基于高斯过程分类的小样本图像识别  

Image Recognition of Small Sample Based on Gaussian Process Classification

作　　者：林宽 李卓雅 李新民 

机构地区：[1]青岛大学数学与统计学院,山东 青岛 [2]青岛海尔智能家电科技有限公司,山东 青岛

出　　处：《人工智能与机器人研究》2022年第4期341-347,共7页Artificial Intelligence and Robotics Research

摘　　要：当今图片分类课题应用广泛,各种神经网络模型层出不穷。计算机性能和图片样本量对模型的训练和预测结果均有影响。针对于图片分类可能遇到的小样本问题,本文提出使用高斯过程分类模型和HSV图片分解的组合方式,与传统机器学习模型和RGB分解方式进行比较,发现该组合方式的准确率最高,效果最好。Nowadays, image classification is widely used, and various neural network models emerge in endlessly. The performance of computer and the sample size of pictures have an influence on the training and prediction results of the model. Aiming at the problem of small samples that may be encountered in image classification, this paper proposes a combination method of the Gaussian process classification model and HSV image decomposition. Compared with the traditional machine learning model and RGB decomposition method, it is found that this combination method has the highest accuracy and the best effect.

关 键 词：图片分类 小样本 高斯过程分类 HSV 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]