离子吸附型稀土开采的高分影像识别尺度选择研究

Analysis of Scale Effect of High Resolution Image Recognition in ion Adsorption Rare Earth Mining

离子吸附型稀土矿矿区分散,矿点众多,采用高空间分辨率遥感影像进行开采监测是一种有效手段,尽管较高的空间分辨率能改善识别质量,但也会带来识别效率和费用问题,合适的识别尺度选择显得尤为重要。以赣南岭北稀土矿区Pleiades-1A高分遥感影像为数据源,结合稀土矿的开采过程,通过沉淀池构建矿点尺度和矿区尺度识别模型,探究稀土矿开采识别的尺度效应。实验以0.5 m为步长,通过重采样获取0.5,1,1.5,2,2.5,3 m分辨率影像,再融合高分辨率影像纹理特征,利用面向对象Cart决策树和模糊分类相结合的方法获取矿点及矿区尺度沉淀池识别情况。结果表明:在矿区尺度下,多分辨率沉淀池识别精度为93.24%,87.83%,78.37%,75.68%,63.51%,62.16%,随着分辨率降低,识别效率显著提高,但识别精度有一定降低。矿点尺度下,通过对1,1.5,2,2.5,3,3.5 m稀土矿点分布情况分析,得出矿点识别精度为96.55%,96.55%,96.55%,93.10%,79.31%,65.51%,在高于2 m分辨率尺度上进行大范围监测矿点识别精度保持不变,低于2 m后识别精度开始降低。说明在大范围监测情况下,分辨率2 m可以保证矿点监测的正确率与监测效率,而在矿区尺度,要对沉淀池进行有效识别,应采用米级以下影像分辨率。

The ion-adsorbing rare earth mining areas are scattered and there are many ore spots.It is an effective means to adopt remote sensing images with high spatial resolution for mining monitoring.Although higher spatial resolution can improve the identification quality,it will also bring problems of identification efficiency and cost.This paper takes Pleiades-1A high-resolution remote sensing image in the northern rare earth mining area of Ganzhou,Jiangxi Provice as the data source,combines with the mining process characteristics of rare earth minerals,constructs the model of ore point scale and mining scale recognition through the characteristic sedimentation tank of rare earth mining,and explores the scale effect of mining recognition of rare earth minerals.In the experiment,0.5,1,1.5,2,2.5 and 3 m resolution images were obtained by sampling with 0.5,1,2,2.5 and 3 m resolution images.Then high-resolution image texture features were fused,and the identification of mining points and scale sedimentation ponds was obtained by using object-oriented Cart decision tree and fuzzy classification method.The results showed that:at the scale of ore point,the identification accuracy of sedimentation tank at each resolution is 93.24%,87.83%,78.37%,75.68%,63.51%,62.16%,respectively.With the reduction of resolution,the identification efficiency is significantly improved,but the identification accuracy is reduced to some extent.Based on the analysis of the distribution of rare earth mining points at the scale of 1,1.5,2,2.5,3 and 3.5 m,it is concluded that the identification accuracy of mining points is 96.55%,96.55%,96.55%,93.10%,79.31%and 65.51%,respectively.In terms of resolution of 2 m,the identification accuracy of mining points in a wide range of monitoring remains unchanged.It indicated that in the case of scale monitoring in the range of mining areas,the resolution 2 m can ensure the accuracy and efficiency of mine point monitoring,while in the case of mine point scale,the image resolution below the meter level should be adopted to effectively identify the sedimentation tank.