基于FCN-LSTM的工业烟尘图像分割

工业生产中常根据林格曼烟气黑度判断工业烟尘的污染等级,一种有效的方式是应用计算机视觉系统对工业烟尘进行监测, 其中对烟尘目标进行准确分割是该系统的关键技术。因为工业烟尘具有形状不固定、和云相似度高等特点,现有算法在复杂场景下对烟尘进行分割时容易受到干扰,分割准确度有待提高。针对这一问题,提出一种基于FCN-LSTM的工业烟尘图像分割方法,在全卷积网络对图像空间特征提取的基础上,使用长短时记忆网络提取图像序列的时间信息,通过烟尘的动态特征对运动的烟尘和背景进行区分,增强复杂场景下的抗干扰能力。实验表明,本文模型相比于全卷积网络,在复杂场景下的抗干扰能力有显著提升,能够有效克服来自云的干扰,对全卷积网络分割结果中易出现干扰点的问题也有改善,IoU指标最高有8.04%的提升。     

基于图像处理技术的烟雾检测在家庭火灾防范中应用

针对家庭火灾防范中缺少更有效、快速的实时烟雾检测问题,本文设计了一种基于图像处理技术的烟雾检测算法,能够对家庭住宅内部进行实时烟雾检测,从而实现火灾防范。本算法主要使用单阶段的端到端YOLO v3算法,对输入的烟雾数据进行学习,提取网络中不同卷积层的特征图进行联合训练。实验结果表明,和双阶段网络的烟雾图像检测方法相比,该方法检测速度快,实时效果好,同时也取得了较高的检测准确率,具有较高的实际应用价值。     

基于模式识别的火灾烟雾探测中抗水雾干扰研究

水雾与火灾烟雾在颜色、形态以及面积扩散等方面均具有相似性,容易对基于图像识别的火灾探测技术形成干扰。在同一地点及环境下使用监控系统录制火灾烟雾和水雾的真实视频,从运动方向特征和纹理特征着手进行分析,并利用支持向量机(support vector machine,SVM)对视频进行火灾烟雾和水雾的识别。实验结果表明,火灾烟雾与水雾在运动方向特征及纹理特征中熵和对比度上有明显差异,火灾烟雾识别时使用运动方向特征及纹理特征既能准确的识别出火灾烟雾,又能极大减小疑似烟雾目标带来的干扰。     

Numerical Simulation of Energy Saving Potential in Dispersing Process of Blasted Smoke in Mines

The dispersing process of the blasted smoke in underground mines by ventilation is very important for the safety of personnel, and its energy consumption is also worthy of concern. In this study, the impact of supply air flow rate on the decay process of harmful gases and fan energy, and the role of harmful gases buried by blasted pile in the smoke dispersing process is analyzed with three-dimensional numerical simulation. According to the results, air flow facilitates quick smoke dispersion and the gas emits from the blasted pile is only significant at the beginning of ventilation. It is thus recommended that large amount of air supply is taken to disperse the smoke quickly until the concentration of harmful gases reaches relevant standards, and reduce the amount of air in the remain time with frequency conversion technology to save energy. In the given study, 12 m/s air flow rate is taken for dispersing smoke and 3 m/s air flow rate is taken for the waiting time. The energy consumption for ventilation is reduced by about 50％. In addition, due to the gases in the blasted pile remains very high concentration after the dispersing process, it is recommended to keep suitable amount of ventilation in the consequent process.     

高层建筑火灾燃烧规律的数值模拟研究

以高层建筑为研究对象,应用FDS软件对高层建筑火灾燃烧规律进行数值模拟计算。得到了高层建筑火灾时的烟气运动、着火点附近区域温度变化、CO2浓度变化等火灾参数变化规律的数值模拟计算结果,计算结果显示,在高层建筑火灾发生300s左右时,火灾开始进入轰燃阶段,火场烟气迅速蔓延,火场温度迅速上升到600℃。数值模拟计算结果为高层建筑性能化防火设计及人员紧急疏散提供一定的理论依据。     

锥形量热法研究木质素基膨胀型阻燃剂对环氧树脂阻燃抑烟效果

以来自自然界储量第二的木质素作为膨胀型阻燃剂的基体,通过接枝氮、磷元素成功合成碳源、酸源、气源三位一体的木质素基膨胀型阻燃剂（Lig-T）,实现了良好的阻燃性能。将Lig?T按照不同含量添加到环氧树脂（EP）中制备EP/Lig-T复合材料,以锥形量热测试考察复合材料的热稳定性能和阻燃性能,并重点考察复合材料在接近真实火灾事故时的热释放和烟释放规律。结果表明,当Lig-T含量为20 %（质量分数,下同）时,复合材料的热释放速率峰值为1 374 kW/m²、热释放总量为41.63 MJ/m²、烟释放总量为1 634 m²/m²,与EP参比试样的数值相比,均呈现下降的趋势,燃烧结束的残炭率从4.26 %增至10.01 %。基于气相和凝聚相的协效阻燃机理,木质素作为膨胀型阻燃剂的碳源使得复合材料在高温条件下具备更好的成炭效果,在燃烧过程中形成稳定且致密的炭层结构,在实现高效阻燃的同时减少有毒烟气的释放,降低火灾的危害。     

基于颜色与运动特征的变电站实时烟雾检测

变电站的环境特殊性要求站内火灾检测准确、快速,烟雾检测是早期火灾检测中的常用方法.针对变电站特殊环境,提出一种快速烟雾检测算法.首先,利用改进的ViBe算法检测变电站实时监控视频中的运动区域,再通过颜色特征确定疑似烟雾区域,然后结合烟雾面积增长特性和移动特性识别烟雾区域.实验结果表明:这种算法能够检测到不同场景中的烟雾,对远距离、稀薄烟雾也能有效检测,满足变电站火灾检测实时性、准确性要求,能够应用于变电站场景烟雾检测.     

基于视频的烟焰检测方法研究

烟焰作为火灾最为明显的标志,对于其进行实时检测,在火灾监控方面有着重要的意义.传统的离子式、光电式、吸烟式等烟雾探测器受到空间和时间的限制,不能进行实时全面的探测[1].本文提出了一种基于视频的烟焰检测算法,采用帧差法提取出差值图像,经过图像处理,获得了理想的烟焰轮廓,根据轮廓大小来确定被监测区域是否有火灾的发生.实验证明,该方法实时准确、灵敏度高、抗干扰能力强、适用范围广,具有广阔的应用前景.     

Up in smoke: I. Smoke-derived germination cues for postfire annual,Nicotiana attenuata torr. Ex. Watson

Some postfire annuals with dormant seeds use heat or chemical cues from charred wood to synchronize their germination with the postfire environment. We report that wood smoke and polar extracts of wood smoke, but not the ash of burned wood, contain potent cue(s) that stimulate germination in the postfire annual plant,Nicotiana attenuata. We examined the responses of seeds from six populations of plants from southwest Utah to extracts of smoke and found the proportion of viable seeds that germinated in the presence of smoke cues to vary between populations but to be consistent between generations. With the most dormant genotypes, we examine three mechanisms by which smoke-derived chemical cues may stimulate germination (chemical scarification of the seed coat and nutritive- and signal-mediated stimulation of germination) and report that the response is consistent with the signal-mediated mechanism. The germination cue(s) found in smoke are produced by the burning of hay, hardwood branches, leaves, and, to a lesser degree, cellulose. Moreover, the cues are found in the common food condiment, liquid smoke, and we find no significant differences between brands. With a bioassay-driven fractionation of liquid smoke, we identified 71 compounds in active fractions by GC-MS and AA spectrometry. However, when these compounds were tested in pure form or in combinations that mimicked the composition of active fractions over a range of concentrations, they failed to stimulate germination to the same degree that smoke fractions did. Moreover, enzymatic oxidation of some of these compounds also failed to stimulate germination. In addition, we tested 43 additional compounds also reported from smoke, 85 compounds that were structurally similar to those reported from smoke and 34 compounds reported to influence germination in other species. Of the 233 compounds tested, 16 proved to inhibit germination at the concentrations tested, and none reproduced the activity of wood smoke. By thermally desorbing smoke produced by cellulose combustions that was trapped on Chromosorb 101, we demonstrate that the cue is desorbed between 125 and 150°C. We estimate that the germination cues are active at concentrations of less than 1 pg/seed and, due to their chromatographic behavior, infer that a number of different chemical structures are active. In separate experiments, we demonstrate that cues remain active for at least 53 days in soil under greenhouse conditions and that the application of aqucous extracts of smoke to soil containing seeds results in dramatic increases in germination of artificial seed banks. Hence, although the chemical nature of the germination cue remains elusive, the stability of the germination cues, their water-solubility, and their activity in low concentrations suggest that these cues could serve as powerful tools for the examination of dormant seed banks and the selective factors thought to be important in the evolution of postfire plant communities.     

FSNet: 基于频率特性的烟雾图像分割网络

烟雾图像分割是对烟雾进行识别与精准定位的基础, 是火灾预警的重要手段. 针对烟雾分割时存在过分割、欠分割以及边界拟合粗糙的问题, 本文提出一种基于频率分离特性的烟雾图像分割网络. 所提出的频率分离模块将特征图中的烟雾区域分离为低频主体部分和高频边界部分, 同时基于多任务学习设计多模块权重自适应损失函数对烟雾整体、主体、边界分别监督学习, 起到细化烟雾边界和改善烟雾整体分割结果的作用; 此外, 结合可变形卷积提出改进的空洞空间金字塔池化模块以解决其信息利用率低和特征关联性差的问题. 在对比实验中, FSNet的烟雾交并比为76.55%, 比基线网络提高了4.25%. 可视化分割结果可以看出, FSNet能有效缓解过分割、欠分割, 所得烟雾边界更平滑, 烟雾图像分割的整体性能获得较大提升.