一种数据增强和多模型集成的细粒度分类算法

针对解决数据缺少和单个卷积网络模型性能的限制造成细粒度分类准确率不高的问 题，提出了一种数据增强和多模型集成融合的分类算法。首先通过镜像、旋转、多尺度缩放、高 斯噪声、随机剪切和色彩增强6 种变换对CompCars 数据集进行增强处理，然后采用差异化采样 数据集的方法训练CaffeNet、VGG16 和GoogleNet 3 种差异化的网络。然后采用多重集成的方法 集成多种模型的输出结果。实验中测试网络结构在不同数据增强算法和不同模型集成下的分类结 果。模型集成的分类准确率达到94.9%，比最好的单GoogleNet 模型的分类精确率提高了9.2 个 百分点。实验结果表明该算法可以有效地提高分类的准确率。

Data Augmentation with Multi-Model Ensemble for Fine-Grained Category Classification

In order to solve low classification precision caused by the lack of training data or the classification performance constraint of single convolutional network model, a fine-grained category classification algorithm based on data augmentation and multi-model ensemble is proposed. Firstly, the paper designs a variety of data augmentation methods to increase the number of pictures in CompCars dataset, including mirroring, rotation, multiscale scaling, Gaussian noise, random cropping and color enhancement. Then 3 differentiated models, CaffeNet, VGG16 and GoogleNet, are trained using the constructed differentiated dataset by different data sampling. A multi-layer ensemble learning method is used to integrate multi-model’s classification results. The experimental results show the fine-grained classification of the differentiated convolution network trained on the different datasets generated by the different data augmentation method. The experiment also shows the classification results of multi-model ensemble with different ensemble strategy. The final classification precision of multi-model ensemble is 94.9%. Compared with the best single model GoogleNet, the classification precision is increased 9.2%. The results verify the effectives of proposed algorithm.