基于浮选泡沫图像预测精矿品位的研究进展  

Research Progress in Predicting Concentrate Grade Based on Flotation Foam Images

作　　者：卜显忠[1] 杨怡琳 宛鹤[1] BU Xianzhong;YANG Yilin;WAN He(School of Resource Engineering,Xi′an University of Architecture and Technology,Xi′an 710055,China)

机构地区：[1]西安建筑科技大学资源工程学院,陕西西安710055

出　　处：《金属矿山》2024年第2期25-38,共14页Metal Mine

基　　金：国家自然科学基金项目(编号:52074206,52374278,52274271)。

摘　　要：随着人工智能技术在矿业生产的广泛应用,利用计算机视觉技术提高精矿品位预测的准确性和效率已成为必然趋势。在综述了传统图像处理算法和深度学习算法在精矿品位预测中的应用与发展历程基础上,并探讨了未来的发展趋势和挑战。传统图像处理技术通过提取泡沫图像的尺寸、颜色、纹理和流速等特征,结合分水岭分割、颜色矩、灰度共生矩阵和局部点特征匹配等算法进行特征提取。这些特征在计算资源有限的场景中具有一定的应用价值,但在应对精矿品位预测任务时精度较低。深度学习技术通过构建合适的模型架构并利用大量数据进行训练,能够提取高层语义特征,具有较高的预测精度,与图形处理单元(GPU)等高效运算设备配合使用,可实现高性能和高效率的统一。介绍了支持向量机(SVM)、极限学习机(ELM)等机器学习算法以及多层感知器(MLP)、全连接层和多尺度特征融合等深度学习算法在特征映射和品位预测中的应用,以及深度学习模型的发展历程。最后综述了工业界视觉检测系统的应用现状,并从数据驱动模型、多模态数据融合、算法实时性和数据集规模等方面分析了该领域所面临的挑战和未来发展趋势。With the widespread application of artificial intelligence in mining production,it has become an inevitable trend to improve the accuracy and efficiency of ore grade prediction using computer vision technology.This article reviews the application and development process of traditional image processing algorithms and deep learning algorithms in ore grade pre-diction,and discusses future trends and challenges.Traditional image processing techniques extract features such as size,color,texture,and flow rate of foam images,and combine algorithms such as watershed segmentation,color moments,gray-level co-oc-currence matrix,and local feature matching for feature extraction.These features have certain application value in scenarios with limited computing resources but have lower accuracy in tackling ore grade prediction tasks.Deep learning techniques,on the other hand,can extract high-level semantic features by constructing suitable model architectures and training them with large amounts of data.They have higher predictive accuracy and can achieve high performance and efficiency when used in conjunction with efficient computing devices such as Graphics Processing Units(GPUs).This article also introduces machine learning algorithms such as Support Vector Machines(SVM),Extreme Learning Machines(ELM),as well as deep learning al-gorithms such as Multilayer Perceptron(MLP),fully connected layers,and multi-scale feature fusion for feature mapping and grade prediction.The development history of deep learning models is also discussed.Finally,the current application status of industrial visual inspection systems is reviewed,and the challenges and future development trends in this field are analyzed from aspects such as data-driven models,multimodal data fusion,algorithm real-time performance,and dataset scale.

关 键 词：精矿品位预测 浮选泡沫 图像处理 计算机视觉 深度学习 

分 类 号：TD923[矿业工程—选矿]