sensing image, y i Represents the category label of the remote sensing image; represents a real number set of remote sensing image data; w, h, c represent the length, width and number of channels of the input remote sensing image respectively, and N represents the number of labeled remote sensing images; in, represents the acquired unlabeled remote sensing image, and M represents the number of unlabeled remote sensing images; Step 2: Use the deep neural network to extract remote sensing image features from the remote sensing image dataset using the method shown in formula (3); Among them, f represents feature extraction of the acquired remote sensing image, represents the set of real numbers, n c Represents the channel dimension of the output, w, h, c represent the length, width and number of channels of the input remote sensing image respectively; Step 3: Use the feature distribution statistics extracted from unlabeled samples to calibrate the feature distribution extracted from labeled samples; Step 4: Augment the labeled samples by sampling from the calibrated distribution and train a robust classifier.

2. The method for generalization enhancement of small samples of UAV remote sensing using distribution alignment according to claim 1, characterized in that: Step 3 is implemented as follows: Step 3.1: Perform Gaussian transformation on the remote sensing image features using the method shown in formula (4); Among them, x' represents the image features after transformation, f(x i ) represents the input image feature; γ represents the conversion amplitude; i represents the sample index; Step 3.2: Perform K-means clustering on the features extracted from the unlabeled samples to form classification result clusters; Step 3.3: Perform Gaussian distribution statistics on the classification result clusters; Step 3.4: Match the feature distribution of the labeled samples with the feature distribution of the unlabeled samples according to the similarity of the distribution means; Calibration 3.5: Calibrate the distribution of the marked feature, and then obtain the mean and covariance of the calibrated distribution as shown in formula (8); Where μ′ is the calibrated mean, ∑′ is the calibrated covariance, N is defined as the nearest neighbor function, β determines the weight of the Gaussian statistics of the nearest cluster in the Gaussian statistics of the calibration distribution, and s′ i is the Gaussian-like distribution of the support set, α and β are hyperparameters, and α determines the degree of dispersion of the features sampled from the calibrated new category distribution.

3. The method for generalization enhancement of small samples of UAV remote sensing using distribution alignment according to claim 2, characterized in that: The implementation method of step 3.3 is: Step 3.3.1: Perform mean statistics on the classification result clusters using the method shown in formula (5); Among them, q′ j is the jth sample of the Gaussian-like distribution feature in the N×M query set, It is represented as a referential function that uses the method shown in formula (6) to distinguish the classes; c n Represented as the cluster of the nth category; Step 3.3.2: Perform covariance statistics on the nth cluster of the classification result cluster using the method shown in formula (7); Where T is the transpose operation of the matrix.

4. The method for generalization enhancement of small samples of UAV remote sensing using distribution alignment according to claim 1, characterized in that: Step 4 is implemented as follows: Step 4.1: Expand the labeled samples to obtain a sampled dataset; Step 4.2: Use full connection to train the classifier in the manner shown in formula (10); Among them, x′ is the sampled data set x generated and randomly sampled samples from the original dataset x, It represents the indicator function, when y i When the label and k categories are consistent The result is 1, and 0 if it is inconsistent. N′ is the data set x after sampling. generated and the total number of samples in the original data set x, K is the total number of categories of samples Pr(y i |x;θ) is the classifier θ that classifies x into y i probability.

5. The method for generalization enhancement of small samples of UAV remote sensing using distribution alignment according to claim 4, characterized in that: Step 4.1 is implemented as follows: Step 4.1.1: Construct the mapping relationship between samples and labels to form sample-label pairs (s i ,y i ); Step 4.1.2: For the label y i The labeled sample features are expanded to obtain the sampling data set shown in formula (9).

6. A distribution-aligned UAV remote sensing small sample generalization enhancement system implementing the method of claim 1, characterized by: Includes remote sensing image acquisition equipment and data processor; Remote sensing image acquisition equipment is used to detect the target area and obtain remote sensing image data sets; which are used as input to the data processor.

7. The distribution-aligned UAV remote sensing small sample generalization enhancement system according to claim 6, characterized in that: The data processor includes an image feature extraction module, a distribution calibration module, a feature sampling module, and a category decision module; The image feature extraction module is used to extract image features from the remote sensing image dataset and use it as input to the distribution calibration module; The distribution calibration module is used to calibrate the feature distribution of the extracted small sample by the feature calibration method of Gaussian distribution statistics, so as to alleviate the distribution deviation caused by the sample noise in the small sample distribution, and thus obtain a robust feature distribution; It serves as the input of the feature sampling module; The feature sampling module is used to sample the calibrated distribution, and the sampled features and features extracted from real data are used together for classifier training, which is used as input to the category decision module; The category decision module is used to identify and classify the target feature semantics and output the classification result of the target category.

8. The distribution-aligned UAV remote sensing small sample generalization enhancement system according to claim 6, characterized in that: The remote sensing image acquisition device is a drone-mounted camera.

9. The distribution-aligned UAV remote sensing small sample generalization enhancement system according to claim 6, characterized in that: The data processor is a computer or an embedded motherboard.

10. The distribution-aligned UAV remote sensing small sample generalization enhancement system according to claim 7, characterized in that: The image feature extraction module is the WideResNet28 pre-trained model.

Images