Experimental Studies on Rock Thin-Section Image Classification by Deep Learning-Based Approaches

Experimental studies were carried out to analyze the impact of optimizers and learning rate on performance deep learning-based algorithms for rock thin-section image classification. A total 2634 images including three types—metamorphic, sedimentary, volcanic rocks—were acquired from an online open-source science data bank. Four CNNs using different optimizer (Adam, SGD, RMSprop) under two learning-rate decay schedules (lambda cosine modes) trained validated. Then, a systematic comparison was conducted based model. Precision, f1-scores, confusion matrix adopted as evaluation indicators. Trials revealed that approaches classification highly effective stable. Meanwhile, experimental results showed mode better option adjustment during training process. In addition, four neural networks confirmed ranked VGG16, GoogLeNet, MobileNetV2, ShuffleNetV2. last step, influence optimization evaluated VGG16 demonstrated capabilities model Adam RMSprop more robust than SGD. The study in this paper provides important practical value high-precision model, which can also be transferred other similar tasks.