Unlock instant, AI-driven research and patent intelligence for your innovation.
Method and device for training and detecting multi-scale feature convolutional neural network
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 convolutional neural network and multi-scale feature technology, applied in the field of deep learning, can solve the problems of low accuracy and recognition rate, long calculation time and backwardness, etc., and achieve the effect of improving recognition speed and reducing calculation amount
Pending Publication Date: 2020-05-01
DATONG POWER SUPPLY COMPANY OF STATE GRID SHANXI ELECTRIC POWER +1
View PDF5 Cites 4 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
However, this technology is relatively backward, and the calculation time of the convolutional neural network is often much longer than the calculation time of the artificially designed features, so it will face the contradiction between the detection speed and the detection effect during detection.
Especially in the recognition of tiny targets in large-scale panoramic pictures, the accuracy and recognition rate of traditional neural networks are even lower
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
Embodiment 1
[0034] as attached figure 1 As shown, in order to realize the detection of tiny targets in panoramic images, the embodiment of the present application provides a multi-scale feature convolutional neural network training method, including the following steps:
[0035] Step S11: mark the target to be recognized in the training image, and generate training data for training;
[0036] Step S12: Input the training data into the multi-scale feature convolutional neural network to obtain multiple feature maps;
[0037] Step S13: Generate target pre-selection boxes on multiple feature maps, and train the multi-scale feature convolutional neural network.
[0038] Specifically, in step S11, a considerable number of panoramas of the target to be detected are collected as the training data of the multi-scale feature convolutional neural network. type. And set the label data frame for the target position in the panorama that contains the target to be detected. The label data frame adopt...
Embodiment 2
[0050] as attached image 3 As shown, the embodiment of the present application provides a detection method of a multi-scale feature convolutional neural network, comprising the following steps:
[0051] Step S21: training a multi-scale feature convolutional neural network;
[0052] Step S22: Input the detection data into the multi-scale feature convolutional neural network;
[0053] Step S23: The detection data obtains multiple feature maps through a multi-scale feature convolutional neural network;
[0054] Step S24: Generate default frames on the acquired multiple feature maps respectively;
[0055] Step S25: Filter the default frame, and output the identified crack image of the porcelain bottle.
[0056] Specifically, in step S21, the multi-scale feature convolutional neural network is trained according to the method disclosed in Embodiment 1 to obtain a multi-scale feature convolutional neural network model. If the neural network has already been trained, this step ca...
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
The invention provides a method for training and detecting a multi-scale feature convolutional neural network and the method comprises the following steps: S11, marking a to-be-recognized target in atraining image, and generating training data for training; S12, inputting the training data into a multi-scale feature convolutional neural network to obtain a plurality of feature maps; and S13, generating a target pre-selection box on the plurality of feature maps, and training the multi-scale feature convolutional neural network. Feature extraction adopted in the method is a multi-feature extraction mode, different features are obtained for different feature extractionlayers, a multi-layer feature fusion detection mode is carried out, the features of different layers are fused to obtain rich and accurate fusion features considering the expression capacity of position information and semantic information, and therefore a more accurate detection result is obtained.
Description
technical field [0001] The invention belongs to the technical field of deep learning, and in particular relates to a training and detection method and device for a multi-scale feature convolutional neural network. Background technique [0002] The existing technology usually uses a BP neural network and a network based on radial basis functions and invariant moments. Based on image operations, a series of regions to be detected with different positions and sizes are selected on the picture, and then the regions are directly input into a volume. A neural network is used to obtain classification results. By properly designing the structure of the convolutional neural network, the computer can directly learn the hidden features in the picture, avoiding artificial design features, and can be more widely used in the detection of various types of objects. However, this technology is relatively backward, and the calculation time of convolutional neural network is often much longer...
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