遥感图像去噪对于遥感图像在后续的分类、检测等任务中有着非常重要的作用.为了让去噪后的图像更好地保留边缘细节信息,同时增强深度网络对噪声污染区域的辨识能力,本文结合注意力机制以及感知损失来处理遥感图像,提出了一种新的基于残...遥感图像去噪对于遥感图像在后续的分类、检测等任务中有着非常重要的作用.为了让去噪后的图像更好地保留边缘细节信息,同时增强深度网络对噪声污染区域的辨识能力,本文结合注意力机制以及感知损失来处理遥感图像,提出了一种新的基于残差自编码器的遥感图像去噪网络ARED-VGG.考虑到遥感图像中不同地物大小不同,该网络首先同时使用图像的空间和光谱信息来提取多尺度特征;然后使用残差自编码器网络结构来进行图像空间-光谱多尺度图像重建.为了增加网络的辨识能力,更多地关注网络中提取的高频特征,网络引入了注意力机制.同时为了让去噪后的结果更符合视觉感观,使用了感知损失混合均方误差作为损失函数.从实验结果知,本文所提出的方法在噪声去除和纹理细节保留方面与NLM3D、BM4D、LRMR、HSID-CNN和3DADCNN相比表现更优.在Washington DC mall遥感图像数据集上进行了仿真实验,平均峰值信噪比以及平均结构相似性指标都有较好的结果;在AVIRIS Indian Pines数据集上进行了真实数据实验,以去噪后的结果地物分类指标作为验证,整体分类精度以及Kappa系数分别为96.90%和0.9647;对网络结构进行了消融实验,在两个数据集下,本文所提出的网络结构都能获得更优的结果.本文提出基于注意力机制和感知损失的深度神经网络进行遥感图像去噪,提高了网络的辨识能力,实现了良好的去噪性能,并且有效保持了图像的细节信息和光谱信息.展开更多
A new electrocardiogram(ECG) data compression method is presented.It employs a repeated coding.In this method beat templates are extracted from original signals according to the features of ECG.The data are divided in...A new electrocardiogram(ECG) data compression method is presented.It employs a repeated coding.In this method beat templates are extracted from original signals according to the features of ECG.The data are divided into three parts:beat template,residual and position parameter.The three separate parts are first encoded with LADT,and then use the lossless entropy encoding to maintain a low level of reconstructed waveform distortion.The entropy encoding based on Huffman encoding is employed here as an example.Selections from the MIT BIH arrhythmia database,show that there is substantial improvement in compression ratio (CR) over other single and simple compression methods for comparable percent root mean square difference(PRD).展开更多
In High Efficiency Video Coding, the Residual Quad-Yree （RQT） coding is used to encode the prediction residual for both intra and inter Coding Units （CU） and provides improved coding gains. However, this results i...In High Efficiency Video Coding, the Residual Quad-Yree （RQT） coding is used to encode the prediction residual for both intra and inter Coding Units （CU） and provides improved coding gains. However, this results in much higher computational complexities. To address this problem, we develop two fast RQT algorithms for intra- and inter-prediction residual coding respectively. For intra coding, the proposed algorithm selects the best prediction mode in the rate distortion mode decision process using a Prediction Unit （PU） size-dependent fast RQT depth decision on a reduced prediction mode candidates set from the rough mode decision process. For inter coding, in addition to CU size-dependent fast RQT depth decisions, we propose a discriminant analy- sis-based fast depth decision algorithm to det- ermine the best transform unit size. Experimental results show that on average, we can real- ise a 21.29% encoding time saving and 0.03% bit-rate reduction for intra coding, while 15% of the encoding time can be saved with a negligible coding performance loss for inter coding.展开更多
In this paper, a CMOS image sensor(CIS) is proposed, which can accomplish both decorrelation and entropy coding of image compression directly on the focal plane. The design is based on predictive coding for image deco...In this paper, a CMOS image sensor(CIS) is proposed, which can accomplish both decorrelation and entropy coding of image compression directly on the focal plane. The design is based on predictive coding for image decorrelation. The predictions are performed in analog domain by 2×2 pixel units. Both the prediction residuals and original pixel values are quantized and encoded in parallel. Since the residuals have a peak distribution around zero,the output codewords can be replaced by the valid part of the residuals' binary mode. The compressed bit stream is accessible directly at the output of CIS without extra disposition. Simulation results show that the proposed approach achieves a compression rate of 2. 2 and PSNR of 51 on different test images.展开更多
文摘遥感图像去噪对于遥感图像在后续的分类、检测等任务中有着非常重要的作用.为了让去噪后的图像更好地保留边缘细节信息,同时增强深度网络对噪声污染区域的辨识能力,本文结合注意力机制以及感知损失来处理遥感图像,提出了一种新的基于残差自编码器的遥感图像去噪网络ARED-VGG.考虑到遥感图像中不同地物大小不同,该网络首先同时使用图像的空间和光谱信息来提取多尺度特征;然后使用残差自编码器网络结构来进行图像空间-光谱多尺度图像重建.为了增加网络的辨识能力,更多地关注网络中提取的高频特征,网络引入了注意力机制.同时为了让去噪后的结果更符合视觉感观,使用了感知损失混合均方误差作为损失函数.从实验结果知,本文所提出的方法在噪声去除和纹理细节保留方面与NLM3D、BM4D、LRMR、HSID-CNN和3DADCNN相比表现更优.在Washington DC mall遥感图像数据集上进行了仿真实验,平均峰值信噪比以及平均结构相似性指标都有较好的结果;在AVIRIS Indian Pines数据集上进行了真实数据实验,以去噪后的结果地物分类指标作为验证,整体分类精度以及Kappa系数分别为96.90%和0.9647;对网络结构进行了消融实验,在两个数据集下,本文所提出的网络结构都能获得更优的结果.本文提出基于注意力机制和感知损失的深度神经网络进行遥感图像去噪,提高了网络的辨识能力,实现了良好的去噪性能,并且有效保持了图像的细节信息和光谱信息.
文摘A new electrocardiogram(ECG) data compression method is presented.It employs a repeated coding.In this method beat templates are extracted from original signals according to the features of ECG.The data are divided into three parts:beat template,residual and position parameter.The three separate parts are first encoded with LADT,and then use the lossless entropy encoding to maintain a low level of reconstructed waveform distortion.The entropy encoding based on Huffman encoding is employed here as an example.Selections from the MIT BIH arrhythmia database,show that there is substantial improvement in compression ratio (CR) over other single and simple compression methods for comparable percent root mean square difference(PRD).
基金supported by the National Natural Science Foundation of China under Grant No.61272502the National Basic Research Program of China (973 Program) under Grant No.2010CB327900the National Science Fund for Distinguished Young Scholars under Grant No.61125206
文摘In High Efficiency Video Coding, the Residual Quad-Yree （RQT） coding is used to encode the prediction residual for both intra and inter Coding Units （CU） and provides improved coding gains. However, this results in much higher computational complexities. To address this problem, we develop two fast RQT algorithms for intra- and inter-prediction residual coding respectively. For intra coding, the proposed algorithm selects the best prediction mode in the rate distortion mode decision process using a Prediction Unit （PU） size-dependent fast RQT depth decision on a reduced prediction mode candidates set from the rough mode decision process. For inter coding, in addition to CU size-dependent fast RQT depth decisions, we propose a discriminant analy- sis-based fast depth decision algorithm to det- ermine the best transform unit size. Experimental results show that on average, we can real- ise a 21.29% encoding time saving and 0.03% bit-rate reduction for intra coding, while 15% of the encoding time can be saved with a negligible coding performance loss for inter coding.
基金Supported by the National Natural Science Foundation of China(No.61036004)Tianjin Research Program of Application Foundation and Advanced Technology(No.13JCQNJC00600)
文摘In this paper, a CMOS image sensor(CIS) is proposed, which can accomplish both decorrelation and entropy coding of image compression directly on the focal plane. The design is based on predictive coding for image decorrelation. The predictions are performed in analog domain by 2×2 pixel units. Both the prediction residuals and original pixel values are quantized and encoded in parallel. Since the residuals have a peak distribution around zero,the output codewords can be replaced by the valid part of the residuals' binary mode. The compressed bit stream is accessible directly at the output of CIS without extra disposition. Simulation results show that the proposed approach achieves a compression rate of 2. 2 and PSNR of 51 on different test images.
