Fine-grained image classification method based on multi-modal learning

The invention discloses a multi-modal learning-based fine-grained image classification method, which comprises the following steps of: downloading original pictures of different species and corresponding additional information files from a known data set, preprocessing the additional information files, training a neural network for extracting multi-modal features and fusion features, and converging the neural network, performing label probability prediction on the fine-grained image through the converged neural network, performing decision correction on the prediction probabilities of the two neural networks, and finally outputting the category of the species in the image according to the correction result. 本发明公开了一种基于多模态学习的细粒度图像分类方法，先从已知数据集中下载不同物种的原始图片及对应的附加信息文件，通过对附加信息文件进行预处理后，用于训练提取多模态特征和融合特征的神经网络并收敛，然后通过收敛的神经网络对应细粒度图像进行标签概率预测，再对两个神经网络的预测概率进行决策修正，最后根据修正结果输出图像中物种的类别。