Method for identifying rock lithology based on lightweight convolutional neural network

Abstract

The rock lithology is identified based on the MobileNets lightweight convolutional neural network, a rock image lithology identification model is established in combination with a transfer learning method, and a method for rapidly and efficiently identifying multiple types of rocks suitable for geological field application is provided. A method for quickly and efficiently identifying multi-class rocks based on a lightweight convolutional neural network in combination with transfer learning is characterized in that a learning result on a large data set ImageNet based on a MobileNet lightweight convolutional neural network in combination with a transfer learning method is migrated to a 28-class rock image data set, and after re-training, the multi-class rocks are quickly and efficiently identified. And screening the compression CNNs model with the best performance and deploying the compression CNNs model to a mobile terminal, and constructing a rock identification model suitable for being used in a geological survey site.

Description

First, technical field [0001] The invention relates to a method of rocky rocky identification, and in particular to a rock rocky identification method of lightweight convolutional neural network binding migration learning. [0002] Second, technical background [0003] Rocky rockness identification and classification is an important area in geological research and an important part of geological investigation work. The traditional identification method requires field geologists to collect fresh rock specimens during the survey, and then returns to the laboratory, from the vertical layer of the rock specimen, the area is approximately 2 cm × 2 cm, which is flat on the grinding machine. On one side, adhesive such as gum is attached to the glass sheet until the other side is flattened to 0.03 mm thick, then bonded with gums with gums; finally, knowledge or experienced geologists observe rock flakes under polarizing microscope Image determines the rock type and structural parameters. ...

AI Technical Summary

Problems solved by technology

At present, heavyweight deep learning models are mainly used to automatically identify rock thin section images. Although this type of method solves the problems of strong subjectivity and high judgment costs, it still needs to make rock samples and obtain rock thin section images under a microscope. Need to improve recognition efficiency and reduce recognition time

However, the model requires more resources during calculation, which is not the best solution for geological survey sites

Images