Unlock instant, AI-driven research and patent intelligence for your innovation.
Feature extraction method based on feature enhancement model
What is Al technical title?
Al technical title is built by PatSnap Al team. It summarizes the technical point description of the patent document.
A feature extraction and model technology, applied in neural learning methods, biological neural network models, pattern recognition in signals, etc., can solve the problem of losing feature position information, achieve accurate features, and improve robustness
Pending Publication Date: 2022-05-27
HARBIN ENG UNIV
View PDF0 Cites 0 Cited by
Summary
Abstract
Description
Claims
Application Information
AI Technical Summary
This helps you quickly interpret patents by identifying the three key elements:
Problems solved by technology
Method used
Benefits of technology
Problems solved by technology
In addition, in order to effectively reduce the number of training parameters, the fully connected layer will lose the feature position information when integrating the features, and the pooling operation will discard the details of the information to a certain extent.
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more
Image
Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
Click on the blue label to locate the original text in one second.
Reading with bidirectional positioning of images and text.
Smart Image
Examples
Experimental program
Comparison scheme
Effect test
specific Embodiment approach 1
[0015] Specific implementation mode 1: refer to figure 1 Specifically describing this embodiment, the feature extraction method based on the feature enhancement model described in this embodiment includes the following steps:
[0016] Step 1: Set the loss threshold, then obtain the training data set, and use the training data set to train the convolutional neural network. When the error of the convolutional neural network is less than or equal to the loss threshold, extract the feature matrix after pooling of the convolutional neural network;
[0017] Step 2: Perform affine transformation on the feature matrix;
[0018] Step 3: For the feature matrix after affine transformation, strengthen the resistance to feature change of the local position in the feature matrix to obtain a feature matrix N;
[0019] Step 4: Input the feature matrix N into the fully connected layer of the convolutional neural network for retraining to obtain the final feature.
specific Embodiment approach 2
[0020] Embodiment 2: This embodiment is a further description of Embodiment 1. The difference between this embodiment and Embodiment 1 is that the affine transformation is expressed as:
[0021]
[0022] where, (x Source ,y Source ) represents the feature points of the original matrix, (x Targea ,y Target ) represents the feature points of the matrix after affine transformation, θ 11 , θ 12 , θ 13 , θ 21 , θ 22 and θ 23 Represents affine transformation coefficients.
specific Embodiment approach 3
[0023] Embodiment 3: This embodiment is a further description of Embodiment 1. The difference between this embodiment and Embodiment 1 is that the enhancement of the ability to resist feature change in the third step includes channel dimension enhancement and space dimension enhancement.
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More
PUM
Login to View More
Abstract
A feature extraction method based on a feature enhancement model relates to the technical field of underwater sound feature extraction, aims to solve the problem that feature position information is lost when features are integrated in the prior art, and comprises the following steps: 1, setting a loss threshold value, then obtaining a training data set, and training a convolutional neural network by using the training data set to obtain a convolutional neural network; when the error of the convolutional neural network is smaller than or equal to a loss threshold value, extracting a pooled feature matrix of the convolutional neural network; 2, performing affine transformation on the feature matrix; 3, for the feature matrix after affine transformation, enhancing the feature change resistance of local positions in the feature matrix to obtain a feature matrix N; and step 4, inputting the feature matrix N into a full connection layer of the convolutional neural network for re-training to obtain a final feature. The problem that feature position information is lost when features are integrated in the prior art is effectively solved.
Description
technical field [0001] The invention relates to the technical field of underwater sound feature extraction, in particular to a feature extraction method based on a feature enhancement model. Background technique [0002] Compared with the deep neural network, the convolutional neural network has different features in the training process, which makes the convolutional neural network have a better tolerance for translation, and the tolerance of the convolutional neural network relative to the rotation. The level has dropped a lot. Moreover, in order to effectively reduce the number of training parameters, the fully connected layer will lose the feature location information when integrating the features, and the pooling operation will discard the details of the information to a certain extent. SUMMARY OF THE INVENTION [0003] The purpose of the present invention is to propose a feature extraction method based on a feature enhancement model, aiming at the problem of losing ...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More
Application Information
Patent Timeline
Application Date:The date an application was filed.
Publication Date:The date a patent or application was officially published.
First Publication Date:The earliest publication date of a patent with the same application number.
Issue Date:Publication date of the patent grant document.
PCT Entry Date:The Entry date of PCT National Phase.
Estimated Expiry Date:The statutory expiry date of a patent right according to the Patent Law, and it is the longest term of protection that the patent right can achieve without the termination of the patent right due to other reasons(Term extension factor has been taken into account ).
Invalid Date:Actual expiry date is based on effective date or publication date of legal transaction data of invalid patent.
Login to View More 
Login to View More