基于电感式传感器阵列成像的废金属识别方法  

Waste Metal Identification Method Based on Inductive Sensor Array Imaging

作　　者：张朝宏 叶文华[1] 周亚楠 ZHANG Chaohong;YE Wenhua;ZHOU Yanan(College of Mechanical&Electrical Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China)

机构地区：[1]南京航空航天大学机电学院,南京210016

出　　处：《有色金属（选矿部分）》2024年第7期71-78,共8页Nonferrous Metals（Mineral Processing Section）

基　　金：江苏省重点研发计划项目(BE2022845)。

摘　　要：在传统物料自动化分选领域,电感式传感器有着抗干扰性强和检测便捷的优势,但因其在识别传送带上随机散布的金属物料时,难以准确提取出每个金属物料对应的检测数据,而很少被独立应用到金属物料在线识别中。为此,提出了一种将电感式传感器阵列检测数据转换为图像的方法,以图像分割完成数据提取,实现对金属物料的在线识别。首先以BP神经网络对检测数据进行插值,改善传感器分布方向上数据稀疏的问题后,通过线性映射得到灰度图像;其次针对成像分辨率低及区域边界模糊的问题,以传统方法对灰度图像进行初步分割后,设计边界蚕食算法,增强对图像山谷处的分割力度,实现图像的完全分割,提取每个金属对应的图像区域,完成数据提取;最后建立模糊C均值聚类算法分类模型,根据每个金属图像区域的灰度最大值和灰度最大梯度值完成金属材质分类,实现种类识别。试验结果表明,相比于其他图像分割方法,边界蚕食分割方法像素保留比例较高,为后续识别步骤保留了更多信息,且最小分割距离不高于2.5 mm,对密集金属群有更强的区分能力;所提出的金属识别方法最终的识别正确率达到了94.6%,对传送带上随机散布的密集金属群具有良好的在线识别能力。In the field of traditional material automatic sorting,inductive sensors have the advantages of strong anti-interference and convenient detection.However,when identifying randomly scattered metals on the conveyor belt,it is difficult to accurately extract the corresponding detection data for each metal,so it is rarely independently applied to online identification of metals.To this end,a method for converting the detection data of inductive sensor array into images is proposed,data extraction is completed by image segmentation to realize online recognition of metals.Firstly,BP neural network is used to interpolate the detection data to improve the problem of sparse data in the sensor distribution direction,a grayscale image is obtained through linear mapping.Secondly,in response to the problems of low imaging resolution and blurred regional boundaries,traditional methods are used for preliminary segmentation of grayscale images.A boundary encroachment algorithm is designed to enhance the segmentation strength at the valley of the image,achieve complete image segmentation,extract the corresponding image area for each metal,and complete data extraction.Finally,the Fuzzy C-Means clustering algorithm classification model is established,and the metal classification is completed according to the maximum gray value and the maximum gray gradient value of each metal image region,so as to realize the type recognition.The experimental results show that,compared to other image segmentation methods,the boundary encroachment segmentation algorithm has a better pixel retention ratio and minimum segmentation distance of no more than 2.5 mm,which preserves more information for subsequent recognition steps,and has stronger discrimination ability against densely scattered metals.And the proposed metal recognition method has a final recognition accuracy of 94.6%,for densely randomly scattered metals on the conveyor belt,which shows good online recognition ability.

关 键 词：电感式传感器阵列 图像分割 数据提取 边界蚕食 金属识别 

分 类 号：TD925[矿业工程—选矿]