基于SE-Inception-v3的星系形态分类模型

随着天文探测技术的快速发展, 海量的星系图像数据不断产生, 能够及时高效地对星系图像进行形态分类对研究星系的形成与演化至关重要. 针对传统的星系形态分类模型特征选择困难、分类速度慢、准确率受限等难题, 提出一种以Inception-v3神经网络为主干结构, 融合压缩激励(Squeeze and Excitation Network, SE)通道注意力机制的星系形态分类模型. 该模型在斯隆数字巡天(Sloan Digital Sky Survey, SDSS)样本的测试集准确率高达99.37%. 旋涡星系、圆形星系、中间星系、雪茄状星系与侧向星系的F1值分别为99.33%、99.58%、99.33%、99.41%与99.16%. 该模型与Inception-v3、MobileNet (Mobile Neural Network)和ResNet (Residual Neural Network)网络模型相比, SE-Inception-v3宽度和深度优势表现出更强的特征提取能力, 可以高效识别不同形态的星系, 为未来大型巡天计划的大规模星系形态分类问题提供了一种新方法.

Galaxy Morphology Classification Model Based on SE-Inception-v3

With the rapid development of astronomical detection technology, there will be a huge torrent of incoming galaxy images in the coming years, making the automatic galaxy morphology classification a challenging task. To solve the problem of feature selection, the low speed and low accuracy of traditional galaxy morphology classification models, a galaxy morphology classification model based on Inception-v3 neural network with SE (Squeeze and Excitation Network) channel attention mechanism is introduced. We select galaxy images from Sloan Digital Sky Survey (SDSS) for the SE-Inception-v3 model. The test results show that the accuracy of SE-Inception-v3 model is as high as 99.37%, and the F1 scores of spiral galaxy, completely round smooth galaxy, in-between smooth galaxy, cigar-shaped smooth galaxy and edge-on galaxy are 99.33%, 99.58%, 99.33%, 99.41% and 99.16%, respectively. Compared with the MobileNet (Mobile Neural Network) and ResNet (Residual Neural Network) models, the width and depth advantages of SE-Inception-v3 make the classification model have stronger feature extraction capabilities, which provides a new galaxy morphology classification approach for future large-scale sky survey programs.