【摘要】 针对神经网络模型在训练过程中遇到的收敛速度慢和测试样本不平衡导致的准确率降低问题，提出了一种基于改进VGG16图像分类模型的LBF-VGG16(Leaky-Bactch-Focal-VGG16)。该模型将原Relu激活函数替换为Leaky Relu,并在卷积层与激活函数之间引入BN层，以优化收敛效果。在训练过程中，采用SGD优化器，并融入Focal Loss损失函数。试验结果表明，LBF-VGG16模型在分类效果和收敛速度方面较改进前均有显著提升更多还原

【Abstract】 Aiming at the image classification issue of the “dying neuron” problem associated with its ReLU activation function and the challenge of imbalanced test samples, a modified version of the VGG 16 model, named LBF-VGG 16(Leaky-Batch-Focal-VGG 16), is proposed. This enhanced model substitutes the traditional ReLU activation function with Leaky ReLU and integrates Batch Normalization layers between the linear and nonlinear components to foster better convergence. During the training phase, the Stochastic Gradient Descent optimizer is employed in conjunction with Focal Loss. Comparative experiments were conducted in two groups: one contrasting VGG 16 with LBF-VGG 16, and the other comparing VGG 13 + Local, VGG 19 + Focal Loss, and LBF-VGG 16. The outcomes demonstrate that the LBF-VGG 16 model outperforms the other models in terms of accuracy and convergence speed.更多还原