基于卷积神经网络的多标签图像分类识别算法研究

Research on Multi-label Image Classification and Recognition Algorithm Based on Convolutional Neural Network

准确获取和判断多标签图像中的信息,需实现该图像的精准分类识别,因此,研究基于卷积神经网络的多标签图像分类识别算法.利用四元数Gabor滤波卷积算法提取该类图像特征,将获取特征向量作为卷积神经网络模型的输入,通过模型的卷积、池化,以及单层感知器的学习和训练识别图像,实现多标签图像的识别;对模型实行双重优化,提升模型收敛速度;同时采用基于实例差异化的多标签图像分类方法,实现多标签图像分类.测试结果表明:对不同复杂程度图像纹理特征提取收敛性能良好,可准确完成多标签图像中的目标识别,Kappa系数均在0.8以上,分类效果良好.

To accurately acquire and judge the information in multi label image,it is necessary to realize the accurate classification and recognition of the image.Therefore,the algorithm of multi label image clas⁃sification and recognition based on convolutional neural network is studied.The feature of the image is ex⁃tracted by the method of Gabor filtering convolution.The acquired feature vector is used as the input of convolution neural network model.The multi label image recognition is realized through the convolution,pooling and learning and training of the single-layer perceptron.The model is optimized by double optimi⁃zation,and the convergence speed of the model is improved.At the same time,the multi label image clas⁃sification method based on case differentiation is adopted to realize the multi label image classification.The test results show that the algorithm has good convergence performance for texture feature extraction of different complexity,and can accurately identify the target in multi label images.Kappa coefficient is above 0.8,and the classification effect is good.