Architecture-Aware Optimization Strategies in Real-time Image Processing

6Embedded Implementation of VHR Satellite Image Segmentation

Remote sensing techniques are increasingly used in geological exploration, natural disaster prevention, monitoring, etc. They usually require very high resolution (VHR) satellite images, which are texturally rich and may raise the running cost of systems. In some cases, for example during volcanic eruptions or flood monitoring, fast and effective processing methods are necessary because the information needs to be extracted and considered as fast as possible. However, image segmentation models usually used to detect objects or other relevant features in VHR satellite images are fundamentally time consuming, which is a real hassle for researchers.

In this chapter, we present a texture region segmentation method for VHR satellite images in a high-abstraction C environment and realize its RTL FPGA implementation with the CDMS4HLS design flow described in the previous chapter. First of all, we base the design on a promising level set method (LSM) segmentation dedicated to VHR satellite images [AKB 12, HIC 13, HUA 11]. Next, the lattice Boltzmann method (LBM) is used as the solver of the level set equation for its highly parallelizable capability. Finally, the algorithm is implemented into the register-transfer level for FPGA using the improved HLS tools.

6.1. LSM description

6.1.1. Background

The LSM refers to the class of active contour models that use the implicit representation of the evolving curve instead of the ...