Spatial and temporal variation of fish assemblages were investigated seasonally from May 2007 to February 2008across 11 study sites in a subtropical small stream, the Puxi Stream, of the Huangshan Mountain. Along the ...Spatial and temporal variation of fish assemblages were investigated seasonally from May 2007 to February 2008across 11 study sites in a subtropical small stream, the Puxi Stream, of the Huangshan Mountain. Along the longitudinal gradientfrom headwater to downstream, fish species richness and abundance increased gradually, but then decreased significantly at thelower reaches. The highest species richness and abundance were observed in August and the lowest in February. Based on analysisof similarities (ANOSIM), fish assemblages were significantly different in spatial variation but not in temporal variation. Althoughdifferences were observed both among sites and among stream orders, the lower R value in order-variation suggestedstream order was not the optimal factor explaining the spatial variation of fish assemblages. In addition, dam construction did notsignificantly alter fish assemblages in the sites adjacent to and immediately downstream to dams. Using cluster analysis andnon-metric Multi Dimensional Scaling analysis (NMS), assemblages were separated into three groups at a Bray-Curtis similarityvalue of 42%: the upper, middle and lower groups. Following analysis of similarity percentages of species contributions (SIMPER),shifts in occurrence or abundance of S. curriculus, Z. platypus, R. bitterling and A.fasciatus contributed most to the differencesamongst the three groups. Standard Deviation Redundancy Analysis (RDA) suggested that habitat structure (such as elevation,substrate, and flow velocity) contributed to the spatial and temporal pattern of fish assemblages in the Puxi Stream. In conclusion,the fish assemblages in Puxi Stream presented significant spatial but not temporal variation. Human disturbance has perhapsinduced the decrease in species diversity in the lower reaches. However, no significant change was observed for fish assemblagesin sites far from and immediately downstream from low-head展开更多
在流式识别方法中,分块识别破坏并行性且消耗资源较大,而限制自注意力机制的上下文识别很难获得所有信息.由此,文中提出轻量化端到端声学架构(CFLASH-Transducer).为了获取细腻的局部特征,采用轻量化的FLASH(Fast Linear Attention with...在流式识别方法中,分块识别破坏并行性且消耗资源较大,而限制自注意力机制的上下文识别很难获得所有信息.由此,文中提出轻量化端到端声学架构(CFLASH-Transducer).为了获取细腻的局部特征,采用轻量化的FLASH(Fast Linear Attention with a Single Head)与卷积神经网络块结合.卷积块中采用Inception V2网络,提取语音信号多尺度的局部特征.再通过Coordinate Attention机制捕获特征的位置信息和多通道之间的相互关联.此外,采用深度可分离卷积,用于特征增强和层间平滑过渡.为了使其可流式化处理音频,采用RNN-T(Recurrent Neural Network Transducer)架构进行训练与解码.将当前块已经计算的全局注意力作为隐变量,传入后续块中,串联各块信息,保留训练的并行性和相关性,并且不会随着序列的增长而消耗计算资源.在开源数据集THCHS30上进行训练与测试,CFLASH-Transducer取得较高的识别率.并且相比离线识别,流式识别精度损失不超过1%.展开更多
基金supported by the National Basic Research Program of China (2009CB119200)Anhui Provincial Natural Science Foundation (090413080)Natural Science Foundation of Anhui Education Bureau (KJ2008B211,KJ2009A110)
文摘Spatial and temporal variation of fish assemblages were investigated seasonally from May 2007 to February 2008across 11 study sites in a subtropical small stream, the Puxi Stream, of the Huangshan Mountain. Along the longitudinal gradientfrom headwater to downstream, fish species richness and abundance increased gradually, but then decreased significantly at thelower reaches. The highest species richness and abundance were observed in August and the lowest in February. Based on analysisof similarities (ANOSIM), fish assemblages were significantly different in spatial variation but not in temporal variation. Althoughdifferences were observed both among sites and among stream orders, the lower R value in order-variation suggestedstream order was not the optimal factor explaining the spatial variation of fish assemblages. In addition, dam construction did notsignificantly alter fish assemblages in the sites adjacent to and immediately downstream to dams. Using cluster analysis andnon-metric Multi Dimensional Scaling analysis (NMS), assemblages were separated into three groups at a Bray-Curtis similarityvalue of 42%: the upper, middle and lower groups. Following analysis of similarity percentages of species contributions (SIMPER),shifts in occurrence or abundance of S. curriculus, Z. platypus, R. bitterling and A.fasciatus contributed most to the differencesamongst the three groups. Standard Deviation Redundancy Analysis (RDA) suggested that habitat structure (such as elevation,substrate, and flow velocity) contributed to the spatial and temporal pattern of fish assemblages in the Puxi Stream. In conclusion,the fish assemblages in Puxi Stream presented significant spatial but not temporal variation. Human disturbance has perhapsinduced the decrease in species diversity in the lower reaches. However, no significant change was observed for fish assemblagesin sites far from and immediately downstream from low-head
文摘在流式识别方法中,分块识别破坏并行性且消耗资源较大,而限制自注意力机制的上下文识别很难获得所有信息.由此,文中提出轻量化端到端声学架构(CFLASH-Transducer).为了获取细腻的局部特征,采用轻量化的FLASH(Fast Linear Attention with a Single Head)与卷积神经网络块结合.卷积块中采用Inception V2网络,提取语音信号多尺度的局部特征.再通过Coordinate Attention机制捕获特征的位置信息和多通道之间的相互关联.此外,采用深度可分离卷积,用于特征增强和层间平滑过渡.为了使其可流式化处理音频,采用RNN-T(Recurrent Neural Network Transducer)架构进行训练与解码.将当前块已经计算的全局注意力作为隐变量,传入后续块中,串联各块信息,保留训练的并行性和相关性,并且不会随着序列的增长而消耗计算资源.在开源数据集THCHS30上进行训练与测试,CFLASH-Transducer取得较高的识别率.并且相比离线识别,流式识别精度损失不超过1%.
