Hyperspectral images clustering on reconfigurable hardware using the k-means algorithm

摘要: Unsupervised clustering is a powerful technique for understanding multispectral and hyperspectral images, k-means being one of the most used iterative approaches. It simple though computationally expensive algorithm, particularly large images into many categories. Software implementation presents advantages such as flexibility low cost complex functions. However, it limitations, difficulties in exploiting parallelism high performance applications. In order to accelerate clustering, hardware could be used. The disadvantage this approach that any change project requires previous knowledge design process can take several weeks implemented. improve methodology, an automatic parameterized has been developed hardware/software codesign approach. An unsupervised uses Euclidian distance calculate pixel centers was case study validate methodology. Two implementations, software one, have Although component operates at 40 MHz, 12.5 times less than operating frequency (PC), approximately 2 faster one.