基于改进差分进化粒子群的煤尘颗粒图像辨识  

Recognition of coal dust image based on improved differential evolution particle swarm optimization

作　　者：王征[1] 潘红光[1] WANG Zheng;PAN Hongguang(College of Electric and Control Engineering,Xi’an University of Science and Technology,Xi’an 710054,China)

机构地区：[1]西安科技大学电控学院

出　　处：《煤炭学报》2020年第2期695-702,共8页Journal of China Coal Society

基　　金：国家自然科学基金资助项目(51804249)

摘　　要：为实现煤尘融合区域内单个颗粒准确分离,进一步明确煤尘特性参数图像分析内在机理,构建改进差分进化粒子群模型(IDE-PSO,全称Improved Differential Evolution Particle Swarm Optimization)对煤尘颗粒群基本形态特征进行研究。指出颗粒分离过程可分为3个阶段:①分析煤尘特性参数变化规律,建立图像特性模型,根据煤尘物理特性参数确定颗粒群重叠域;②推导图像参数和煤尘特性的关系表达式,定位边缘特征点,实现边缘平滑处理并在重叠域边缘特征点内提取重叠颗粒交点计算模型;③确定粒子群各行粒子适应度函数值,更新粒子群位置矩阵,获取粒子群全局最优位置交点,运用改进差分进化粒子群算法实现煤尘重叠颗粒分离。结果表明:通过参数模型建立颗粒群基本形态特征判定规则;通过特征点定位,可剔除大多数干扰点,交点提取计算量明显降低,利于寻找有效交点;理论计算模型加入变异算子可保证粒子群多样性,避免粒子过早收敛。当粒径<20μm时,提出算法模型的识别率为80.85%,随粒径范围增大到粒径<75μm时,提出算法模型的识别率为86.54%,当75μm<粒径<200μm和粒径> 200μm时,提出的模型算法的识别率分别为88.04%和91.16%,与其他识别算法相比,改进算法对煤尘颗粒图像识别精度具有明显优势。在煤尘物理特性分析的基础上,建立图像特性分析模型,推导出其图像参数和煤尘特性的关系表达式,易于编程实现,在矿山实际运用中具有较强的可行性,也为以后煤尘爆炸浓度检测可靠性研究提供了借鉴。To separate the individual particle from the coal dust image particle fusion region accurately and confirm the internal mechanism of image characteristic parameters,an effective model based on improved differential evolution particle swarm optimization is proposed.It includes three steps as follows:firstly,the change rules of coal dust parameters are analyzed and image characteristic model is set up.After that,the particle swarm overlap region is determined.Secondly,the relationship between image parameters and dust characteristics are derived and the intersection model is set up in the edge feature points.Finally,the particle fitness function values are determined and the global optimal position is obtained.The results show that the proposed characteristics rules yield more effective decision.The most interfering points can be eliminated and the computational burden is reduced significantly.Furthermore,the diversity of particle swarms is maintained effectively by adding mutation operator,so particles premature convergence is avoided and overlapping particles are separated effectively.When the particle size is <20 μm,the recognition rate of the proposed algorithm is 80.85%;when the particle size is <75 μm,the recognition rate is 86.54%;when the particle size is between 75 μm,and 200 μm and > 200 μm,the recognition rates are 88.04% and 91.16%,respectively.Compared with other recognition algorithms,the proposed algorithm has obvious improvement on image recognition accuracy.Based on the coal dust physical properties,the image characteristics model is established and is easily implemented by programming.It is feasible in the mine application and provides a new approach to examine the reliability of coal dust explosion concentration.

关 键 词：重叠颗粒 图像特性 边缘特征 支撑区域 全局最优 

分 类 号：TD76[矿业工程—矿井通风与安全]