A convolutional network operation unit, a reconfigurable convolutional neural network processor, and a method for implementing image denoising processing
A convolutional neural network and convolutional network technology, applied in image communication, color signal processing circuits, electrical components, etc., can solve problems that are not suitable for large-scale and extensive applications, low efficiency of deep learning networks, and failure to meet performance requirements , to achieve the effect of improving hardware performance and flexibility, easy hardware implementation, and high power consumption
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Publication Date
- 2018-12-07
Smart Images

Figure 1 
Figure 2 
Figure 3
Abstract
Description
technical field
[0001] The invention relates to the field of image processing, in particular to a convolutional network operation unit, a reconfigurable convolutional neural network processor and a method for realizing image denoising processing. Background technique
[0002] Image raindrop and dust removal is of great significance for image processing applications, especially video surveillance and navigation systems. It can be used to restore images or videos polluted by raindrops and dust, and can also be used as a pre-processing operation to help subsequent image recognition or classification.
[0003] Most of the current image noise removal methods are completed by Gaussian filtering, median filtering, and bilateral filtering. These methods have poor processing effects and often cannot meet the needs of specific image processing applications. Therefore, a better method is needed to remove image noise, and the method of convolutional neural network becomes a good choice...
Examples
Embodiment Construction
[0031] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.
[0032] refer to figure 1 , the convolutional network operation unit used in the reconfigurable convolutional neural network processor in the present invention includes 2 reconfigurable separate convolution modules, a nonlinear activation function unit and a multiply-accumulator unit; the first reconfigurable The output of the separation convolution module is the input of the nonlinear activation function unit, the output of the nonlinear activation function unit is the input of the multiplication accumulator unit, and the output of the multiplication accumulator unit is the input of the second reconfigurable separation convolution module;
[0033] The image signal and configuration network parameter signal are input to the first reconfigurable separation convolution module; the first reconfigurable separation convolution module completes the ...