Deep learning non-neighborhood equalization processing method and system for water color remote sensing image

Abstract

The invention discloses a deep learning non-neighborhood equalization processing method and system for a water color remote sensing image, and relates to the technical field of image processing, and the method comprises the steps: collecting a to-be-processed noise-containing satellite water color image; inputting a to-be-processed original image into a non-local mean value model of a preset deepconvolutional neural network framework; processing the input original image by adopting convolution kernels in N convolution layers of a non-local mean value model of a preset deep convolutional neural network framework, and outputting a noise image corresponding to the original image after network learning; and performing superposition processing on the noise image and the original image to obtain a denoised image. The system comprises an acquisition module, a calculation module, a processing module I, a processing module II, a processing module III, a processing module IV, a processing module V and an output module. On the basis of a deep convolutional network model, a non-local mean value module is added, and random noise can be effectively eliminated.

Description

Technical field [0001] The present invention relates to the technical field of image processing, and more specifically, to a method and system for deep learning non-neighborhood averaging processing of water color remote sensing images. Background technique [0002] With the rapid development of computer images, imaging has been widely used in military, medical, commercial and daily life. In the special application environment, due to the limitation of the imaging system itself and the limitation of the amount of light in the surrounding environment, the difference between the image noise and the information contained in the image itself is small, and the image signal-to-noise ratio after imaging is too low, which leads to the target It is blurred or even covered by background noise, which leads to the need to perform image enhancement processing on images with low signal-to-noise ratio in practical applications. [0003] The signal-to-noise ratio performance of satellite images d...

AI Technical Summary

Problems solved by technology

[0004] At present, the main problem that restricts the improvement of the signal-to-noise ratio of single-frame images is that the signal / noise resolution needs to obey certain prior assumptions, and the lack of sufficient intelligence in complex scenes, especially remote sensing images with large frames, ground objects, etc. Due to the characteristics of complex types and rich texture types, when the similarity between image detail texture and noise is high, it often improves the signal-to-noise ratio and smoothes the detailed components of the ground features too much, which affects the spatial resolution of the image.

[0005] First of all, the histogram equalization method is one of the most effective methods to improve image contrast and expand the dynamic range of gray scale, but because the algorithm does not select the processed data when statistical probability distribution, it may increase the contrast of background and noise and Reducing the contrast of the target signal reduces the gray value of the transformed image, causing some details to disappear, making the final displayed image unclear; secondly, with the continuous development of pattern recognition and machine learning, the signal-to-noise ratio based on machine learning The boosting method was proposed. By establishing a reliable training sample set, the filter can adaptively learn the noise distribution law in various complex scenes. The SNR boosting method using machine learning has stronger adaptability and robustness. However, due to the limited optical aperture of the monitor, the sensitivity of the detector and the limited exposure time, the imaging SNR is low

Images