Classification and Area Computation Modeling of Remote Sensing Images Using Histogram and Convolutional Neural Network

EasyChair Preprint 6429

Abstract

Remote Sensing is an important field in science and technology and consists of the images of Earth and satellites taken by the means of artificial satellites or aircraft. Satellite images or high-resolution aerial images, is flexible to work with and easy to monitor. Since, the total area of the earth is so large, high resolution remote sensing images produce vast amount of data, even image processing is time consuming. This work represents a combination of unsupervised and supervised process to classify high spatial resolution satellite images so that minimal human intervention is needed. For this purpose, histogram peak-based classification approach is used to classify remote sensing image into subcategories like urban land, vegetation land, water body etc. To detect different objects, present in the image, convolutional neural network-based approach is used. The neural network model is trained using custom dataset. Then object localization operation is performed to get the co-ordinates of the object present in the image. Then histogram-based segmentation operation is performed to compute the area of different objects present in the image. After that 3d model is constructed using the co-ordinates obtained. Georeferencing technique is used to calculate the area of different classes observed.

Keyphrases: 3D modeling, Convolutional Neural Net-work, image classification, remote sensing

@booklet{EasyChair:6429,
  author    = {Swarna Kamal Pradhan and Dipon Das and Ujjwal Mondal},
  title     = {Classification and Area Computation Modeling of Remote Sensing Images Using Histogram and Convolutional Neural Network},
  howpublished = {EasyChair Preprint 6429},
  year      = {EasyChair, 2021}}