Hyperspectral image classification using ResNet with ...

1 Hyperspectral image classification using ResNet with squeeze and excitation block ECE 539 Fall 2018 Jiayu Wang Question to solve I am currently working as a part-time software engineer intern at a company which delivers Hyperspectral imaging system. I want to explore image classification using Hyperspectral images. Dataset The company allows me to use some of the existing Hyperspectral images for this project. There is a total of 588 images from five classes. The specific composition is listed in the table below. image class Training size Testing size Total Almond 189 21 210 Amber 108 12 120 Shell 97 11 108 Stone 113 13 126 Wood Chip 21 3 24 Total 528 60 588 image size: 100 x 100 pixel The images are not labelled.

2 I need to do that. Plan I m going to use deep convolutional neural network [1]. The architecture is ResNet [2] with squeeze & excitation [3] block. I will use TensorFlow with keras and Python3 to implement it. There are implementations of this architecture online. But since my input is of different size, I will reference those and either build my own version if applicable or adjust my input dimensions. Next step: 1. Build the architecture (ongoing) 2. Preprocess data. Read in RAW image files as numpy array with selected channels. 3. Make labels. Use 1 to 5 as categories. 4. Separate into training and testing sets. Reference [1] A. Krizhevsky, I.

3 Sutskever, and G. E. Hinton, ImageNet classification with Deep Convolutional Neural Networks, Commun. ACM, vol. 60, no. 6, pp. 84 90, May 2017. [2] K. He, X. Zhang, S. Ren, and J. Sun, Deep Residual Learning for image Recognition, [cs], Dec. 2015. [3] J. Hu, L. Shen, and G. Sun, squeeze -and- excitation Networks, [cs], Sep. 2017.