A Method for Generating Stacked Training Samples Using Generative Adversarial Networks
A technology for training samples and network generation, applied in instruments, computing, character and pattern recognition, etc., can solve the problems of difficult to collect samples, difficult to collect samples, etc., to achieve accurate and robust detection models, increase quantity and quality, and reduce costs. Effect
Active Publication Date: 2020-04-07
南京中设航空科技发展有限公司
View PDF7 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0008] The purpose of the present invention is to provide a method for generating accumulation training samples using a generative confrontation network to solve the problem in the background that it is difficult to collect samples due to the identification of aerial accumulations, and the types of accumulations are varied and ever-changing. Difficult to collect enough samples to train a robust model
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 moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment Construction
[0028] Embodiments of the present invention will be described in further detail below in conjunction with the accompanying drawings.
[0029] A method of generating accumulation training samples by generating confrontation network of the present invention comprises the following steps:
[0030] S1: Create a one-to-one correspondence training set between accumulations and points. When generating annotations, manually label samples of different accumulations to form a one-to-one correspondence between accumulations and samples;
[0031] Mark random points on the same geometric plane and perform random distribution to produce more indicator deposits with different textures to form a one-to-one data pair between labels and samples. The specific correspondence is as follows figure 1 shown.
[0032] S2: Use the discriminator and generator to train the GAN model. The GAN model structure is as follows figure 2 As shown, the data is calibrated one by one, and U-Net is used as the ge...
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 discloses a method for generating a deposit training sample by using a generative adversarial network. The method comprises the following steps of creating a one-to-one correspondence training set of deposits and points; using a discriminator and a generator to train a GAN model; enabling the random generator to generate a random distribution set of the points of the plane; generating more samples by using the generated random points; and optimizing the GAN model by using the generated sample. According to the method, a more robust model can be obtained, the number and the quality of rare samples can be increased by utilizing the model, the data acquisition cost is reduced, the detection model of the traffic administration is more accurate and robust, and the detection of thetraffic administration is better served.
Description
technical field [0001] The invention belongs to the technical field of computer image processing for randomly generating more training data according to existing training samples (such as white garbage and accumulations) of road administration law enforcement and maintenance, and specifically relates to a method for generating training samples of accumulations by using a generative confrontation network method. Background technique [0002] The first neural network model was proposed around 1960. During this period, through the continuous efforts of many scholars, the neural network has been continuously optimized. However, due to the lack of a large amount of data and the limitation of computing power of the computer, the neural network algorithm has not played its role. There is potential. Until the introduction of deep learning and its theory in 2006, combined with the massive data and efficient computing power of the Internet, deep learning has received great attention ...
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
Login to View More Patent Type & Authority Patents(China)
IPC IPC(8): G06K9/62
Inventor 周敏朱志超王勇杨健曾元图尔荪艾力
Owner 南京中设航空科技发展有限公司