Real-time defogging method for aerial image and UAV based on FPGA convolutional neural network

Abstract

The present invention provides a real-time defogging method for aerial images based on FPGA convolutional neural network and an unmanned aerial vehicle. Dehazing parameters; select the dehazing parameters, and instantiate the dehazing parameters in the RAM inside the FPGA; obtain the i-1th frame image and the i-th frame image, obtain the dark channel image of the i-1th frame image and calculate the first Atmospheric light value A of i-1 frame image i‑1 ;Convolutional neural network operation is performed on the i-th frame image, and the rough transmittance t' of the i-th frame image is calculated i (x); Get the grayscale image of the i-th frame image; set the rough transmittance t' i (x) Conduct guided filtering with the grayscale image, and calculate and obtain the fine transmittance t of the i-th frame image i (x); put A i‑1 , t i (x) and i-th frame image I i (x) operation, to obtain the defogged image J of the i-th frame image i (x). It has better universality and can meet the real-time requirements at the same time. The invention is applied in the fields of image processing and computer vision.

Description

technical field [0001] The present invention relates to the fields of image processing and computer vision, in particular to a method for real-time defogging of aerial images based on FPGA convolutional neural network and an unmanned aerial vehicle. Background technique [0002] At present, UAVs are more and more widely used in surveying and mapping, patrolling, disaster prevention and other fields. When UAVs operate or enable visual navigation, they largely rely on UAV images with high imaging quality. When UAVs fly at medium and high altitudes, under adverse weather conditions such as fog and haze, due to the scattering effect of aerosols in the atmosphere, the visibility of the atmosphere is reduced, and the quality and clarity of aerial images are seriously affected, resulting in The problem of degradation and blurring occurs, the natural degree of image color and image contrast are greatly reduced, and the basic information features in the image are distorted and damag...

AI Technical Summary

Problems solved by technology

[0008] In the technical research introduced above, the dehazing algorithm based on image enhancement does not consider the reasons for image degradation, and can improve the contrast of foggy images and highlight image details. However, it will cause a certain loss of image information, and the dehazing effect is average.

[0009] The premise of the dehazing algorithm based on physical model image restoration is that the atmosphere is single-scattering and the medium is uniform. This assumption is not universal, especially the sky area does not meet this assumption.

This type of method often suffers from distortion when processing haze images with large areas of sky

The patent "A FPGA-based real-time water surface dense fog image enhancement method, application number: 201711419415.2, inventor and designer: Zhang Hong, etc." is mainly used in the treatment of water surface dense fog, and cannot be applied to the defog treatment of scenes such as mountainous areas and urban areas. ;

[0010] The image defogging algorithm based on deep learning, this type of algorithm obtains the defogging parameters through training, and has good adaptability to various scenes, but the current research mainly focuses on the back-end line defogging, which has poor real-time performance and is not suitable for airborne Image processing in real time

Images