摘要
针对岩石智能识别的研究总体较少,且识别准确率较低的问题,提出一种结合迁移学习和集成学习方法的岩石识别卷积神经网络模型TF_MDR_Fusion,其主要结合迁移学习思想,对MobileNetV3_Large、DenseNet121和ResNet50这3种模型在ImageNet数据集上进行预训练,通过集成学习方法将3种模型结合进一步提高岩石识别的性能。通过对比实验和消融实验,其结果均表明TF_MDR_Fusion模型比单个模型表现更出色,其对于岩石图像识别的准确率为77.60%,可为岩石智能识别提供有力支持。
A convolutional neural network model named TF_MDR_Fusion is proposed for rock identification.The model is combined with transfer learning and ensemble learning methods.It primarily embraces the concept of transfer learning,involving pretraining the MobileNetV3_Large,DenseNet121,and ResNet50 models on the ImageNet dataset.Three models are combined using ensemble learning techniques to further elevate the performance of rock identification.Through comparative experiments and ablation experiments,the results consistently demonstrate that TF_MDR_Fusion model outperforms individual models,with an accuracy of 77.60% in rock image identification.
出处
《工业控制计算机》
2024年第5期83-84,87,共3页
Industrial Control Computer
基金
国家自然科学基金资助项目(62062011)。
关键词
卷积神经网络
迁移学习
集成学习
岩石识别
convolutional neural network
transfer learning
integrated learning
rock identification
