N-CNNs算法在图像分类中的实际性能优化

Actual Performance Optimization of N-CNNs Algorithm in Image Classification

传统卷积神经网络模型在嵌入式设备上的图像数据分类任务中,存在存储和计算方面的局限性。为解决这类问题,本文研究卷积神经网络与视觉转换器相结合的N-CNNs模型的图像分类能力。首先采用卷积神经网络来对图像进行处理,然后采用视觉转换器拆解多个独立卷积神经网络子模型,并调整各子模型的超参数和优化器设置。实验的结果表明,在保持准确率的前提下,N-CNNs模型损失度更低,具备更强的泛化能力。

Traditional convolutional neural network models have limitations in storage and computation for image data classification tasks on embedded devices.To solve such problems,this paper proposes an N-CNNs model that combines CNNs with ViT.Firstly,CNNs are used to process the image,and then ViT is used to decompose multiple independent convolutional neural network sub models and adjust the hyperparameters and optimizer settings of each sub model.The experimental results indicate that,while maintaining accuracy,the N-CNNs model has lower loss and stronger generalization ability.