An Improved Vehicle Fine-Grain Identification Algorithm Based on Stochastic Weight Average

In the field of motor vehicle recognition, use neural network models has become standard, and tuning hyperparameters loss functions been shown to be an effective way improve performance these models. However, when using classical convolutional architectures (e.g., ImageNet) training them on images with random labels, overparameterization problem can lead suboptimal results increased risk recognition failure. P. Ismailova et al. proposed a solution this weight averaging, which resulted in development simple Stochastic Weight Averaging (SWA) optimizer. paper, we apply SWA method optimize original model demonstrate significant improvements accuracy through different learning rate schemes various traditional optimizers. We also identify suitable hyperparameter values enhance model's generalization abilities several experiments, reducing waste resources task improving fine-grained general, thus increasing efficiency related fields.