基于深度学习算法的图像分类方法
|
摘要
图像分类是提高大规模图像检索速度的一项关键技术,当前图像分类方法均存在不同程度的不足,图像分类结果不能满足图像处理的实际应用需求,为了解决当前图像分类方法存在的一些难题,以进一步提高图像分类的效果,提出了基于深度学习算法的图像分类方法。首先对图像分类的国内外研究现状进行综述,指出各种图像分类方法的局限性,然后提取图像分类特征,将图像特征向量作为深度学习算法的输入,图像分类结果作为深度学习算法的输出,通过学习建立图像分类器,最后在Matlab 2016平台与经典图像分类方法的性能进行仿真对照实验。结果表明,提出的图像分类准确性更优,图像分类的平均时间大幅度降低,图像分类整体结果得到了有效改善,较好地克服了当前图像分类方法的不足,具有较高的实际应用价值。
Image classification is a key technology to improve the speed of large-scale image retrieval. The current image classification methods have deficiencies, so they cannot meet the actual application needs of image classification. In order to solve problems existing in current image classification methods, and further improve the effect of image classification, a novel image classification method based on depth learning algorithm is proposed in this paper. Firstly, the research status of image classification at home and abroad is introduced, and the limitations of various image classification algorithms are pointed out. Secondly the features of image classification are extracted while image feature vector machine is used as the input, and image classification is used as the output of depth learning algorithm, and the image classifier is constructed by learning. At last, the performance is tested and compared with classical image classification methods in Matlab 2016 platform. The simulation results show that the image classification accuracy of the proposed method is obviously better, the average time of image classification is greatly reduced, the overall result of image classification is effectively improved. It overcomes the shortcomings of current image classification methods, and has higher application value.
Image classification is a key technology to improve the speed of large-scale image retrieval. The current image classification methods have deficiencies, so they cannot meet the actual application needs of image classification. In order to solve problems existing in current image classification methods, and further improve the effect of image classification, a novel image classification method based on depth learning algorithm is proposed in this paper. Firstly, the research status of image classification at home and abroad is introduced, and the limitations of various image classification algorithms are pointed out. Secondly the features of image classification are extracted while image feature vector machine is used as the input, and image classification is used as the output of depth learning algorithm, and the image classifier is constructed by learning. At last, the performance is tested and compared with classical image classification methods in Matlab 2016 platform. The simulation results show that the image classification accuracy of the proposed method is obviously better, the average time of image classification is greatly reduced, the overall result of image classification is effectively improved. It overcomes the shortcomings of current image classification methods, and has higher application value.
引文
[1]张鸿,伍萍.基于线性解码和深度回归预测的图像分类算法[J].计算机应用与软件,2016,33(11):130-134.
[2]李康顺,王福滨,张丽霞,等.基于改进BOF算法的图像识别和分类[J].中南大学学报(自然科学版),2016,47(5):1599-1605.
[3]王玉晶.基于最优特征加权的图像分类算法[J].中北大学学报(自然科学版),2017,38(2):196-201.
[4]胡明娣,臧艺迪,徐家慧.基于FABEMD的纹理图像分类算法[J].西安邮电大学学报,2018,23(1):53-58.
[5]刘盛清,孙季丰,余家林,等.基于多尺度特征融合Hessian稀疏编码的图像分类算法[J].计算机应用,2017,37(12):3517-3522.
[6]朱威,屈景怡,吴仁彪.结合批归一化的直通卷积神经网络图像分类算法[J].计算机辅助设计与图形学学报,2017,29(9):1650-1657.
[7]范敏,韩琪,王芬,宿晓岚,等.基于多层次特征表示的场景图像分类算法[J].吉林大学学报(工学版),2017,47(6):1909-1917.
[8]张光辉,徐学红.基于多特征融合与稀疏分类的图像检索算法[J].控制工程,2016,23(11):1796-1801.
[9]陈瑞瑞.基于深度卷积神经网络的图像分类算法[J].河南科技学院学报(自然科学版),2018,46(4):56-60,67.
[10]史静,朱虹.多特征融合的场景图像分类算法[J].计算机系统应用,2018,27(5):171-175.
[11]古险峰,冯学晓.基于深度无监督学习的图像分类算法[J].平顶山学院学报,2018,33(2):67-70.
[12]王改华,李涛,吕朦,等.采用无监督学习算法与卷积的图像分类模型[J].华侨大学学报(自然科学版),2018,39(1):146-151.
[13]吴志攀,郑中韦.基于深度学习与运动信息的动作识别算法[J].计算机工程与设计,2018,39(8):2668-2674.
[2]李康顺,王福滨,张丽霞,等.基于改进BOF算法的图像识别和分类[J].中南大学学报(自然科学版),2016,47(5):1599-1605.
[3]王玉晶.基于最优特征加权的图像分类算法[J].中北大学学报(自然科学版),2017,38(2):196-201.
[4]胡明娣,臧艺迪,徐家慧.基于FABEMD的纹理图像分类算法[J].西安邮电大学学报,2018,23(1):53-58.
[5]刘盛清,孙季丰,余家林,等.基于多尺度特征融合Hessian稀疏编码的图像分类算法[J].计算机应用,2017,37(12):3517-3522.
[6]朱威,屈景怡,吴仁彪.结合批归一化的直通卷积神经网络图像分类算法[J].计算机辅助设计与图形学学报,2017,29(9):1650-1657.
[7]范敏,韩琪,王芬,宿晓岚,等.基于多层次特征表示的场景图像分类算法[J].吉林大学学报(工学版),2017,47(6):1909-1917.
[8]张光辉,徐学红.基于多特征融合与稀疏分类的图像检索算法[J].控制工程,2016,23(11):1796-1801.
[9]陈瑞瑞.基于深度卷积神经网络的图像分类算法[J].河南科技学院学报(自然科学版),2018,46(4):56-60,67.
[10]史静,朱虹.多特征融合的场景图像分类算法[J].计算机系统应用,2018,27(5):171-175.
[11]古险峰,冯学晓.基于深度无监督学习的图像分类算法[J].平顶山学院学报,2018,33(2):67-70.
[12]王改华,李涛,吕朦,等.采用无监督学习算法与卷积的图像分类模型[J].华侨大学学报(自然科学版),2018,39(1):146-151.
[13]吴志攀,郑中韦.基于深度学习与运动信息的动作识别算法[J].计算机工程与设计,2018,39(8):2668-2674.