基于深度学习的可扩展图像分类研究
首发时间:2020-12-28
摘要:鉴于图像分类算法中对未曾训练过的样本的不兼容性,本文提出了一种基于深度学习的可扩展图像分类研究,可以对未曾训练过的样本进行识别并加以分类。首先对从二分类开始测试,通过正负样本的选取,对数据集进行预处理,对比选取合适的网络结构并编写相应基于BP的训练方法,然后开始添加分类到十分类,并对之前选取的网络结构和编写好的训练方法稍加修改,对每一个分类的种类单独训练并测试结果。通过实验结果证明,测试中添加新的未曾训练过的图像数据,在本研究神经网络中也可以识别出来并得到相对很好的实验结果。
关键词: 计算神经网络 图像分类 可扩展性 训练方法 BP算法
For information in English, please click here
Research on Expansibility Image Classification Based on deep learning
Abstract:In view of the incompatibility of the untrained samples in image classification algorithm, this paper proposes an expansible image classification research based on deep learning, which can identify and classify the untrained samples. First of all, the test starts from the two classification, through the selection of positive and negative samples, preprocesses the data set, compares and selects the appropriate network structure and compiles the corresponding training method, then starts to add the classification to the ten categories, and slightly modifies the selected network structure and the prepared training method, and each classification type is trained and tested separately. The experimental results show that adding new untrained image data in the test can also be recognized in the neural network of this study and relatively good experimental results can be obtained.
Keywords: computational neural network image classification expansible training method BP algorithm
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
基于深度学习的可扩展图像分类研究
评论
全部评论0/1000