基于深度学习的车辆类别检测研究与实践

Research and practice of vehicle category detection based on deep learning

使用基于深度神经网络的VGG模型,能有效的解决传统目标检测需要人工提取,费时费力的缺点,并提高模型的泛化能力。通过研究分析VGG模型,利用Keras框架实现了此模型在迁移学习上的应用。实验通过对不同品牌的车辆训练集训练与调参,测试集的数据准确率提高到近90%。

In traditonal target recognition, manually extracting features from different environments is time-consuming and laborious.The VGG model based on deep neural network can effectively solve these problems and improve the generalization ability of the model.By research on the VGG model, the application of VGG model in transfer learning is realized by using Keras framework. Through the training and parameter adjustment of vehicle training sets of different brands, the data accuracy of the test set is improved to nearly 90%.