基于融合边缘优化的煤矿图像语义分割方法北大核心

由于煤矿井下环境恶劣,使得煤矿井下视频获取的图像严重降质,而现有的基于深度学习的语义分割模型在图片清晰化后存在边缘分割模糊的问题,提出了一种采用融合边缘优化模块处理边界信息并运用门控卷积层连接传统特征提取模块并行处理信息的方法;为监督学习轮廓信息,采用二元交叉熵损失函数提高学习效果,并与常规分支的损失函数共同优化模型分割效果。试验结果表明:对已完成清晰化的煤矿井下图像进行语义分割任务时,基于融合边缘优化模块的方法与其他方法相比整体语义分割精度得到提升并且边缘分割精度更高。     

基于多尺度卷积神经网络和LBP算法的浮选工况识别

针对泡沫浮选加药状态检测困难、识别效率低和主观性强等问题,提出了一种结合多尺度卷积神经网络(CNN)特征及改进局部二值模式(LBP)计算方法的核随机权神经网络(K-RVFLNs)浮选工况识别方法。首先,对泡沫浮选图像进行非下采样Shearlet多尺度分解,将原始图像分解为不同频率尺度,设计多通道CNN网络对多尺度图像进行特征提取;再通过改进LBP算法提取特征作为补充,将CNN提取的图像特征与LBP特征进行融合;最后,通过核随机权神经网络映射到更高维空间进行分类决策,实现浮选加药状态的精确识别。实验结果表明,采用多尺度CNN及LBP-TOP特征融合的方法识别的精度比传统LBP算法提高了5.34%,比采用单CNN特征的方法提高了3.76%,结合K-RVFLNs实现浮选工况分类准确率高达96.38%,识别精度和稳定性较现有方法有较大提升,且减少了人工干预,有利于提高生产效率。     

基于长程时空特征与多尺度外观特征的锌精选工况识别北大核心

锌精选作为锌浮选的最后一道流程,其工况直接决定锌浮选最终产品质量。现有基于卷积网络的浮选工况识别方法具备挖掘隐藏特征的能力,取得了良好效果,但仍存在表征能力有限、模型参数大等问题。为此,提出了基于长程时空特征与外观特征的锌精选工况识别模型。首先,提出基于分离三维卷积网络(Separable 3D Convolutional Neural Network,S3D CNN)与注意力机制的泡沫视频相邻帧间短程时空特征提取方法,获得特征聚焦的泡沫视频相邻帧间短程时序信息。然后,在短程时空特征的基础上采用双向卷积长短时记忆网络(Bi-directional Convolutional Long Short-Term Memory,BiConvLSTM)提取泡沫视频帧间的长程时空特征,获取泡沫视频帧间的长程动态时序信息。最后,采用基于残差网络和迁移学习的二维卷积网络提取泡沫图像的多尺度外观特征,并融合长程时空特征,对锌精选工况进行识别。实验结果表明,与现有卷积网络方法相比,所提模型在工况识别精度和模型参数上性能更佳。     

基于递归门控卷积和上下文注意力的煤块检测算法

针对煤矿带式输送机上煤块检测因光照不均存在的漏检与误检问题，提出一种基于门控卷积和上下文注意力机制的改进YOLOv5煤块检测算法。首先，将主干网络中残差模块替换为递归门控卷积模块，通过逐步融合特征信息，提取高阶语义特征，增强模型的特征提取能力。其次，在特征融合结构中加入GCA注意力机制，将全局上下文信息融入坐标注意力模块中，加强感兴趣区域的全局表示，增强多尺度特征融合能力，提高模型对煤块边缘特征的敏感度。最后，采用SIoU损失函数，加速网络模型的收敛。试验结果表明，改进的算法在自建煤块数据集上平均精度均值达到92.8%,召回率达到85.9%,检测速度达到38帧/s。既提高了检测精度，又满足了检测的实时性。     

边缘融合的多字典超分辨率图像重建算法

由于煤矿井下特殊的环境导致井下图像一般比较模糊和边缘特征比较弱,而目前基于字典学习的图像超分辨率重建算法通常是将全部图像块利用训练的单一字典进行重建,忽视了各图像块之间的差异性,不利于重建边缘不清晰的矿井图像。结合矿井图像特征提出一种边缘融合的多字典超分辨率图像重建方法,该算法根据各图像块的梯度统计信息将图像块进行分类并训练对应的字典库,重建时将不同字典重建的图像块融合成完整的高分辨率图像;此外为了提高图像的边缘信息,预处理阶段低分辨率图像进行边缘融合以增强边缘特征,重建的高分辨率图像利用学习的先验知识进行边缘融合以修正重建过程中出现的误差。实验表明,该算法的重建效果优于其它基于字典学习的超分辨率图像重建方法,能够很好地重建图像的边缘细节,并抑制重建过程中产生的重影和振铃效应,平均PSNR值提高1.19 d B。     

一种基于多尺度特征复用残差网络的矿山图像重建算法

针对矿山图像重建中细节损失导致重建质量低下的问题,提出了一种基于多尺度特征复用残差网络的矿山图像重建算法,旨在提高矿山场景下图像重建的精度和效率。首先,设计了一个多尺度特征提取模块,通过堆叠多个并行的卷积层和池化层,结合局部残差网络构建图像特征提取模块,通过不同尺度的多路组合网络,从输入图像中充分提取图像的多尺度细节特征。这些特征表示具有不同的语义信息和空间分辨率,能够捕捉到图像中的不同细节和纹理结构。然后,引入了特征复用模块,将不同尺度的特征进行融合和复用,以增强图像重建的准确性。通过多尺度的特征交互和信息传递,可以有效地利用全局和局部的上下文信息,提高图像的重建性能。通过在自建的矿山图像重建数据集上进行试验,结果表明:所提出的算法在重建精度和效率方面均得到了显著提升,与其他深度学习模型相比,该算法在重建图像的细节保留和结构准确性方面表现出更好的性能。此外,该算法具有较快的训练和推断速度,适用于实际应用场景。     

一种改进分水岭算法的浮选泡沫图像分割方法

为了解决煤泥浮选泡沫图像分割中传统分水岭算法的过分割问题,提出了一种基于自适应标记提取的改进分水岭算法。该方法首先对浮选泡沫图像进行高斯滤波,再运用基于形态学的扩展最大值技术从泡沫图像中自适应提取标记,利用标记对梯度图像进行修改,最后使用分水岭算法对修正后的梯度图像进行分割。试验结果表明,改进后的算法克服了标记提取需要先验知识、分割过程繁琐等问题,使参数选取更加合理,分割结果更加准确。     

煤泥浮选泡沫数字图像处理中的边缘检测算法研究

通过分析煤泥浮选泡沫图像的特点和应用常规边缘检测算法所存在的问题, 提出了综合应用煤泥浮选泡沫图像灰度和梯度信息进行煤泥浮选状态监控的观点, 介绍了使用基于Sobel算子改进的方向算子对煤泥浮选泡沫图像进行处理的方法。     

加快煤矿智能化建设推进煤炭行业高质量发展

引入图像语义分割技术,对矿井次光照环境中的目标物进行分割,将图像分成原始清晰图像和次光照图像两类,采用基于深度学习的图像增强方法对次光照条件下拍摄的图像增强细节后替换,再利用单应变换算法对数据集进行扩充,进而构建矿井巷道图像语义分割标准数据集。提出一种基于自注意力机制的轻量级编码—解码结构网络：以DeepLab V3+编码—解码网络为基础网络,在编码结构中,提取矿井图像深、浅层语义特征信息,将深层语义特征信息经由轻量级自注意力机制模块进行特征激活,而浅层语义特征信息直接送入解码器中,在解码结构中拼接深、浅层语义特征信息,恢复原始图像尺寸,输出分割结果。与传统算法就图像预测进行对比实验,结果表明：在网络复杂度方面,对于3通道512×512像素大小的图像,算法的网络理论计算量FLOPs仅48.80 G,参数量仅11.90 M;在网络分割精度方面,平均交并比76.50%,平均像素精度87.75%,领先其他主流网络;在速度方面,推理一张图像的速度能达到0.032 s,可满足轻量级网络的要求。

     

复杂混合浮选泡沫图像分割方法研究

针对浮选泡沫图像难以直接识别的问题,提出了一种基于形态学的分水岭改进算法,利用CLAHE、最小最大值滤波对泡沫图像进行预处理,采用形态学方法进行图像重建,最后利用分水岭算法对重建后的图像进行分割。仿真结果表明,该算法可有效地对复杂混合泡沫图像进行分割。     

The effect of mothers on bubble size distributions in flotation pulp phases and surface froths

Froth dynamics, i.e. the stability and mobility of the froth, are crucial indicators of various important features of flotation systems. For example, it is desirable that the froth collapses as soon as possible after it is skimmed off the cell in order to curtail losses in throughput. On the other hand, if the froth is too prone to collapse, it will not be sufficiently stable to support its load prior to skimming. Likewise, the mobility of the froth gives similar information on the performance of the flotation cell. For example, a sharp contrast can be observed between dry viscous (immobile) froths and watery runny froths with high mobility. Several authors have recently shown that analysis of the structure of the froth in a flotation cell can be used to assess the performance of the cell. This implies that there is a close relationship between the bubble size distribution in the pulp and froth phases.Until very recently, it was not possible to verify this hypothesis direct, since reliable measurements of bubble size distributions in especially the froth phase could not be obtained. With recent improvements in the machine vision technology originally developed at the University of Stellenbosch, it is now possible to measure bubble size distributions and stability in froth structure with a high degree of accuracy. Unlike previous methods, these improved algorithms can provide a detailed map of flow patterns in the froths, which can give a significantly better idea of operating conditions in the flotation cell. Consequently, in this experimental study the bubble size distribution in the pulp phase of a laboratory flotation cell was measurement with a capillary tube system (UCT bubble size analyser), while the bubble size distribution in the froth phase was measured by use of digital image analysis. The relationship between these bubble size distributions in the pulp and froth phases is discussed.     

基于改进的SuperGlue模型的浮选泡沫稳定度检测方法研究

在浮选过程中,浮选泡沫保持一定的稳定度对保证浮选指标的稳定至关重要。由于浮选泡沫本身的复杂性和现有检测方法的局限性,目前还无法对工业现场泡沫的稳定度进行定量检测和评估。为此,开发了一种基于改进的SuperGlue模型的浮选泡沫图像特征匹配算法,用于对浮选泡沫的稳定度进行测量。该算法采用改进的SuperPoint模型网络对泡沫图像进行特征点提取,将原有的VGG网络进行了改进,利用特征匹配模型对所得到的两组特征点进行匹配,再进行误匹配精筛选,设置置信度阈值进一步提升匹配精度。对比了在匹配算法中应用比较多的GMS算法,本文算法的有效特征点匹配对数提升了19.58%,匹配精度达99.85%。与传统灰度差值法对比,本文的泡沫稳定度测量方法对不同状态的泡沫可辨识性提升明显,极大地提高了图像检测灵敏度,可以满足生产对泡沫稳定度测量的要求。     

Evaluation of the selective detachment process in flotation froth

The improved selectivity between particles of varying degrees of hydrophobicity in flotation froths has been well documented in literature, especially in the deep froths utilized in flotation columns. The phenomenon is believed to be due to the selective detachment process whereby the least hydrophobic particles are released from the bubble surface upon bubble coalescence. To quantify the selective detachment process, column flotation experiments were performed under various operating conditions that provided varying amounts of reflux between the froth and collection zones. Entrainment was eliminated by the use of relatively coarse 250 × 75 micron material. The flotation column incorporated the ability to provide instantaneous stoppage of the process streams and separation between the collection and froth zones after ensuring steady-state operation of the column. The samples collected from the two zones and process streams were evaluated to quantify the flotation rate distribution of the particles comprising each sample. The flotation rate was used as an indicator of the degree of hydrophobicity and thus a relative measure of the binding force between the particle and bubble in the froth zone. The flotation rate data was used as input into well known flotation models to obtain the froth zone recovery rate and the quantity of material that refluxes between the collection and froth zones.     

煤泥浮选泡沫图像灰度行程及其统计纹理特征

针对煤泥浮选泡沫灰度图像统计纹理特征问题,通过实验室浮选柱试验,采集了大量的煤泥浮选泡沫图像,分析了泡沫图像的类别及特征,提出了描述浮选泡沫纹理特征的灰度行程矩阵提取算法,并进一步提取灰度行程矩阵的行程因子特征参数来描述浮选泡沫的视觉纹理特征,分析了各特征参数随浮选时间的变化关系.研究表明,泡沫灰度行程因子特征参数能够表征浮选泡沫图像纹理特征,并与特定的泡沫状态相关,可为煤泥浮选视觉监控系统提供泡沫状态信息.     

The effect of reagent addition strategy on the performance of coal flotation

The recovery and grade of concentrates obtained by flotation is partly dependent upon froth structure. The flotation froth structure can be modified, and hence the flotation performance modified, by the addition of reagents.A direct control system was developed which permitted pre-determined addition profiles of reagent and water to be added to a cell. A series of experiments were performed using different strategies of reagent (Triton x-405) addition for the flotation of Bickershaw coal.A transputer enhanced Micro-Vax Computer was employed to grab images of flotation froths and to process them quantitatively in a parallel procedure. Using image processing techniques characterization of froth structures was achieved. The image processing procedures that were intended to run on-line, were examined and developed using video tapes of experimental runs, off-line.An essential stage in the development of such a system is the understanding of the adsorption behaviour of the reagent in terms of the kinetics, the distribution of reagent between the solid and solution phases, and the resulting effect on the froth structure. A kinetic model was developed to describe the adsorption of the reagent on the particles. The experimental recoveries and froth structure were investigated in terms of reagent distribution and the froth structure.     

Effect of particle size on flotation performance of complex sulphide ores

Flotation processes occurring in the bulk and froth phases have a characteristic influence on the structural features and dynamics of the flotation froth. It is recognized that the structure and texture of a mineral froth is a good indicator of flotation separation performance. The surface froth feature and dynamics are presented by three features extracted from the digitized images of the froths, i.e. SNE, a rough indication of the average bubble size of the froth, froth stability and the average grey level of the froth, an indication of mineral loading. Particle size is an important parameter in flotation operation. Nowadays, particle size is often measured and controlled in flotation concentrators. In this study the dependence of the froth structures on the particle size variation was investigated on the batch flotation of a sulfide ore from the Merensky reef in South Africa, and the size by size recovery curves were studied as well. In general medium particles produced bubbles smaller than those observed in the presence of fine and coarse particles, and the recovery rates were larger. Entrainment was a contributory mechanism to the recovery of fine particles. The fluctuation of flotation indices on the particle size change can be diagnosed and predicted by the froth structures change with a high degree of accuracy.     

Effect of ultrasound on separation selectivity and efficiency of flotation

The effect of the use of ultrasound in the froth phase on the flotation performance has been investigated in relation to the flotation rate of a complex sulphide ore. A series of kinetic flotation tests with and without ultrasound were conducted in a flotation machine with a 2-L cell, in which an ultrasonic probe is located in the froth zone. The results indicate that there is a considerable effect of ultrasound on separation selectivity and efficiency in the flotation of a complex sulphide ore at intermediate and high level airflow rates whereas, no significant differences in the separation performance were obtained from the flotation with and without ultrasound at low airflow rates. In addition, the results of the size-by-size analysis show that a much better cleaning action in the froth was promoted for coarse particles rather than fine particles as a result of the use of ultrasound. As a result of increase in the bubble coalescence, it was found that the use of ultrasound in the froth is more effective at shallow froths. Therefore, either effective pulp volume can be increased with a negligible loss of flotation performance or the pulp density can be decreased to obtain better product quality with the use of ultrasound in shallow froths.