针对目前基于深度学习的脑肿瘤分类算法参数多、计算复杂的问题,提出了一种基于改进MobileViT的轻量级脑肿瘤图像分类模型。首先,在轻量化模型MobileViT中加入卷积块注意力模块(CBAM)以有效地增强局部和全局特征。其次,采用迁移学习方...针对目前基于深度学习的脑肿瘤分类算法参数多、计算复杂的问题,提出了一种基于改进MobileViT的轻量级脑肿瘤图像分类模型。首先,在轻量化模型MobileViT中加入卷积块注意力模块(CBAM)以有效地增强局部和全局特征。其次,采用迁移学习方法加快网络模型在脑肿瘤图像上的学习速度,并在训练过程中使用余弦退火算法来优化所提出的轻量化模型,使得模型更好地收敛。最后,在真实脑肿瘤数据集上对本文模型的有效性进行评估,并与现有的最新基线模型(ResNet、DenseNet121、ShuffleNet、EfficientNet、MobileNet和MobileViT)进行比较。实验结果表明,相比于基线模型,本文所提出的模型不仅显著提高了脑肿瘤图像分类的准确性,而且计算复杂度较低,符合在边缘计算中部署深度学习模型的需求。To address the challenges presented by current deep learning-based brain tumor classification algorithms, which involve numerous parameters and complex computations, we propose a lightweight brain tumor image classification model based on an enhanced version of MobileViT. Firstly, a convolutional block attention module (CBAM) is added to the lightweight model MobileViT to effectively enhance the local and global feature maps. Secondly, a transfer learning approach is used to accelerate the learning speed of the network model on brain tumor images. Additionally, we employ the cosine annealing algorithm to optimize the training process of our proposed lightweight model, facilitating better convergence. Finally, we evaluate the effectiveness of our proposed model on a real brain tumor dataset, comparing it with several state-of-the-art baselines including ResNet, DenseNet121, ShuffleNet, EfficientNet, MobileNet, and MobileViT. The experimental results show that compared to the baseline model, the proposed model in this paper not only significantly improves the accuracy of brain tumor image classification, but also has a lower computational complexity, which meets the requirements of deploying deep learning models in edge computing.展开更多
针对传统大球盖菇人工分级劳动强度大、效率低、一致性差等问题,提出基于MobileViT模型的改进方法。通过设计自适应分支的多尺度模块、增加局部与全局特征融合、引入双重注意力模块等,提高特征提取能力,增强模型鲁棒性。实验结果表明,...针对传统大球盖菇人工分级劳动强度大、效率低、一致性差等问题,提出基于MobileViT模型的改进方法。通过设计自适应分支的多尺度模块、增加局部与全局特征融合、引入双重注意力模块等,提高特征提取能力,增强模型鲁棒性。实验结果表明,改进后的XCA-MobileViT对实验平台上5个级别的大球盖菇数据集平均识别准确率达97.71%,相较于Mobile Vi T模型准确率提高2.34%,参数量和计算量分别下降0.401M和140.2 M。通过对两个菌菇公开数据集进行的验证实验发现,XCA-MobileViT的F1分及准确率超越对比的其他模型,具有良好的泛化性。展开更多
文摘针对目前基于深度学习的脑肿瘤分类算法参数多、计算复杂的问题,提出了一种基于改进MobileViT的轻量级脑肿瘤图像分类模型。首先,在轻量化模型MobileViT中加入卷积块注意力模块(CBAM)以有效地增强局部和全局特征。其次,采用迁移学习方法加快网络模型在脑肿瘤图像上的学习速度,并在训练过程中使用余弦退火算法来优化所提出的轻量化模型,使得模型更好地收敛。最后,在真实脑肿瘤数据集上对本文模型的有效性进行评估,并与现有的最新基线模型(ResNet、DenseNet121、ShuffleNet、EfficientNet、MobileNet和MobileViT)进行比较。实验结果表明,相比于基线模型,本文所提出的模型不仅显著提高了脑肿瘤图像分类的准确性,而且计算复杂度较低,符合在边缘计算中部署深度学习模型的需求。To address the challenges presented by current deep learning-based brain tumor classification algorithms, which involve numerous parameters and complex computations, we propose a lightweight brain tumor image classification model based on an enhanced version of MobileViT. Firstly, a convolutional block attention module (CBAM) is added to the lightweight model MobileViT to effectively enhance the local and global feature maps. Secondly, a transfer learning approach is used to accelerate the learning speed of the network model on brain tumor images. Additionally, we employ the cosine annealing algorithm to optimize the training process of our proposed lightweight model, facilitating better convergence. Finally, we evaluate the effectiveness of our proposed model on a real brain tumor dataset, comparing it with several state-of-the-art baselines including ResNet, DenseNet121, ShuffleNet, EfficientNet, MobileNet, and MobileViT. The experimental results show that compared to the baseline model, the proposed model in this paper not only significantly improves the accuracy of brain tumor image classification, but also has a lower computational complexity, which meets the requirements of deploying deep learning models in edge computing.
文摘针对传统大球盖菇人工分级劳动强度大、效率低、一致性差等问题,提出基于MobileViT模型的改进方法。通过设计自适应分支的多尺度模块、增加局部与全局特征融合、引入双重注意力模块等,提高特征提取能力,增强模型鲁棒性。实验结果表明,改进后的XCA-MobileViT对实验平台上5个级别的大球盖菇数据集平均识别准确率达97.71%,相较于Mobile Vi T模型准确率提高2.34%,参数量和计算量分别下降0.401M和140.2 M。通过对两个菌菇公开数据集进行的验证实验发现,XCA-MobileViT的F1分及准确率超越对比的其他模型,具有良好的泛化性。
