基于深度学习的黑钨矿图像识别选矿方法

黑钨矿图像识别是代替黑钨选矿手选抛废的一条高效途径,但存在无法识别黑钨矿石与围岩废石的问题。本文利用深度学习中卷积神经网络进行迁移学习来解决,该方法具有收敛快速、所需数据集小和分类准确的优点。首先,对黑钨原矿彩色图像采用旋转、平移等方法进行数据增广降低样本不平衡性。其次,基于Keras框架使用本文优化的神经网络进行全新训练。结果表明:黑钨矿石与围岩两类识别中Wu-VGG19迁移网络矿石识别率最高,为97.51%。此外,本文加入石英脉石类别继续实验,得出修改的Wu-v3迁移网络矿石识别率最高,为99.6%。

Flotation performance of anisic hydroxamic acid as new collector for tungsten and tin minerals

In order to improve the recovery of tungsten ores containing tin minerals,anisic hydroxamic acid(p-methoxy benzohydroxanic acid,PMOB)was synthesized and introduced as novel collector in the flotation of scheelite,wolframite and cassiterite.The flotation performance and adsorption mechanism were investigated by micro/batch flotation,zeta potential measurements and density functional theory(DFT).The micro flotation results showed that the recoveries of scheelite,wolframite and cassiterite using PMOB as collector are 97.45%,95.77% and 90.08%,respectively,and the corresponding recoveries are 91.00%,84.30% and 84.67% for benzohydroxamic acid(BHA).The batch flotation results revealed that the collector dosage could be reduced by about 45% for PMOB compared with BHA,in the case of similar flotation indicators.Zeta potential measurements indicated that PMOB could be adsorbed on the mineral surfaces by chemisorption.Moreover,density functional theory(DFT)calculation results showed that the substituent group—OCH_(3)endues PMOB stronger electron donation ability and hydrophobicity compared with benzohydroxamic acid(BHA),pmethyl benzohydroxamic acid(PMB)and p-hydroxyl benzohydroxamic acid(PHB).