کنفرانس مهندسی برق ایران

صفحه اصلی / سی و دومین کنفرانس بین المللی مهندسی برق

HyperSpectral Image Classification using a 3D Convolutional Mixer Block

نویسندگان :

Sara Dianat¹ Mehran Yazdi²

1- دانشگاه شیراز 2- دانشگاه شیراز

کلمات کلیدی :

HyperSpectral imaging،classification،deep learning،transformers

چکیده :

HyperSpectral Image (HSI) processing has many applications in agriculture, Earth Change Monitoring, etc. Classification of HSIs is an important stage in most applications and applying a precise method can improve the overall performance, transformers have shown outstanding performance in HSI classification. In this paper, we propose a novel classification algorithm using the convolutional mixer for HSI classification. Convolutional mixer is similar to transformers and is used to mix spatial and spectral information which are gathered separately by using depth wise convolution followed by a point wise convolution. We have used 3D convolutions in mixer block due to the 3D shape of hyperspectral cuboids to extract spectral and spatial features simultaneously, which have not been done before, to the best of our knowledge. Experimental results show that our method is superior to similar state of the art works in terms of overall accuracy and Kappa Coefficient.