基于改进的SuperGlue模型的浮选泡沫稳定度检测方法研究

Study on Flotation Froth Stability Detection Method Based on Improved SuperGlue Model

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。

During flotation,it is crucial for the flotation foam to maintain a certain degree of stability to ensure the stability of flotation indexes.Due to the complexity of the flotation froth itself and the limitations of the existing detection methods,it is currently not possible to quantitatively detect and evaluate the stability of the froth at industrial sites.For this reason,a flotation froth image feature matching algorithm has been developed based on the improved SuperGlue model for the measurement of flotation froth stability.The algorithm uses the improved SuperPoint model network to extract feature points from this foam image,the original VGG network is improved,and the feature matching model is utilized to match the two sets of feature points obtained,and then mismatch refinement screening is carried out,and the confidence threshold is set to further improve the matching accuracy.Compared with the GMS algorithm,which is more widely used in matching algorithms,the effective feature point matching pairs of this paper's algorithm are improved by 19.58%,and the matching accuracy reaches 99.85%.Compared with the traditional grayscale difference method,this foam stability measurement method in this paper improves the recognizability of foam in different states significantly,greatly improves the image detection sensitivity,and can meet the requirements of foam stability measurement in production.