Research On Intelligent Separation Technology Of Vanadium-Titanium Ore Based On Deep Learning

Steel is the lifeblood of the industry,and vanadium-titanium ore is one of the main sources of iron ore.Moreover,vanadium is an important stabilizing agent in smelting iron and steel,and the high-quality steel can be obtained by directly smelting vanadium titanium ore.At the same time,although China is rich in mineral resources,the high purity vanadium titanium ore resources have been almost consumed,the remaining resources are difficult to exploit,and the yield is low.Therefore,how to improve the utilization rate of vanadium-titanium ore resources is one of the main problems of vanadium-titanium ore industry in China.In order to obtain iron ore resources,it is usually necessary to select the ore with high purity from many waste rocks.However,at present,this method is mostly manually implemented,which has low efficiency and is greatly affected by subjective factors.Therefore,in this paper,a deep-learning based vanadium titanite detection algorithm is proposed,which can greatly improve the efficiency of vanadium titanite selection,laying a solid foundation for automatic beneficiation.Firstly,a machine vision model is designed in this paper,which includes the selection of camera,imaging method and so on.The high quality image data set of vanadium-titanium ore is obtained by using the model,and then the data set is extended by means of data enhancement,and then the color,texture and grayscale features of vanadium-titanium ore images are extracted.Subsequently,the traditional machine learning and deep learning methods are used to detect vanadium-titanium ore.In traditional machine learning,Local Binary Pattern(LBP),gray-level co-occurrence Matrix(GLCM)and Gabor filter are used to extract texture features,and then Support Vector Machine(SVM)and Fuzzy Support Vector Machine(FSVM)classifiers are trained to complete the beneficiation identification of vanadium-titanium ore,and then Discrete Wavelet Transformation(DWT)is used to optimize the structure of Gabor filter,simplifying its dimensions and improve its recognition speed.At the same time,the color and gray characteristics of vanadium-titanium ore are classified by FSVM classifier.Due to the poor effect,the method of feature fusion is adopted for classification.The average accuracy of the fusion method reaches 85.3%.While in deep learning,VGG19,Inception V3 and Res Net network structures are adopted to detect vanadium-titanium ore images through transfer learning,Vt VGG-3 freezes the first 16 layers of the network and retrains the last three fully connected layers.Vt VGG-all retrains the entire network;Vt Inception V3-f is the model that freezes all the layers before the last two layers and trains only the last two layers,Vt Inception V3 directly retrains the entire network;and Vt Res Net-f freezes the network's input and the first four modules and retrain the last fully connected layer,and Vt Res Net-all retrains the entire network.The results show that Vt Inception V3 has better performance in the identification of vanadium-titanium ore,and its accuracy rate reaches 99.1%,and F1 is 97.25%.Therefore,Vt Inception V3 model is better for completing the task of vanadium-titanium ore image detection.Finally,the network structure is optimized and combined with the SVM classifier,which has a better detection effect on sulfur-bearing ores.The final network recognition is 1/4 faster than the original one,and the accuracy rate is also improved to 99.3%.