空洞U-Net神经网络的移动机器人视觉图像增强

Visual Image Enhancement of Mobile Robot Based on Hollow U-Net Neural Network

为了提升移动机器人视觉图像对比度、信息量以及整体质量,提出一种新的基于空洞U-Net神经网络的移动机器人视觉图像增强方法。在由编码器、解码器与跳层连接构成的U-Net网络中,引入残差网络与空洞卷积部分,构建空洞U-Net神经网络,以融合不同层次的像素特征块,并根据灰度等级与频数直方图,增强图像对比度。针对图像中待处理的像素点灰度值,利用其邻域像素点灰度值的中间值滤除图像噪声。根据像素向量场,利用梯度下降法锐化图像边缘,实现视觉图像增强。在实验阶段,选取部分样本训练空洞U-Net神经网络,获取最优网络参数,经测试验证所提方法的图像对比度、信息量以及整体质量上都有大幅提升,具有优越的视觉图像增强效果。

In order to improve the contrast,information and overall quality of mobile robot visual image,a new visual image enhancement of mobile robot based on hollow U-Net neural network is proposed.In the U-Net network composed of encoder,decoder and layer hopping connection,the residual network and hole convolution part are introduced to construct a hole U-Net neural network to fuse different levels of pixel feature blocks,and enhance the image contrast according to the gray level and frequency histogram.For the gray value of the pixel to be processed in the image,the middle value of the gray value of the adjacent pixel is used to filter the image noise.According to the pixel vector field,the gradient descent method is used to sharpen the image edge and realize visual image enhancement.In the experimental stage,some samples are selected to train the hollow U-Net neural network and obtain the optimal network parameters.The test shows that the image contrast,amount of information and overall quality of the proposed method are greatly improved,and has superior visual image enhancement effect.