Incremental Learning Method for Remote Sensing Image Segmentation

    As a powerful visual model, Incremental Learning Method (ILM) have demonstrated remarkable performance in various visual recognition problems and attracted considerable attention in recent years. However, due to the highly correlated bands and insufficient training samples of hyper spectral image data, it remains a challenging problem to effectively apply the ILN models on hyper-spectral images. In this Matlab design, an efficient ILM and SVM architecture has been designed using the below mentioned feature extraction methods to boost its discriminative capability for hyper spectral image classification, in which the original data is used as the input and the final ILN, SVM outputs are the predicted class-related results.
Feature Extraction Methods

  • Full trained AlexNet
  • Fine-tuned AlexNet
  • Pre-trained AlexNet
  • Full trained CaffeNet
  • Fine-tuned CaffeNet
  • Pre-trained CaffeNet
  • Full trained GoogLeNet
  • Fine-tuned GoogLeNet
  • Pre-trained GoogLeNet
  • Full trained VGGConvNets
  • Fine-tuned VGGConvNets
  • Pre-trained VGGConvNets

The proposed ILM is compared with SVM for the various feature extraction methods. The experiments on Zurich Dataset V1.0 have demonstrated using Matlab 2018a version and results shows remote sensing image segmentation based on the segment classes: Roads, Buildings, Trees, Grass, Bare Soil, Water, Railways and Swimming pools

Reference Paper-1: Convolutional Neural Networks for Hyper Spectral Image Classification

Author’s Name: Shiqi Yu, Sen Jia, and ChunyanXu

Source: Elsevier-Neurocomputing


Reference Paper-2: Towards better Exploiting Convolutional Neural Networks for Remote Sensing

Author’s Name: Keiller Nogueira, Otavio A.B.Penatti, and  Jefersson A.dos Santos

Source: Elsevier-Pattern Recognition


Request design files for academic purpose, contact +91 7904568456 by whatsapp or

REQUEST SIMULATION VIDEO DEMO