A graphic processing unit (GPU)-accelerated biological species recognition method using partially connected neural evolutionary network model is introduced in this paper. The partial connected neural evolutionary network adopted in the paper can overcome the disadvantage of traditional neural network with small inputs. The whole image is considered as the input of the neural network, so the maximal features can be kept for recognition. To speed up the recognition process of the neural network, a fast implementation of the partially connected neural network was conducted on NVIDIA Tesla C1060 using the NVIDIA compute unified device architecture (CUDA) framework. Image sets of eight biological species were obtained to test the GPU implementation and counterpart serial CPU implementation, and experiment results showed GPU implementation works effectively on both recognition rate and speed, and gained 343 speedup over its counterpart CPU implementation. Comparing to feature-based recognition method on the same recognition task, the method also achieved an acceptable correct rate of 84.6% when testing on eight biological species.
In this paper,a new type of neural network model - Partially Connected Neural Evolutionary (PARCONE) was introduced to recognize a face gender. The neural network has a mesh structure in which each neuron didn't connect to all other neurons but maintain a fixed number of connections with other neurons. In training,the evolutionary computation method was used to improve the neural network performance by change the connection neurons and its connection weights. With this new model,no feature extraction is needed and all of the pixels of a sample image can be used as the inputs of the neural network. The gender recognition experiment was made on 490 face images (245 females and 245 males from Color FERET database),which include not only frontal faces but also the faces rotated from-40°-40° in the direction of horizontal. After 300-600 generations' evolution,the gender recognition rate,rejection rate and error rate of the positive examples respectively are 96.2%,1.1%,and 2.7%. Furthermore,a large-scale GPU parallel computing method was used to accelerate neural network training. The experimental results show that the new neural model has a better pattern recognition ability and may be applied to many other pattern recognitions which need a large amount of input information.
In this paper, the 3-D Wavelet-Fractal coder was used to compress the hyperspectral remote sensing image, which is a combination of 3-D improved set partitioning in hierarchical trees （SPIHT） coding and 3-D fractal coding. Hyperspectral image date cube was first translated by 3-D wavelet and the 3-D fractal compression ceding was applied to lowest frequency subband. The remaining coefficients of higher frequency sub-bands were encoding by 3-D improved SPIHT. We used the block set instead of the hierarchical trees to enhance SPIHT＇s flexibility. The classical eight kinds of affme transformations in 2-D fractal image compression were generalized to nineteen for the 3-D fractal image compression. The new compression method had been tested on MATLAB. The experiment results indicate that we can gain high compression ratios and the information loss is acceptable.
人机交互研究领域中行为分析与识别是当前研究的一个热点,行为序列分割是行为分析与识别的基础.鉴于强度摄像机视频在进行行为分割时对光线、视角变化过于敏感,提出了一种由深度视频提取的骨架信息,基于本征维数与置信度二次判断的无监督行为序列分割算法.首先,通过Kinect跟踪人体20个骨骼关节点数据,获得视频中人的姿态,通过提取关节点极坐标位置信息来描述行为特征;然后通过奇异值分解(sigular value decomposition,SVD)估计行为序列的本征维数,确定数据对应的低维流形,通过检测特征数据在该流形上投影误差的突变来找到分割帧,并对分割出来的行为序列进行类别标记.每找到一个分割帧就对当前标记类包含样本和当前标记类的前一类包含样本进行基于置信度的二次判断,找到前一类最优分割帧并初始化继续分割.最后采用随机森林模型对分割结果进行识别验证.实验结果表明采用本文算法可以明确分割出代表不同模式的行为片段.
