Image Recognition of Ore Specimens Based on Deep Neural Network

Ore and mineral identification is an excellent obstacle for today’s geological informatization popularization work. At the same time, identification plays a vital role in many research areas. Therefore, intelligent identification of ores and minerals is the basis for pushing for the work to popularize geological science. Today’s computer vision technology and deep learning theory made the intelligent identification of ore minerals possible. This paper introduces artificial intelligence to the field of geological science by utilizing a combination of a cloud server, recognition model, convolutional neural network, and deep learning. This integration realizes the “face ID” for minerals and rocks, which can help people in the industry and the public quickly identify minerals and rocks found in various environments. At the same time, it can widely promote geology’s informatization and the work’s popularization.Through the research and analysis of ore identification-related technologies, the author selected the Pytorch framework to optimize the ResNet50 deep learning model to identify ore samples. The main contents of the essay are as follows:(1)Fully studied and researched the related technologies in computer vision technology and ore identification, developed an in-depth understanding of the identification framework, analyzed its shortcomings, and finally chose the Pytorch framework to identify ore samples.(2)Eleven ore samples were selected to make the dataset, including quartz, biotite, bornite, malachite, granite, limestone, and malachite.(3)The image is denoised. Then, combined with the preprocessed images, data augmentation is performed on the ore color image by rotation and translation to reduce the sample imbalance.(4)The ResNet50 model is optimized based on the Pytorch framework, and the optimized model is used to prepare, evaluate, and validate the data set. As a result, the resulting model achieved an impressive recognition accuracy rate of 98%.(5)Combining the model with practice, this paper’s research results were concluded, the optimized contents were proposed, and the focus of future research was suggested.