Low-Complexity Reconfigurable Computing Based Online One-Class Classification Using High-Resolution Hyperspectral Imagery

Material characterization and recognition can be accomplished with the help of hyperspectral remote sensing techniques. In recent years, the practical implementation of online classification using hyperspectral imagery has been gaining attention, especially in the remote sensing community, due to its ability to enable a plethora of novel applications. In the coming years, there will be considerable demand for still image inspection and moving image monitoring applications, particularly in the aerospace and industrial automation fields. They require cost-effective, low-power usage and portable image analysis systems with high-performance computing framework abilities. Field-programmable gate arrays (FPGAs) are a suitable solution that can accommodate the above-listed requirements. But, designing and developing the FPGA-based hardware design is known to be a complex process. This paper presents a low-complexity FPGA design approach that performs the one-class support vector machine (OCSVM) classification using ultra-high resolution hyperspectral imagery.