Artificial Intelligent Identification Of Apatite-fission Tracks Based On Machine Learning

In the past half-century,the fission track dating method has been widely used in the study of earth science.By using apatite fission track dating technology,the age and length of the fission track of samples can be obtained,and a reasonable annealing model can be selected to establish a time-temperature function,which can simulate the thermal evolution history of rock samples and restore the stripping and burial process of samples.However,the traditional fission track dating method needs to observe and count with the naked eye under a microscope,which needs to eliminate impurities such as scratches to accurately identify fission tracks.The whole counting process is not only time-consuming and laborious but also prone to counting errors.Therefore,it is urgent to overcome the shortcomings of traditional fission track identification and counting methods and develop efficient and accurate track fission identification methods.In this study,the intelligent identification of apatite fission track is taken as the research object,and two machine learning methods-Tensorflow Object Detection API based on Faster R-CNN model and Open CV cascade classifier based on LBP feature are used as the identification methods of apatite fission track,and the apatite fission track sample and Durango standard sample photographed by external detector method under the optical microscope are used as the training data and verification data of the two methods.In the method of using the Open CV cascade classifier,the method of preprocessing data samples is to divide the experimental data into positive samples containing only apatite fission tracks and negative samples without apatite fission tracks.The experimental environment of the Open CV cascade classifier method based on LBP features is built,and positive and negative samples are input into the built model for training.In the method of using Tensorflow Object Detection API,the method of preprocessing data samples is to classify training samples into transparent and opaque labels according to the transparency of apatite fission tracks,and label apatite fission tracks with Labelimg software.The experimental environment of Tensorflow Object Detection API based on Faster R-CNN is built,and the model is trained after the construction.We use three indexes,Precision,Recall,and F1-Score,to measure the recognition effect of the two experimental methods on apatite fission tracks.The average Precision,Recall,and F1-Score results of the Open CV cascade classifier based on the LBP feature are 75.9%,76.4%,and 75.7% respectively,and the average Precision,Recall,and F1-Score results of Tensorflow Object Detection API based on Faster R-CNN model are94.4%,84.9%,and 88.8% respectively.The research results show that the three indexes of Tensorflow Object Detection API are higher than the cascade classifier using Open CV.We think that the method of Tensorflow Object Detection API based on the Faster R-CNN model is better than that of the Open CV cascade classifier based on LBP features in this experiment.In addition,the Precision,Recall,and F1-Score of the two methods for apatite fission track identification are all higher than 75%,which shows that the two methods have high accuracy for apatite fission track identification,thus proving the reliability of machine learning method in apatite fission track target detection research.