A compression sensing network based on full-image observation and a sensing loss reconstruction method
A technology of compressed sensing and network, applied to instruments, character and pattern recognition, computer components, etc., can solve problems such as blurred images, long training process time, and unobvious semantic information, and achieve the effect of strengthening sampling and reconstruction
A technology of compressed sensing and network, applied to instruments, character and pattern recognition, computer components, etc., can solve problems such as blurred images, long training process time, and unobvious semantic information, and achieve the effect of strengthening sampling and reconstruction
CN108537104AInactive Publication Date: 2018-09-14XIDIAN UNIV
Image
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
[0031] Embodiments and effects of the present invention will be further described in detail below in conjunction with the accompanying drawings.
[0032] refer to figure 1 , the compressed sensing network for full-image observation in the present invention is composed of two sub-networks, one is the observation sub-network, and the other is the reconstruction sub-network. in:
[0033] Observation sub-network, including the first convolutional layer;
[0034] Reconstruct the sub-network, including the deconvolution layer, the second convolution layer, the first Relu activation layer, the third convolution layer, the second Relu activation layer and the fourth convolution layer; the first Relu activation layer is obtained from the first Relu activation layer The output x of the second convolutional layer 1 and 0 take the maximum value as the output f of the first Relu activation layer 1 , namely f 1 =max(0,x 1 ), the second Relu activation layer is the output x from the th...
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 compression sensing network based on full-image observation and a sensing loss reconstruction method. The invention mainly solves the problem of block effect of reconstructionimages recovered by conventional networks. The network comprises an observation sub-network and a reconstruction sub-network. The observation sub-network comprises a first convolution layer and the reconstruction sub-network comprises a deconvolution layer, a second convolution layer, a first Relu activation layer, a third convolution layer, a second Relu activation layer and a fourth convolutionlayer. The first convolution layer, the deconvolution layer, the second convolution layer, the first Relu activation layer, the third convolution layer, the second Relu activation layer and the fourth convolution layer are connected successively from left to right successively, and an output end of the deconvolution layer is connected with an output end of the fourth convolution layer. The network is characterized in that sensing loss is introduced in a training process, so that structure information of reconstruction images is clearer. When the network is used for image reconstruction, blockeffect is prevented, image recovery quality is improved, and the semantic information of reconstruction images is enhanced. The network and the method can be used for image processing.
Description
technical field [0001] The invention belongs to the technical field of compressed sensing, and mainly relates to a compressed sensing network for full-image observation and a reconstruction method for perceptual loss, which can be used for image processing. Background technique [0002] In a large number of practical problems, people tend to collect as little data as possible, or have to collect incomplete data due to objective conditions. Traditional image compression is based on Nyquist sampling for data collection, and starts from the characteristics of the data itself to find and eliminate the hidden redundancy in the data. The result of this is that data compression must be done after the data is completely collected, and the compression process requires complex algorithms, which is contradictory to the performance of equipment that collects and processes a large number of signals. The concept of compressed sensing is proposed to solve this problem. It can directly col...
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
14 Sep 2018
Publication
CN108537104A
- IPC
- G06K9/00; G06K9/62
- CPC
- G06F2218/08; G06F18/214
- Inventors
- η³ε ζ; ζζ±